Edited, memorised or added to reading list

on 10-Feb-2020 (Mon)

Do you want BuboFlash to help you learning these things? Click here to log in or create user.

#ml #snorkel

Aim of Babble Labble: Can we use natural language as a form of weak supervision?

Natural language supervision would involve parsing the semantics and then using these as labeling functions.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Ratner_et_al-2017-dawn,cs,stanford,edu-Snorkel_and_The_Dawn
e extracting structured data from data that has some structure such as tables embedded in PDFs and webpages? We’ve been working on a system called Fonduer to make this fast and easy in Snorkel! <span>Learning from natural language supervision : Can we use natural language as a form of weak supervision, parsing the semantics of natural language statements and then using these as labeling functions? We’ve done some exciting preliminary work here! <span>




#knowledge-base-construction #machine-learning

Non-functional Requirements of an AKBC system are performance, scaling, usability and support for transparency and fairness.

An AKBC system should be performant when training a model or applying inferences.

An AKBC system must be able to scale in order to process a large corpus of potentially billions of documents, containing, in turn, billions of figures and tables.

The ability of an AKBC system to scale is increasingly relevant as larger and larger data sets become available.

An AKBC system must not require end users to learn technical details of underlying algorithms.

An AKBC sys- tem should not require writing complex extraction functions (in the form of programs or scripts).

An AKBC system should provide the capability to choose between differ- ent features (and even models), as this can allow end users to decide if any features or models do not meet desired properties (such as fairness)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning
For feature extraction in AKBCs, low-level techniques and big-data frameworks, e.g.Hadoop and Condor., have been used.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Flashcard 4846354828556

Tags
#knowledge-base-construction #machine-learning #unfinished
Question
A mention is a span of text that refers to an [...].
Answer
entity


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
A mention is a span of text that refers to an entity.

Original toplevel document (pdf)

cannot see any pdfs







#knowledge-base-construction #machine-learning #unfinished

Fonduer: standard NLP pre-processing tools are used to generate linguistic attributes, such as lemmas, parts of speech tags, named entity recognition tags, dependency paths, etc., for each Sentence.

Fonduer: Structural and tabular attributes of a Sentence, such as tags, and row/column information, and parent attributes, can be captured by traversing its path in the data model.

Fonduer: Visual attributes for the document are recorded by storing bounding box and page information for each word in a Sentence.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#nlp #reading-group #transformer #unfinished
Transformers: As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Alammar-2018-The_Illustrated_Transformer-jalammar,github,io
eet or to the animal? It’s a simple question to a human, but not as simple to an algorithm. When the model is processing the word “it”, self-attention allows it to associate “it” with “animal”. <span>As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word. If you’re familiar with RNNs, think of how maintaining a hidden state allows an RNN to incorporate its representation of previous words/vectors it has processed with the current one it’




#bert #nlp #unfinished
There are two steps in the BERT framework: pre-training and fine-tuning.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#machine-learning #management #software-engineering #unfinished

By coupling research and de- velopment, Google's goal is to minimize or even eliminate the traditional technol- ogy transfer process.

The traditional technology transfer process has proven challenging at other companies.

Most of Google's projects involve people work- ing with a given technology from the research stage through to the product stage.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#machine-learning #software-engineering #unfinished

If a problem does not have a clear specfication, it is not amenable#to#provable#algorithms.

Mac hin e#learnin g#replaces a missing specification by lots of data. As aspects#of#the#da ta#cannot#be#descri bed#by#a#compact# mathematical# statemen, experimentation#is# necessary!

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Behind the Agile manifesto: What the authors came to value after developing software by doing it and helping others do it.

Manifesto for Agile Software Development: Individuals and interactions over processes and tools
Manifesto for Agile Software Development: Working software over comprehensive documentation
Manifesto for Agile Software Development: Customer collaboration over contract negotiation
Manifesto for Agile Software Development: Responding to change over following a plan

Manifesto for Agile Software Development: X "over" Y means that while there is value in X, Agile authors value Y more.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Unknown title
Manifesto for Agile Software Development Manifesto for Agile Software Development We are uncovering better ways of developing software by doing it and helping others do it. Through this work we have come to value: Individuals and interactions over processes and tools Working software over comprehensive documentation Customer collaboration over contract negotiation Responding to change over following a plan That is, while there is value in the items on the right, we value the items on the left more. Kent Beck Mike Beedle Arie van Bennekum Alistair Cockburn Ward Cunningham Martin Fowler James Grenning Jim Highsmith Andrew Hunt Ron Jeffries Jon Kern Brian Marick Robert C. Martin Stev




#bert

The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens.

  • The first token of every input sequence in BERT is the special classification token – [CLS].
  • BERT: The [CLS] token is used in classification tasks as an aggregate of the entire sequence representation.
  • BERT: The [CLS] token is ignored in non-classification tasks.
  • BERT: For single text sentence tasks, the [CLS] token is followed by the WordPiece tokens and the separator token – [SEP].
  • BERT: For sentence pair tasks, the WordPiece tokens of the two sentences are separated by a [SEP] token. This i
  • For both single and pair sentence inputs in BERT, the sequence ends with the [SEP] token.

Sentence Pair Input

  • BERT: For sentence pair input, a token's sentence embedding indicates Sentence A or Sentence B.
  • BERT: Sentence embeddings are similar to token/word embeddings with a vocabulary of 2.
  • BERT: positional embeddings indicate each token's position in the sequence.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

BERT Explained – A list of Frequently Asked Questions – Let the Machines Learn
ctives allow it to be used on any single sequence and sequence-pair tasks without substantial task-specific architecture modifications. How is the input text represented before feeding to BERT? <span>The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence is the special classification token – [CLS]. This token is used in classification tasks as an aggregate of the entire sequence representation. It is ignored in non-classification tasks. For single text sentence tasks, this [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. Single Sentence Input For sentence pair tasks, the WordPiece tokens of the two sentences are separated by another [SEP] token. This input sequence also ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A positional embedding is also added to each token to indicate its position in the sequence. Which Tokenization strategy is used by BERT? BERT uses WordPiece tokenization. The vocabulary is initialized with all the individual characters in the language, and then the most freque




Flashcard 4967545572620

Tags
#machine-learning #software-engineering #unfinished
Question
The experimental paradigm in machine learning is reaching [...].
Answer
its limits


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
The experimental paradigm in machine learning is reaching its limits. This is challenging the speed of scientific progress in the area.

Original toplevel document (pdf)

cannot see any pdfs







#knowledge-base-construction #machine-learning #unfinished
Fonduer: structural and visual features are generated automatically (see Section 4.2).

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
networks [ 18 ]) struggle to capture the multimodality of richly formatted data. We introduce a multimodal LSTM network that combines textual context with universal features that correspond to <span>structural and visual properties of the input documents. These features are inherently captured by Fonduer ’s data model and are generated automatically (see Section 4.2). We also introduce a series of data layout optimizations to ensure the scalability of Fonduer to millions of document-wide candidates (see Appendix C). <span>

Original toplevel document (pdf)

cannot see any pdfs




  • Wireframes help to: Connect the site’s information architecture to its visual design by showing paths between pages
  • Wireframes help to: Clarify consistent ways for displaying particular types of information on the user interface
  • Wireframes help to: Determine intended functionality in the interface
  • Wireframes help to: Prioritize content through the determination of how much space to allocate to a given item and where that item is located

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Wireframing | Usability.gov
d intended behaviors. For these reasons, wireframes typically do not include any styling, color, or graphics. Wireframes also help establish relationships between a website’s various templates. <span>The Value of Wireframes Wireframes serve multiple purposes by helping to: Connect the site’s information architecture to its visual design by showing paths between pages Clarify consistent ways for displaying particular types of information on the user interface Determine intended functionality in the interface Prioritize content through the determination of how much space to allocate to a given item and where that item is located Creating Wireframes It’s important to keep in mind that wireframes are guides to where the major navigation and content elements of your site are going to appear on the page. Since the




Flashcard 4996083879180

Tags
#knowledge-base-construction #machine-learning
Question
In contrast to KBC from text or tabular data, Fonduer aims to extract relations conveyed via tabular, structural, [...] , and visual expressions
Answer
textual


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
In contrast to KBC from text or tabular data, KBC from richly formatted data aims to extract relations conveyed jointly via textual, structural, tabular, and visual expressions.

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 4996086762764

Tags
#knowledge-base-construction #machine-learning
Question
In contrast to KBC from text or tabular data, Fonduer aims to extract relations conveyed via tabular, textual, [...] , and visual expressions
Answer
structural


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
In contrast to KBC from text or tabular data, KBC from richly formatted data aims to extract relations conveyed jointly via textual, structural, tabular, and visual expressions.

Original toplevel document (pdf)

cannot see any pdfs







#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Pièges diagnostiques :

  • Douleurs du genou secondaire à une atteinte de la hanche (douleurs projetées)
  • Un épisode traumatique antérieur (chute et/ou coup), banal et quotidien chez les enfants, est souvent évoqué par les parents comme cause de l'impotence fonctionnelle : le médecin doit rester vigilant dans sa démarche diagnostique.
  • Un psoïtis est susceptible d'entrainer des difficultés à la marche (ex. appendicite aigüe)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

L'observation de la boiterie est essentielle :

  • Dans la boiterie d'épaule ou d'équilibration, qui traduit une pathologie de la hanche, l'enfant transfert son centre de gravité à chaque appui du côté de la hanche pathologique
  • Dans la boiterie d'esquive, qui est antalgique, l'enfant évite ou raccourcit le temps d'appui du côté pathologique et déporte le poids de son corps du côté opposé.

On recherche ensuite une attitude vicieuse du membre inférieur, une lésion cutanée, en particulier au niveau de la plante du pied.

Une amyotrophie quadricipitale est un signe formel d'organicité et d'ancienneté

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

La palpation de tout le membre inférieur doit être méthodique, en commençant par son extrémité distale et en remontant jusqu'à la hanche.

La diaphyse tibiale est palpée avec une attention particulière (possible fracture sous-périostée).
Dans le même objectif, on lui imprime un mouvement de torsion

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

  • Dans les maladies de hanche, l'abduction et la rotation interne sont limitées.

  • On recherche enfin une douleur à la palpation ou une raideur au niveau du rachis.
    Le refus de la station assise est parfois le seul signe objectif d'atteinte du rachis.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
Tout examen pour douleur du genou, même isolée, comprend obligatoirement l'examen de la hanche

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Les radiographies (face + profil) sont centrées sur la zone douloureuse éventuellement identifiée.

Les clichés comparatifs sont rarement justifiés.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

L'échographie (comparative) peut objectiver un épanchement articulaire de la hanche ou de la cheville.

Dans le cadre d'une ostéomyélite, elle peut retrouver un épaississement des parties molles ou un abcès sous-périosté

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
La scintigraphie osseuse est utile chez le jeune enfant difficile à examiner, afin de localiser le site de la lésion osseuse ou articulaire responsable de l'impotence fonctionnelle (recherche d'une hyperfixation pathologique). Dans un second temps, elle pourra guider une imagerie ciblée sur la lésion (TDM ou IRM)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

En cas de suspicion d'infection ostéoarticulaire les prélèvements bactériologiques sont à effectuer en urgence, avant de débuter l'antibiothérapie :

  • Hémocultures (en faire deux rapprochées)
  • Prélèvements au bloc opératoire sous anesthésie générale de liquide articulaire en cas d'arthrite ou du pus d'un abcès sous-périosté.
    Dans les ostéomyélites, la ponction osseuse diagnostique n'est pas systématique ; elle est effectuée par certaines équipes

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

La fièvre peut être modérée, en particulier chez le nourrisson ou dans les localisations rachidiennes (spondylodiscites).

La CRP est le plus souvent > 20 mg/dL ; la polynucléose peut être modérée ou absente

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
Le retard diagnostique et/ou thérapeutique de ces infections peut être sanctionné à court terme par un sepsis (Staphylococcus aureus, streptocoque de groupe A…) et, à moyen terme, par des séquelles orthopédiques lourdes : destruction du cartilage articulaire responsable de coxarthrose précoce, atteinte du cartilage de croissance responsable de sa stérilisation puis de déformation ou d'inégalité de longueur des membres inférieurs

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
Une IOA se développe par voie hématogène. Elle atteint les membres inférieurs dans plus de 50 % des cas, en particulier les zones fortement vascularisées (métaphyses osseuses, synoviale)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

L'IOA survient le plus souvent chez un enfant sain.

Les principaux germes impliqués dans les infections ostéoarticulaires de l'enfant sont :

  • À tout âge : Staphylococcus aureus
  • Chez le jeune nourrisson (âge < 3 mois) :
    • Streptocoque de groupe B et Escherichia coli (penser à l'infection néonatale)
  • Chez l'enfant âgé de moins de 4 ans :
    • Kingella kingae (avant tout)
    • Streptocoque de groupe A
    • Rarement, pneumocoque.
  • Chez l'enfant drépanocytaire :
    • Salmonella sp.

Les portes d'entrée à rechercher sont cutanées (traumatisme, plaie) et ORL

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

S. aureus est le germe en cause dans 50 % des infections ostéoarticulaires de l'enfant.

Kingella kingae est le pathogène le plus fréquent chez les enfants âgés de moins de 4 ans

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Ostéomyélite aiguë

Tableau clinique habituel :

  • Début brutal avec fièvre > 39 °C ou présentations subaiguës (fièvre peu élevée)
  • Localisation préférentielle à l'extrémité inférieure du fémur ou supérieure du tibia (métaphyses osseuses les plus vascularisées : « près du genou, loin du coude »)
  • Boiterie douloureuse ou appui impossible
  • Augmentation de la température locale (rougeur et tuméfaction à un stade tardif)
  • Palpation métaphysaire très douloureuse

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

L'infection osseuse est suspectée sur ces premiers éléments mais le diagnostic devra être confirmé secondairement par :

  • IRM : montre une zone en hyposignal T1 et hypersignal T2 ; si elle est disponible, c'est le meilleur examen pour le diagnostic d'IOA, elle est plus sensible et plus spécifique que la scintigraphie osseuse

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Arthrite septique

Tableau clinique habituel :

  • Survenue brutale d'un syndrome douloureux fébrile avec impotence fonctionnelle partielle ou totale
  • Localisations les plus fréquentes : hanche ou genou
  • Mobilisation de l'articulation extrêmement douloureuse
  • Épanchement articulaire (par exemple, choc rotulien au niveau du genou).

  • Examens d'imagerie pouvant permettre de préciser le diagnostic :
    • Échographie : particulièrement précieuse pour objectiver un épanchement intra-articulaire de la hanche ou de la cheville

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
Un épanchement articulaire en contexte fébrile est une arthrite septique jusqu'à preuve du contraire

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Le bilan inflammatoire (NFS, CRP) est souvent perturbé mais il peut être normal.

Les prélèvements bactériologiques (hémocultures, prélèvements bactériologiques profonds), faits avant antibiothérapie, permettent seuls la confirmation étiologique

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
La biologie moléculaire (PCR dans le liquide de ponction articulaire) améliore le diagnostic étiologique des arthrites septiques à K. kingae

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
En cas de suspicion d'IOA, l'enfant doit être hospitalisé, avec avis auprès d'un chirurgien pédiatrique. Si l'indication d'un drainage chirurgical est retenue, l'enfant est maintenu à jeun et transféré en orthopédie pédiatrique en urgence

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

L'antibiothérapie initiale est urgente, intraveineuse et probabiliste :

  • Efficace sur S. aureus (sensible à la méticilline) et K. kingae (germes les plus fréquents)
    • Amoxicilline + acide clavulanique, céfamandole ou céfazoline

  • Elle est secondairement adaptée au germe retrouvé et à son antibiogramme.
  • Sous traitement adapté, dans les cas d'évolution favorable (disparition de la fièvre et des douleurs), le relais antibiotique per os est possible 3 à 5 jours après le début du traitement IV.
    • L'antibiothérapie est secondairement adaptée en cas d'identification du germe.
      • Si le germe n'est pas identifié, l'antibiothérapie orale sera l'association amoxicilline + acide clavulanique, efficace sur S. aureus et K. kingae

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
La durée totale de l'antibiothérapie des IOA d'évolution rapidement favorable est de 2 semaines pour les arthrites septiques et de 3 semaines pour les ostéomyélite

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie avant 3 ans :

Fracture sous-périostée du tibia (fracture « en cheveu d'ange »)

  • C'est la première cause de boiterie à cet âge.
  • Cette lésion est très particulière pour plusieurs raisons :
    • Elle est liée à un traumatisme bénin par torsion du membre, souvent passé inaperçu : simple chute de sa hauteur ou pied pris entre les barreaux du lit
    • Le périoste est, chez le jeune enfant, un fourreau très solide qui ne se déchire pas facilement.
      Il assure donc la tenue mécanique de l'os, malgré la fracture.
      C'est ainsi la seule fracture de jambe qui soit compatible avec l'appui et la marche

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie avant 3 ans :

Fracture sous-périostée en cheveux d'ange :

  • Aucun traitement n'est nécessaire à l'exception, le cas échéant, d'une immobilisation plâtrée antalgique.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie entre 3 et 8 ans :

Synovite aiguë transitoire :

  • Aussi appelée « rhume de hanche ».
  • Elle survient habituellement avant 5 ans.

  • C'est une boiterie aiguë non fébrile observée le matin au réveil.
    La douleur peut être absente ou importante.

  • L'examen clinique trouve une limitation de la rotation interne et de l'abduction de la hanche.
    Le reste de l'examen est normal.

  • La radiographie de bassin de face et de la hanche douloureuse de profil est normale ; l'échographie peut mettre en évidence un épanchement intra-articulaire.
    Il n'y a pas de syndrome infectieux biologique.

  • Le retour à la normale est obtenu en quelques jours, sans traitement autre qu'une mise au repos de la hanche (réduction des sorties).
    Les récidives sont rares

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie entre 3 et 8 ans

Ostéonécrose fémorale supérieure (ostéochondrite primitive de hanche, ou maladie de Legg-Calvé-Perthes) :

  • Elle touche habituellement le garçon (80 % des cas) âgé de 4 à 8 ans.
    C'est la nécrose ischémique de l'épiphyse fémorale supérieure survenant au cours de la période de croissance.

  • La boiterie est d'installation insidieuse, augmentant à l'effort et en fin de journée.
    Elle a la particularité de n'être ni permanente ni toujours douloureuse, de sorte que le diagnostic peut être retardé de quelques semaines

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie entre 3 et 8 ans :

Ostéonécrose fémorale supérieure :

  • La radiographie de face du bassin et de la hanche de profil montre des signes variables en fonction du stade évolutif de la maladie et de son extension :
  • Au début, les signes radiologiques sont discrets :
    • Une diminution de la hauteur, un aplatissement du noyau épiphysaire ou une image en « coup d'ongle » sous-chondral
    • À un stade plus avancé, le diagnostic devient évident

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie entre 3 et 8 ans :

Ostéonécrose fémorale supérieure :

  • La guérison est constante et spontanée, mais elle survient avec des séquelles morphologiques plus ou moins graves.
  • C'est sur l'importance de ces séquelles que l'on peut agir par différentes méthodes thérapeutiques orthopédiques ou chirurgicales

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie chez le préadolescent et l'adolescent :

Épiphysiolyse fémorale supérieure (EFS)

  • C'est la principale cause de boiterie de hanche de la période pubertaire.
  • C'est une maladie du cartilage de croissance responsable du glissement de l'épiphyse fémorale supérieure sur la métaphyse.
    Dans plus de la moitié des cas, les patients présentent une surcharge pondérale.

  • Il s'agit d'une maladie non exceptionnelle trop souvent méconnue : le délai moyen entre les premiers signes et le diagnostic est encore actuellement de 3 mois.
    Pourtant, le diagnostic tardif expose au risque de séquelles.

  • La maladie devient bilatérale dans 20 % des cas.

  • Elle se manifeste sous deux formes cliniques complètement différentes : stable ou instable

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie chez le préadolescent et l'adolescent

EFS : forme stable

  • L'épiphyse reste solidaire du col du fémur.

  • L'appui est possible mais la boiterie est permanente, avec rotation externe du membre inférieur lors de la marche.
    La douleur est plutôt inguinale, mais elle peut être projetée et siéger au niveau du genou.

  • La rotation interne et l'abduction de la hanche sont toujours diminuées.
    La mise en rotation interne et en abduction réveille une douleur vive

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie chez le préadolescent et l'adolescent :

EFS forme stable :

  • C'est le cliché de profil qui met en évidence le signe direct du glissement : bascule postérieure de l'épiphyse fémorale supérieure.
  • À un stade plus évolué, la bascule épiphysaire est évidente sur les clichés de face : la classique ligne de Klein, tangente au bord supérieur du col fémoral, ne croise plus le pôle supérieur de l'épiphyse

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Boiterie chez le préadolescent et l'adolescent

EFS : forme instable

  • Les circonstances du diagnostic sont complètement différentes.
  • L'adolescent présente une impotence fonctionnelle complète et hyperalgique du membre inférieur évoquant une fracture. Mais le traumatisme causal est minime, voire inexistant.
  • Presque constamment, on retrouve un antécédent de boiterie douloureuse, négligée depuis plusieurs semaines.

  • La radiographie de bassin de face est suffisante pour porter le diagnostic ; il n'est le plus souvent pas possible ni souhaitable d'obtenir un cliché de profil chez un malade réveillé en raison de la douleur que provoque la mobilisation de la hanche.

  • Dans tous les cas, un retard thérapeutique expose à trois graves complications : la raideur et la coxite laminaire ou la nécrose fémorale supérieure

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Cinq critères cliniques pour diagnostiquer une épiphysiolyse fémorale supérieure :

  • Il s'agit d'un adolescent ou préadolescent
  • Qui se plaint de douleurs de la hanche ou du genou
  • Qui marche en rotation externe
  • Avec une boiterie de hanche
  • Et qui présente à l'examen une douleur et une raideur en abduction et rotation interne

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

  • Les tumeurs osseuses bénignes, rarement douloureuses, peuvent fragiliser l'os qui, en se fissurant, peut entraîner une boiterie.

  • Les tumeurs osseuses malignes de l'enfant sont dominées par les sarcomes osseux.
    Ils sont plus souvent localisés au niveau du genou (extrémité inférieure du fémur ou supérieure du tibia).
    Malheureusement, leur diagnostic est trop souvent retardé alors qu'une tendinite ou une douleur de croissance ont été d'abord évoquées.

  • Les localisations osseuses des hémopathies malignes peuvent être révélées par une boiterie rebelle

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
Les inégalités de longueur des membres inférieurs ne peuvent être causes de boiterie que si elles sont de grande amplitude (plus de 3 à 5 cm selon l'âge)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie
La hanche est identifiée à l'origine de moins de 40 % des boiteries de l'enfant. L'examen du rachis fait partie intégrante de l'examen clinique de tout enfant présentant un trouble de la démarche

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#153 #52 #Boiterie #Cours #Facultaires #IOA #Médecine #Pédiatrie

Une tumeur de la moelle spinale (primitive ou secondaire) peut se révéler par une compression médullaire ou une atteinte de la queue-de-cheval.

Le signe clinique à rechercher est la raideur du rachis

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
La traumatologie est la première cause de mortalité entre 1 an et 19 ans. Le sex-ratio est de 2 garçons pour 1 fille.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Les structures ligamentaires et capsulaires sont, chez l'enfant, plus résistantes que l'os ou les cartilages de croissance. Ainsi, le même traumatisme qui entraîne une entorse voire une luxation chez l'adulte provoque plus volontiers une fracture ou un décollement épiphysaire (traumatisme du cartilage de croissance) chez l'enfant

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

  • Un traumatisme axial provoquera une fracture en « motte de beurre »
  • Un mécanisme en torsion sera responsable d'une fracture spiroïde, d'abord sous-périostée puis complète

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




[unknown IMAGE 5000983350540]
#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes #has-images

Salter et Harris

Les risques de complications sont directement liés à l'atteinte ou au respect de la couche germinative du cartilage de croissance qui se trouve située sur le versant épiphysaire du cartilage :

  • Type I : le trait passe intégralement par la physe ; il passe dans la couche dégénérative de celle-ci et respecte de ce fait la couche germinative.

  • Type II : le trait passe par la physe avec un refend vers la métaphyse ; de ce fait, la couche germinative a peu de chance d'être touchée

  • Type III : le trait passe par la physe avec un refend vers l'épiphyse ; la couche germinative est obligatoirement traversée par le trait.

  • Type IV : le trait de fracture va de la métaphyse à l'épiphyse transversalement par rapport à la physe ; la couche germinative est obligatoirement traversée par le trait.

  • Type V : il complète cette classification
    • Il résulte d'un traumatisme axial qui lèse la couche germinative.
    • Il est exceptionnellement observé isolément ; en revanche, un traumatisme à haute énergie peut introduire une composante de type V dans les quatre autres types

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Épiphysiodèse

  • C'est la complication la plus grave des traumatismes du cartilage de croissance, qu'ils soient accidentels ou liés à un traitement inapproprié.
  • Elle entraîne un arrêt de croissance en créant un pont osseux définitif

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Tout défaut ne se corrige pas avec la croissance. La reconstruction anatomique des traumatismes du cartilage de croissance et des fractures épiphysaires est obligatoire

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Périoste

  • Il corrige de façon parfois spectaculaire un cal vicieux angulaire, d'autant plus qu'il se situe proche d'une physe fertile et chez le jeune enfant

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Périoste

  • En revanche, il ne faut en aucun cas compter sur son aptitude à corriger des cals vicieux rotationnels ou dans des zones à faible potentiel de croissance

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
En prévision d'une éventuelle anesthésie générale, on précise l'heure du dernier repas et l'enfant doit être maintenu à jeun

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Toute suspicion de fracture impose une immobilisation transitoire par attelle, première étape du traitement antalgique

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Le bilan radiographique doit porter sur tous les segments de membre suspects et englobe systématiquement les articulations sus- et sous-jacentes.

Il s'efforce de donner des incidences orthogonales.

Si l'enfant est trop douloureux et qu'il présente à l'évidence une fracture déplacée, la radiographie pourra se limiter à une incidence simple confirmant le diagnostic qui sera affiné sous anesthésie par les incidences complémentaires

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Le cartilage de croissance est radiotransparent. Un décollement Salter I non déplacé ne sera identifiable que sur la localisation de la douleur et sur l'épaississement des parties molles (œdème)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Le traitement orthopédique (immobilisation par attelle ou appareillage circulaire en plâtre ou en résine) est privilégié chez le jeune enfant, avant 6 ans, pour toutes les fractures diaphysaires ou métaphysaires

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
En revanche, le risque de raideur est important dans les suites de fractures articulaires. C'est une des raisons pour lesquelles on préfère les solutions chirurgicales

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Les complications thromboemboliques sont exceptionnelles. Le traitement anticoagulant n'est discuté que chez l'enfant pubère

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
On préférera le recours très large à l'autorééducation (reprise d'activités physiques non violentes, idéalement la natation), qui permet de récupérer les amplitudes articulaires physiologiques sans retard

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
La rééducation est inutile voire dangereuse dans la très grande majorité des cas.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Le membre est immobilisé dans la position inverse de celle qui a provoqué la fracture (poignet fléchi, par exemple pour une fracture déplacée en extension).

Il faut toujours immobiliser les articulations sus- et sous-jacentes.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
La surveillance de la mobilité des doigts ou des orteils, du temps de recoloration pulpaire et l'évaluation de la douleur sont indispensables. Toute modification d'une seule des trois composantes doit faire évoquer le syndrome de loge et conduire à l'ablation de l'appareillage

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
La consolidation des fractures ou des décollements épiphysaires est plus rapide chez le petit enfant, en moyenne 45 jours. Les fractures non déplacées (« cheveu d'ange » ou « motte de beurre ») ne justifient une immobilisation qu'antalgique (0 à 10 jours)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
En revanche, après l'âge de 5 ans, les délais de consolidation des fractures diaphysaires sont identiques à ceux de l'adulte. Il est donc nécessaire d'immobiliser un segment de membre fracturé jusqu'à 90 jours

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Fractures de l'enfant :

Traitement chirurgical

  • Indications
    Il est discuté, voire obligatoire dans plusieurs situations :
    • Chez le polytraumatisé pour éviter les contentions qui compliquent le nursing de l'enfant
    • Pour certaines fractures pathologiques dont la consolidation risque d'être retardée (fragilité osseuse ou lésions kystiques)
    • Pour les fractures articulaires ou les décollements épiphysaires dont la réduction doit être anatomique
    • Pour les enfants d'âge scolaire qui pourront plus rapidement écrire (fracture des deux os de l'avant-bras) ou marcher (fracture du fémur) avec une ostéosynthèse interne

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Les deux principaux moyens sont l'embrochage centromédullaire élastique stable (ECMES) des fractures diaphysaires et l'ostéosynthèse par broches fines des fractures articulaires.

L'utilisation des moyens d'ostéosynthèse plus rigides utilisés chez l'adulte (clous) est possible chez l'adolescent dès lors que les cartilages de croissance arrivent à maturité.

La fixation externe est précieuse pour les fractures ouvertes ou avec perte de substance.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Les fractures du coude sont les plus fréquentes après celles du poignet.

  • Les cartilages de croissance de cette région sont peu fertiles, de sorte que les capacités de remodelage sont faibles.
  • Elles sont dominées par les fractures supracondyliennes (extra-articulaires), suivies par les fractures du condyle latéral (articulaires).
  • Plus que jamais, la rééducation de ces fractures est dangereuse et peut conduire à l'enraidissement du coude

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Fracture supracondylienne

  • Sa grande fréquence s'explique par la fragilité de la palette humérale et par l'hyperextension naturelle du coude des enfants.
  • Le déplacement est en grande majorité (95 %) postérieur

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Fracture supracondylienne

L'enfant est hyperalgique, le membre supérieur ballant, en extension

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Fracture supracondylienne

La classification est fondée sur le déplacement fracturaire, en quatre stades :

  • Stade 1 : rupture de la corticale antérieure.
  • Stade 2 : rupture bicorticale avec déplacement minime (bascule postérieure).
  • Stade 3 : déplacement important avec persistance d'un contact entre les fragments.
  • Stade 4 : absence de contact entre les fragments

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Fracture supracondylienne

À l'exception du stade 1 qui peut être immobilisé sans réduction, les autres stades nécessitent une réduction et une immobilisation sous anesthésie générale

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Fractures du condyle latéral

  • Il s'agit d'une fracture articulaire, décollement épiphysaire Salter 4.
  • La plus grosse difficulté est de diagnostiquer une forme non ou peu déplacée.
    C'est la douleur clinique qui doit faire rechercher le diagnostic.

  • La fixation chirurgicale par broches percutanées ou à ciel ouvert est toujours nécessaire pour éviter le risque de déplacement secondaire.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Toute fracture supracondylienne déplacée est une vraie urgence en raison de son risque majeur de complications vasculo-nerveuses

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
À la fin de la croissance, la soudure du cartilage de croissance de l'extrémité inférieure du tibia est centrifuge. Quand un décollement épiphysaire survient chez l'adolescent, il touche donc exclusivement les zones de fragilité persistantes qui sont périphériques (malléole médiale ou latérale)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

  • Décollement épiphysaire Salter II de l'extrémité inférieure du tibia : le refend métaphysaire est habituellement postérieur.
  • La réduction et l'immobilisation sont la règle.
  • Le risque de composante Salter V par écrasement partiel lors de l'accident impose une surveillance de la croissance à distance

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Décollements épiphysaires de l'extrémité inférieure du tibia Salter III et Salter IV :

  • Les plus fréquentes lésions portent sur la malléole médiale (fracture de MacFarland)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
La fracture de Tillaux est un décollement épiphysaire articulaire du tubercule antérolatéral du tibia ; l'arrachement du fragment est dû à la traction par le ligament tibio-fibulaire antérieur

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Les entorses vraies de cheville sont rares chez l'enfant

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Recherche de signes évocateurs d'une fracture de la base du crâne :

  • Hémotympan
  • Otorrhée
  • Rhinorrhée
  • Ecchymoses péri-orbitaires
  • Hématome rétro-auriculaire
  • Atteinte des paires crâniennes

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

TC de l'enfant :

  • Points forts :
    • Glasgow
    • PC
    • Embarrure
    • Réflexe pupillaire
    • Signes de localisation

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes
Le scanner cérébral sans injection est l'imagerie de choix pour apprécier l'importance des lésions intracrâniennes. Il est à réserver aux seuls TC considérés à « risque élevé » de complications

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Mesures thérapeutiques urgentes

Le maintien des fonctions vitales est une priorité en cas de TC grave (Glasgow ≤ 8) :

  • Décubitus dorsal à 30°, tête dans l'axe, monitoring cardiorespiratoire, pose d'une VVP et perfusion de sérum physiologique (jamais de glucosé)
  • Dans ce contexte, la surveillance régulière de la glycémie est un point fondamental, surtout chez le nourrisson de moins de 2 ans.
  • Libération des voies aériennes, oxygénation au masque, intubation précoce
  • Mannitol IV (en cas de signes d'HTIC)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

On distingue habituellement trois niveaux de risque :

  • Élevé : maintien des fonctions vitales, TDM cérébrale en urgence, surveillance prolongée
  • Intermédiaire : surveillance aux urgences avant décision de TDM cérébrale
  • Faible : surveillance à domicile avec conseils

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Risque élevé :

  • Glasgow pédiatrique ≤ 14
  • Signes neurologiques focaux
  • Convulsion persistante ou à distance de l'impact
  • Signes de fracture de la base du crâne, embarrure, plaie pénétrante
  • Persistance d'anomalies de l'interaction, du tonus
  • Fontanelle bombée, augmentation du PC

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#330 #360 #Cours #Facultaires #Fractures #Médecine #Pédiatrie #Traumatismes

Risque intermédiaire

  • Âge < 3 mois avec des stigmates de TC (hématome)
  • Perte de connaissance initiale
  • Convulsion brève à l'impact
  • Vomissements répétés
  • Anomalie de l'interaction ou du tonus
  • Hématome large du scalp
  • Céphalées intense

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Perspective is an API developed by Jigsaw and Google's Counter Abuse Technology team. It uses ML to spot abuse and harassment online, and scores comments based on the perceived impact they might have on a conversation

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

How Google's Jigsaw Is Trying to Detoxify the Internet -
bator, formerly known as Google Ideas, is beginning to prove that machine learning (ML) fashioned into tools for human moderators can change the way we approach the internet's toxicity problem. <span>Perspective is an API developed by Jigsaw and Google's Counter Abuse Technology team. It uses ML to spot abuse and harassment online, and scores comments based on the perceived impact they might have on a conversation in a bid to make human moderators' lives easier. Perspective Amidst the Shouting Matches The open-source tech was first announced in 2017, though development on it started a few years e




#knowledge-base-construction #machine-learning #unfinished
LSTMs

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
LSTMs: gates control the flow of information and can capture long-term dependencies

Original toplevel document (pdf)

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
gates

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
LSTMs: gates control the flow of information and can capture long-term dependencies

Original toplevel document (pdf)

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
control the flow of information

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
LSTMs: gates control the flow of information and can capture long-term dependencies

Original toplevel document (pdf)

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
capture long-term dependencies

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
LSTMs: gates control the flow of information and can capture long-term dependencies

Original toplevel document (pdf)

cannot see any pdfs




Behind the Agile manifesto: What the authors came to value after developing software by doing it and helping others do it.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
Behind the Agile manifesto: What the authors came to value after developing software by doing it and helping others do it. Manifesto for Agile Software Development: Individuals and interactions over processes and tools Manifesto for Agile Software Development: Working software over comprehensive documentati

Original toplevel document

Unknown title
Manifesto for Agile Software Development Manifesto for Agile Software Development We are uncovering better ways of developing software by doing it and helping others do it. Through this work we have come to value: Individuals and interactions over processes and tools Working software over comprehensive documentation Customer collaboration over contract negotiation Responding to change over following a plan That is, while there is value in the items on the right, we value the items on the left more. Kent Beck Mike Beedle Arie van Bennekum Alistair Cockburn Ward Cunningham Martin Fowler James Grenning Jim Highsmith Andrew Hunt Ron Jeffries Jon Kern Brian Marick Robert C. Martin Stev




#knowledge-base-construction #machine-learning #unfinished
Due to the significantly increased context needed for KBC from richly formatted data, na ¨ ıvely materializing all possible can- didates is intractable as the number of candidates grows combina- torially with the number of relation arguments. This combinatorial explosion can lead to performance issues for KBC systems.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
just 100 documents can gen- erate over 1M candidates. In addition, we find that the majority of these candidates do not express true relations, creating a significant class imbalance that can hinder learning performance [19].

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
example, in the ELECTRONICS domain, just 100 documents can gen- erate over 1M candidates. In addition, we find that the majority of these candidates do not express true relations, creating a significant class imbalance that can hinder learning performance [19].

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
To address this combinatorial explosion, Fonduer allows users to specify throttlers, in addition to matchers, to prune away excess candidates. We find that throttlers must: • Maintain high accuracy by only filtering negative candi- dates. • Seek high coverage of the candidates. Throttlers can be viewed as a knob that allows users to trade off precision and recall and promote scalability by reducing the number of candidates to be classified during KBC.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
increased throttling does not monotonically improve quality since it hurts recall.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
as a rule of thumb, we recommend that users apply throttlers to balance negative and positive candidates. Fonduer provides users with mechanisms to evaluate this balance over a small holdout set of labeled candidates.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The Supreme Court also ruled that the arrest and detention in the first part of Article 22 do not cover arrest under the orders of a court, civil arrest, arrest on failure to pay the income tax, and deportation of an alien. They apply only to an act of a criminal or quasi-criminal nature or some activity prejudicial to public interest.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
In cases where tabular or structural features are noisy—including nearly all documents converted from PDF to HTML by generic tools—visual features can provide a complemen- tary view of the dependencies among text.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
Visual features encode many highly predictive types of semantic information implicitly, such as position on a page, which may imply when text is a title or header.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The 44th Amendment Act of 1978 has reduced the period of detention without obtaining the opinion of an advisory board from three to two months. However, this provision has not yet been brought into force, hence, the original period of three months still continues.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #machine-learning #unfinished
Data programming theory (see Appendix A.2) shows that, with a suf- ficient number of labeling functions, data programming can still achieve quality comparable to using manually labeled data.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




preventive detention existed even during the British rule.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#machine-learning #management #software-engineering #unfinished
Google has been able to hire a tal- ented team across the entire engineer- ing operation.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
Google has been able to hire a tal- ented team across the entire engineer- ing operation. This gives us the op- portunity to innovate everywhere, and for people to move between projects, whether they be primarily research or primarily engineering.

Original toplevel document (pdf)

cannot see any pdfs




#machine-learning #management #software-engineering #unfinished
Having a talented team has given Google the op- portunity to innovate everywhere, and for people to move between projects, whether they be primarily research or primarily engineering.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
Google has been able to hire a tal- ented team across the entire engineer- ing operation. This gives us the op- portunity to innovate everywhere, and for people to move between projects, whether they be primarily research or primarily engineering.

Original toplevel document (pdf)

cannot see any pdfs




Further, the Parliament (under Article 32) can empower any other court to issue these writs. Since no such provision has been made so far, only the Supreme Court and the high courts can issue the writs and not any other court.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The Supreme Court can issue writs only for the enforcement of fundamental rights whereas a high court can issue writs not only for the enforcement of Fundamental Rights but also for any other purpose. The expression ‘for any other purpose’ refers to the enforcement of an ordinary legal right. Thus, the writ jurisdiction of the Supreme Court, in this respect, is narrower than that of high court. 2. The Supreme Court can issue writs against a person or government throughout the territory of India whereas a high court can issue writs against a person residing or against a government or authority located within its territorial jurisdiction only or outside its territorial jurisdiction only if the cause of action arises within its territorial jurisdiction. 15 Thus, the territorial jurisdiction of the Supreme Court for the purpose of issuing writs is wider than that of a high court. 3. A remedy under Article 32 is in itself a Fundamental Right and hence, the Supreme Court may not refuse to exercise its writ jurisdiction.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




I've worked in businesses and industries where the process is designed around the project/problem/business. In fact, that's pretty much the norm in many industries.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
I've worked in businesses and industries where the process is designed around the project/problem/business. In fact, that's pretty much the norm in many industries. Ultimately Agile isn't agile because it's only good for a very particular type of problem, where complexity is low, and the goal is iterative refinement, and deadlines/budgets aren't fi

Original toplevel document

AGILE must be destroyed, once and for all - Erik Meijer : programming
says they saw a rainbow in this business, you'll soon lose sight of what actually matters: making useful and/or entertaining stuff. Continue this thread level 2 DevIceMan 2 points · 4 years ago <span>I've worked in businesses and industries where the process is designed around the project/problem/business. In fact, that's pretty much the norm in many industries. Ultimately Agile isn't agile because it's only good for a very particular type of problem, where complexity is low, and the goal is iterative refinement, and deadlines/budgets aren't fixed. Even then, Agile might not be the best practice. Before I started my Software-Dev career, and while applying places, I thought Agile was some kind of professional way programmers worked together, perhaps resolving merge conflicts, or




Ultimately Agile isn't agile because it's only good for a very particular type of problem, where complexity is low, and the goal is iterative refinement, and deadlines/budgets aren't fixed. Even then, Agile might not be the best practice.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
I've worked in businesses and industries where the process is designed around the project/problem/business. In fact, that's pretty much the norm in many industries. Ultimately Agile isn't agile because it's only good for a very particular type of problem, where complexity is low, and the goal is iterative refinement, and deadlines/budgets aren't fixed. Even then, Agile might not be the best practice.

Original toplevel document

AGILE must be destroyed, once and for all - Erik Meijer : programming
says they saw a rainbow in this business, you'll soon lose sight of what actually matters: making useful and/or entertaining stuff. Continue this thread level 2 DevIceMan 2 points · 4 years ago <span>I've worked in businesses and industries where the process is designed around the project/problem/business. In fact, that's pretty much the norm in many industries. Ultimately Agile isn't agile because it's only good for a very particular type of problem, where complexity is low, and the goal is iterative refinement, and deadlines/budgets aren't fixed. Even then, Agile might not be the best practice. Before I started my Software-Dev career, and while applying places, I thought Agile was some kind of professional way programmers worked together, perhaps resolving merge conflicts, or




Flashcard 5001170259212

Tags
#bert #nlp #unfinished
Question
There are two steps in the BERT framework: [...] and fine-tuning.
Answer
pre-training


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
There are two steps in the BERT framework: pre-training and fine-tuning.

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001171832076

Tags
#bert #nlp #unfinished
Question
There are two steps in the BERT framework: pre-training and [...].
Answer
fine-tuning


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
There are two steps in the BERT framework: pre-training and fine-tuning.

Original toplevel document (pdf)

cannot see any pdfs







#machine-learning #management #software-engineering #unfinished
By coupling research and de- velopment, Google's goal is to minimize or even eliminate the traditional technol- ogy transfer process.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
By coupling research and de- velopment, Google's goal is to minimize or even eliminate the traditional technol- ogy transfer process. The traditional technology transfer process has proven challenging at other companies. Most of Google's projects involve people work- ing with a given technology from the research stage

Original toplevel document (pdf)

cannot see any pdfs




#machine-learning #management #software-engineering #unfinished
The traditional technology transfer process has proven challenging at other companies.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
By coupling research and de- velopment, Google's goal is to minimize or even eliminate the traditional technol- ogy transfer process. The traditional technology transfer process has proven challenging at other companies. Most of Google's projects involve people work- ing with a given technology from the research stage through to the product stage.

Original toplevel document (pdf)

cannot see any pdfs




#machine-learning #management #software-engineering #unfinished
Most of Google's projects involve people work- ing with a given technology from the research stage through to the product stage.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
- velopment, Google's goal is to minimize or even eliminate the traditional technol- ogy transfer process. The traditional technology transfer process has proven challenging at other companies. <span>Most of Google's projects involve people work- ing with a given technology from the research stage through to the product stage. <span>

Original toplevel document (pdf)

cannot see any pdfs




The power to make laws under Article 33 is conferred only on Parliament and not on state legislatures

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#ml #snorkel
Aim of Babble Labble: Can we use natural language as a form of weak supervision?

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
Aim of Babble Labble : Can we use natural language as a form of weak supervision? Natural language supervision would involve parsing the semantics and then using these as labeling functions.

Original toplevel document

Ratner_et_al-2017-dawn,cs,stanford,edu-Snorkel_and_The_Dawn
e extracting structured data from data that has some structure such as tables embedded in PDFs and webpages? We’ve been working on a system called Fonduer to make this fast and easy in Snorkel! <span>Learning from natural language supervision : Can we use natural language as a form of weak supervision, parsing the semantics of natural language statements and then using these as labeling functions? We’ve done some exciting preliminary work here! <span>




#ml #snorkel
Natural language supervision would involve parsing the semantics and then using these as labeling functions.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
Aim of Babble Labble : Can we use natural language as a form of weak supervision? Natural language supervision would involve parsing the semantics and then using these as labeling functions.

Original toplevel document

Ratner_et_al-2017-dawn,cs,stanford,edu-Snorkel_and_The_Dawn
e extracting structured data from data that has some structure such as tables embedded in PDFs and webpages? We’ve been working on a system called Fonduer to make this fast and easy in Snorkel! <span>Learning from natural language supervision : Can we use natural language as a form of weak supervision, parsing the semantics of natural language statements and then using these as labeling functions? We’ve done some exciting preliminary work here! <span>




Article 34 provides for the restrictions on fundamental rights while martial law is in force in any area within the territory of India.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




the expression ‘martial law’ has not been defined anywhere in the Constitution.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Article 35 contains the following provisions: 1. The Parliament shall have (and the legislature of a state shall not have) power to make laws with respect to the following matters: (a) Prescribing residence as a condition for certain employments or appointments in a state or union territory or local authority or other authority (Article 16). (b) Empowering courts other than the Supreme Court and the high courts to issue directions, orders and writs of all kinds for the enforcement of fundamental rights (Article 32). (c) Restricting or abrogating the application of Fundamental Rights to members of armed forces, police forces, etc. (Article 33). (d) Indemnifying any government servant or any other person for any act done during the operation of martial law in any area (Article 34). 2. Parliament shall have (and the legislature of a state shall not have) powers to make laws for prescribing punishment for those acts that are declared to be

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Originally, the right to property was one of the seven fundamental rights under Part III of the Constitution. It was dealt by Article 19(1)(f) and Article 31. Article 19(1)(f) guaranteed to every citizen the right to acquire, hold and dispose of property. Article 31, on the other hand, guaranteed to every person, whether citizen or non-citizen, right against deprivation of his property.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




new Article 300A in Part XII under the heading ‘Right to Property’. It provides that no person shall be deprived of his property except by authority of law. Thus, the right to property still remains a legal right or a constitutional right, though no longer a fundamental right. It is not a part of the basic structure of the Constitution.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The court held that judicial review is a ‘basic feature’ of the constitution and it could not be taken away by putting a law under the Ninth Schedule.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




We find that the model represents the steps of the traditional NLP pipeline in an interpretable and localizable way, and that the regions responsible for each step appear in the expected sequence: POS tagging, parsing, NER, semantic roles, then coreference. Qualitative analysis reveals that the model can and often does adjust this pipeline dynamically, revising lower-level decisions on the basis of disambiguating information from higher-level representations.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

[1905.05950] BERT Rediscovers the Classical NLP Pipeline
ed text encoders have rapidly advanced the state of the art on many NLP tasks. We focus on one such model, BERT, and aim to quantify where linguistic information is captured within the network. <span>We find that the model represents the steps of the traditional NLP pipeline in an interpretable and localizable way, and that the regions responsible for each step appear in the expected sequence: POS tagging, parsing, NER, semantic roles, then coreference. Qualitative analysis reveals that the model can and often does adjust this pipeline dynamically, revising lower-level decisions on the basis of disambiguating information from higher-level representations. Comments: Presented at ACL 2019 Subjects: Computation and Language (cs.CL) Journal reference: Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics (20




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

  • L'anorexie est caractérisée par une diminution, souvent rapide, de l'indice de masse corporelle (IMC).
  • L'hyperphagie boulimique entraîne au contraire une augmentation trop rapide de l'IMC, avec changement de couloir.
  • Dans la boulimie nerveuse, l'IMC est relativement stable.

Le passage d'un TCA à l'autre est relativement fréquent

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale

  • Sa prévalence est estimée à 1 % chez les femmes et 0,1 % chez les hommes.
  • Les formes subsyndromiques sont plus fréquentes.
  • Les premiers symptômes apparaissent fréquemment au cours de l'adolescence

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale

  • La recherche systématique par une question du type « Est-ce que tu as un problème avec ton alimentation ? » ou en utilisant le questionnaire DFTCA

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

En cas d'aménorrhée secondaire ou de cassure pondérale chez une adolescente → rechercher une restriction alimentaire.

Dépistage par le questionnaire DFTCA

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale

Questionnaire DFTCA. (Adaptation du SCOFF anglophone)

Deux réponses positives sont fortement prédictives d'un trouble du comportement alimentaire

  • Te fais-tu vomir parce que tu te sens mal d'avoir trop mangé ?
  • T'inquiètes-tu d'avoir perdu le contrôle de ce que tu manges ?
  • As-tu récemment perdu plus de 6 kg en 3 mois ?
  • Penses-tu que tu es gros(se) alors que d'autres te trouvent trop mince ?
  • Dirais-tu que la nourriture domine ta vie ?

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

On distingue deux types d'anorexie mentale :

  • L'anorexie restrictive
  • L'anorexie de type accès hyperphagiques/purgatifs

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Le type est dit restrictif (par opposition au type accès hyperphagiques/purgatifs ) si au cours des 3 derniers mois, la personne :

  • N'a pas eu d'accès récurrents d'hyperphagie (gloutonnerie)
  • Ni recouru à des vomissements provoqués ou des comportements de purge (laxatifs, diurétiques, lavements)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Intent classification is the automated association of text to a specific purpose or goal.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Intent Classification: How to Identify What Customers Want
k] [imagelink] Subscribe Create new value from your data Try MonkeyLearn Subscribe Try MonkeyLearn Intent Classification: How to Identify What Customers Want [emptylink] [emptylink] [emptylink] <span>Intent classification is the automated association of text to a specific purpose or goal. In essence, a classifier analyzes pieces of text and categorizes them into intents such as Purchase, Downgrade, Unsubscribe, and Demo Request. This is useful to understand the intention




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale

Sur le plan social, il existe souvent un appauvrissement relationnel et un surinvestissement dans la scolarité sur un mode anxieux.

Les relations familiales sont tendues : le patient cherche à contrôler son environnement, veut souvent faire les courses, la cuisine.

Les calories sont parfois comptées.
La balance est généralement omniprésente.

Un déni concernant la gravité de la situation est fréquent, affiché ou réel

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale

S'associent généralement à la perte de poids une aménorrhée (sauf si anorexie préménarchale, prise de pilule ou garçon !), une dysmorphophobie (le ventre et les cuisses sont souvent perçus comme trop gros) et une peur de reprendre du poids

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale - Examen clinique :

Sont particulièrement importants :

  • La recherche d'une hypotension orthostatique
  • L'évaluation de l'état cutané
    • Lanugo
    • Acrocyanose
    • Purpura
    • Hématomes lombaires liés à des séances d'abdominaux
    • Excoriations volontaires
  • Le degré de fonte adipeuse et musculaire
  • L'évaluation d'une constipation

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale

Les principaux diagnostics différentiels différentiels à évoquer sont :

  • Les maladies inflammatoires du tube digestif
  • L'hyperthyroïdie
  • Le diabète insipide
  • Le diabète insulinodépendant
  • Une tumeur du SNC
  • Une achalasie de l'œsophage
  • Une dépression avec perte d'appétit

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale

Bilan paraclinique minimal :

  • NFS-plaquettes (diminution possible des trois lignées ; la neutropénie est la plus habituelle)
  • Ionogramme sanguin avec glycémie, urée, créatininémie, phosphorémie
    • Hypoglycémie d'adaptation
    • Alcalose hypochlorémique et hypokaliémie possibles en cas de vomissements provoqués
    • Insuffisance rénale fonctionnelle
    • Hyponatrémie en cas d'intoxication à l'eau
  • Bilan hépatique (cytolyse ?)
  • CRP (normale)
  • ECG (troubles ioniques, mesure du QTc).

Un dosage de TSH n'est pas recommandé sans signes d'appel

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Une échographie cardiaque peut retrouver un épanchement péricardique et un prolapsus mitral ; elle n'est pas systématique.

Une ostéodensitométrie est indiquée en cas d'aménorrhée supérieure à 6 mois (forme chronique, début de surveillance osseuse).

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Because BERT is a pretrained model that expects input data in a specific format, we will need:

  • special tokens to mark the beginning ([CLS]) and separation/end of sentences ([SEP])
  • tokens that conforms with the fixed vocabulary used in BERT
  • token IDs from BERT’s tokenizer
  • mask IDs to indicate which elements in the sequence are tokens and which are padding elements
  • segment IDs used to distinguish different sentences
  • positional embeddings used to show token position within the sequence

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Unknown title
# Load pre-trained model tokenizer (vocabulary) tokenizer = BertTokenizer.from_pretrained('bert-base-uncased') 100%|██████████| 231508/231508 [00:00<00:00, 416614.61B/s] 2. Input Formatting <span>Because BERT is a pretrained model that expects input data in a specific format, we will need: special tokens to mark the beginning ([CLS]) and separation/end of sentences ([SEP]) tokens that conforms with the fixed vocabulary used in BERT token IDs from BERT’s tokenizer mask IDs to indicate which elements in the sequence are tokens and which are padding elements segment IDs used to distinguish different sentences positional embeddings used to show token position within the sequence Luckily, this interface takes care of some of these input specifications for us so we will only have to manually create a few of them (we’ll revisit the other inputs in another tutorial




a naive cross-product of all `Mentions` of text in a document would result in an intractable, combinatorial explosion of candidates. Matchers serve to limit the number of `Mentions` generated. Matchers operate on individual `Spans`.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Unknown title
on` classes we want to extract, the next step is generating those candidates. To do so, Fonduer allows users to provide _matchers_. These inputs are valuable because with richly formatted data, <span>a naive cross-product of all `Mentions` of text in a document would result in an intractable, combinatorial explosion of candidates. Matchers serve to limit the number of `Mentions` generated. Matchers operate on individual `Spans`.\n", "\n", "### Matchers\n", "One convenient way to think about matchers is to think of them as a way to define what each component of your relation is. In our example, we can provide a




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale - Citères de gravité

Cliniques :

  • IMC < 12,7 kg/m 2 à 13–14 ans
  • IMC < 13,2 kg/m 2 à 15–16 ans
  • IMC < 14 kg/m 2 au-delà de 17 ans

  • Fatigabilité voire épuisement, ralentissement idéique et verbal, confusion

  • Bradycardies extrêmes (FC < 40/min), tachycardies
  • PA < 80/50 mm Hg, hypotension orthostatique ou tachycardie orthostatique

  • Hypothermie < 35,5 °C, hyperthermie

  • Syndrome occlusif

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale - Critères de gravité

Paracliniques :

  • Acétonurie à la BU, hypoglycémie < 0,6 g/L
  • Hypokaliémie, hyponatrémie, hypophosphorémie, hypomagnésémie
  • Élévation de la créatininémie (> 100 μmol/L)
  • Cytolyse hépatique (ASAT) > 4 N
  • Leuconeutropénie < 1 000/mm3 , thrombopénie < 60 000/mm3

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale - Critères de gravité

Anamnestiques :

  • Perte de poids rapide : > 2 kg/semaine
  • Aphagie totale, refus de boire
  • Lipothymies ou malaises d'allure orthostatique
  • Fatigabilité voire épuisement évoqué par le patient

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

L'hospitalisation sous contrainte ne doit être utilisée que lorsque le risque vital est engagé et que des soins consentis sont impossibles.

On rappelle qu'il n'existe pas de soins psychiatriques à la demande de tiers (SPDT, qui a remplacé l'HDT) chez les mineurs.
Un signalement au procureur avec demande d'OPP sur l'hôpital doit rester exceptionnel

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Anorexie mentale

Prise en charge médicale et nutritionnelle attentive :

  • Reprise progressive d'une alimentation diversifiée et normocalorique
  • Suppléments nutritionnels parfois utiles
  • Supplémentation en phosphore les premiers jours de réalimentation, vitamine D

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Prise en charge psychothérapeutique :

  • Thérapies de soutien, psychothérapies d'inspiration analytique, comportementales
  • Thérapies familiales (seule psychothérapie ayant montré une efficacité dans des essais randomisés), groupes de parole parents, groupes de parole adolescents

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Premières mesures aux urgences en cas de dénutrition grave

  • Repos au lit, réchauffement.
  • Supplémentation initiale et systématique en micronutriments : phosphore, polyvitamines, oligoéléments.

  • Si IMC < 12 et/ou ralentissement psychomoteur et/ou désordres hydroélectriques : perfusion de sérum physiologique + 2 g/L de KCl (ne pas apporter de glucose par voie IV).

  • Réintroduction alimentaire progressive et fractionnée en cinq repas avec poursuite des apports en micronutriments.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles
Devant des critères de gravité et l'absence d'alimentation spontanée, une nutrition entérale (NE) par sonde peut être prescrite (mélange isocalorique 1 mL = 1 kcal), administrée à débit continu sur 12 heures ou 24 heures, par pompe, en augmentant progressivement les apports

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Prise en charge des hypoglycémies (aphagie, syndrome de renutrition) :

  • Surveiller les glycémies capillaires systématiquement à l'admission puis toutes les 8 heures, si IMC < 12.
  • Devant tout malaise chez une patiente ayant un TCA, un épisode hypoglycémique doit être suspecté.
    La mesure de la glycémie ne doit pas retarder le resucrage.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

En l'absence de troubles de conscience, resucrage per os (10 g de sucre rapide).

En cas d'impossibilité de resucrage per os (refus de la patiente, troubles de conscience) :

  • Resucrage prioritairement par voie entérale (bolus de 100 mL d'un soluté isocalorique puis 500 mL/24 heures en continu)
  • Si la voie entérale est impossible, injecter 1 ampoule de G10 % en IV qui sera obligatoirement suivie de la mise en place d'une perfusion de type glucosé à 5 % avec KCl 2 g/L et NaCl 4 g/L, à la place du sérum physiologique, pour éviter les rebonds d'hypoglycémies.

Le glucagon est inefficace

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Prévention et traitement de l'hypophosphorémie (syndrome de renutrition)

  • Une hypophosphorémie inférieure à 0,6 mmol/L impose l'arrêt temporaire de l'alimentation et la correction de la phosphorémie avant la reprise de la nutrition, qui se fera en poursuivant la supplémentation phosphorée

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Hypokaliémie modérée de 2,5 à 3 mmol/L et absence de signes ECG :

  • Surveillance scopique
  • Perfusion de chlorure de potassium 10 % sur 3 heures associée à une ampoule de chlorure de magnésium
  • Contrôle du ionogramme 1 heure après la fin de la perfusion (à répéter si nécessaire)

Hypokaliémie sévère < 2,5 mmol/L et/ou signes ECG :

  • Transfert en réanimation ou unité de soins continus
  • Traitement identique à celui de l'hypokaliémie modérée

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Évolution

  • L'évolution à 5 ans est favorable, avec deux tiers des patients guéris, qui n'ont plus de TCA.
  • La mortalité est de l'ordre de 0,5 % par année d'évolution (suicide, dénutrition extrême, hypokaliémie, syndrome de renutrition).
  • L'anorexie du garçon a une évolution similaire

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#latent-features #nlp
e propose a novel end-to-end recurrent neural model which incorporates an entity-aware attention mechanism with a latent entity typing (LET) method.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#latent-features #nlp
s more interpretable by visualizing attention mechanisms applied to our model and results of LET.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#latent-features #nlp
Recently, many studies therefore propose end-to- end neural models without the high-level features. Among them, attention-based models, which focus to the most important semantic information in a sentence, show state-of-the-art results in a lot of NLP tasks. Since these models are mainly proposed for solving translation and language modeling tasks, they could not fully utilize the information of tagged entities in relation classifica- tion task. However, tagged entity pairs could be pow- erful hints for solving relation classification task. For example, even if we do not consider other words except the crash and attack, we intuitively know that the en- tity pair has a relation Cause-Effect(e1,e2) 1 better than Component-Whole(e1,e2) 1 in Figure

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#latent-features #nlp
To address these issues, We propose

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#latent-features #nlp
Entity-aware attention focuses on the most im- portant semantic information considering entity pairs with word positions relative to these pairs and latent types obtained by LET.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#latent-features #nlp
We propose an novel end-to-end recurrent neural model and an entity-aware attention mechanism with a LET which focuses to semantic information of entities and their latent types;

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#latent-features #nlp
We show that our model is more inter- pretable since it’s decision making process could be vi- sualized with self attention, entity-aware attention, and LET.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#ai-safety #nlp

Explainable attention; using label-specific information, state of the art results

https://arxiv.org/pdf/1911.03875.pdf

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on




#machine-learning #software-engineering #unfinished
Mathematical #abstractions#do#not#leak - When#we#use#a#theorem,#we#do#not#need#to#revisit#its#proof

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
Mathematical #abstractions#do#not#leak - When#we#use#a#theorem,#we#do#not#need#to#revisit#its#proof In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, so we Desi gn #wi th#contracts. • In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, In#practice,#th

Original toplevel document (pdf)

cannot see any pdfs




#machine-learning #software-engineering #unfinished
In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, so we Desi gn #wi th#contracts. •

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
Mathematical #abstractions#do#not#leak - When#we#use#a#theorem,#we#do#not#need#to#revisit#its#proof In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, so we Desi gn #wi th#contracts. • In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, In#practice,#there#are#abstraction#leaks#. Abstractions in computer programs leak, but less than those in machine lea

Original toplevel document (pdf)

cannot see any pdfs




#machine-learning #software-engineering #unfinished
In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, In#practice,#there#are#abstraction#leaks#.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
ical #abstractions#do#not#leak - When#we#use#a#theorem,#we#do#not#need#to#revisit#its#proof In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, so we Desi gn #wi th#contracts. • <span>In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, In#practice,#there#are#abstraction#leaks#. Abstractions in computer programs leak, but less than those in machine learning Desi gn #wi th#contracts – kn owi ng#th e#specification#of#a#module#is#enou gh. <span>

Original toplevel document (pdf)

cannot see any pdfs




#machine-learning #software-engineering #unfinished
Abstractions in computer programs leak, but less than those in machine learning

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
r#programs#are#similar#to#mathematic al#proo fs, so we Desi gn #wi th#contracts. • In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, In#practice,#there#are#abstraction#leaks#. <span>Abstractions in computer programs leak, but less than those in machine learning Desi gn #wi th#contracts – kn owi ng#th e#specification#of#a#module#is#enou gh. <span>

Original toplevel document (pdf)

cannot see any pdfs




#machine-learning #software-engineering #unfinished
Desi gn #wi th#contracts – kn owi ng#th e#specification#of#a#module#is#enou gh.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
• In theory, Compu ter#programs#are#similar#to#mathematic al#proo fs, In#practice,#there#are#abstraction#leaks#. Abstractions in computer programs leak, but less than those in machine learning <span>Desi gn #wi th#contracts – kn owi ng#th e#specification#of#a#module#is#enou gh. <span>

Original toplevel document (pdf)

cannot see any pdfs




#nlp #soft-skills
There is so much plagiarism/sentence-borrowing in recent NLP papers (2019+, say). The medium of an academic paper to present results seems like a thin-layered formality.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on




#bert
The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence in BERT is the special classification token – [CLS]. BERT: The [CLS] token is used in classification tasks as an aggregate of the entire sequence

Original toplevel document

BERT Explained – A list of Frequently Asked Questions – Let the Machines Learn
ctives allow it to be used on any single sequence and sequence-pair tasks without substantial task-specific architecture modifications. How is the input text represented before feeding to BERT? <span>The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence is the special classification token – [CLS]. This token is used in classification tasks as an aggregate of the entire sequence representation. It is ignored in non-classification tasks. For single text sentence tasks, this [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. Single Sentence Input For sentence pair tasks, the WordPiece tokens of the two sentences are separated by another [SEP] token. This input sequence also ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A positional embedding is also added to each token to indicate its position in the sequence. Which Tokenization strategy is used by BERT? BERT uses WordPiece tokenization. The vocabulary is initialized with all the individual characters in the language, and then the most freque




#bert
The first token of every input sequence in BERT is the special classification token – [CLS].

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence in BERT is the special classification token – [CLS]. BERT: The [CLS] token is used in classification tasks as an aggregate of the entire sequence representation. BERT: The [CLS] token is ignored in non-classification tasks. BERT: For sing

Original toplevel document

BERT Explained – A list of Frequently Asked Questions – Let the Machines Learn
ctives allow it to be used on any single sequence and sequence-pair tasks without substantial task-specific architecture modifications. How is the input text represented before feeding to BERT? <span>The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence is the special classification token – [CLS]. This token is used in classification tasks as an aggregate of the entire sequence representation. It is ignored in non-classification tasks. For single text sentence tasks, this [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. Single Sentence Input For sentence pair tasks, the WordPiece tokens of the two sentences are separated by another [SEP] token. This input sequence also ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A positional embedding is also added to each token to indicate its position in the sequence. Which Tokenization strategy is used by BERT? BERT uses WordPiece tokenization. The vocabulary is initialized with all the individual characters in the language, and then the most freque




#bert
BERT: The [CLS] token is used in classification tasks as an aggregate of the entire sequence representation.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
xt sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence in BERT is the special classification token – [CLS]. <span>BERT: The [CLS] token is used in classification tasks as an aggregate of the entire sequence representation. BERT: The [CLS] token is ignored in non-classification tasks. BERT: For single text sentence tasks, the [CLS] token is followed by the WordPiece tokens and the separator token – [SEP].

Original toplevel document

BERT Explained – A list of Frequently Asked Questions – Let the Machines Learn
ctives allow it to be used on any single sequence and sequence-pair tasks without substantial task-specific architecture modifications. How is the input text represented before feeding to BERT? <span>The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence is the special classification token – [CLS]. This token is used in classification tasks as an aggregate of the entire sequence representation. It is ignored in non-classification tasks. For single text sentence tasks, this [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. Single Sentence Input For sentence pair tasks, the WordPiece tokens of the two sentences are separated by another [SEP] token. This input sequence also ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A positional embedding is also added to each token to indicate its position in the sequence. Which Tokenization strategy is used by BERT? BERT uses WordPiece tokenization. The vocabulary is initialized with all the individual characters in the language, and then the most freque




#bert
BERT: The [CLS] token is ignored in non-classification tasks.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
t token of every input sequence in BERT is the special classification token – [CLS]. BERT: The [CLS] token is used in classification tasks as an aggregate of the entire sequence representation. <span>BERT: The [CLS] token is ignored in non-classification tasks. BERT: For single text sentence tasks, the [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. BERT: For sentence pair tasks, the WordPiece tokens of the two

Original toplevel document

BERT Explained – A list of Frequently Asked Questions – Let the Machines Learn
ctives allow it to be used on any single sequence and sequence-pair tasks without substantial task-specific architecture modifications. How is the input text represented before feeding to BERT? <span>The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence is the special classification token – [CLS]. This token is used in classification tasks as an aggregate of the entire sequence representation. It is ignored in non-classification tasks. For single text sentence tasks, this [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. Single Sentence Input For sentence pair tasks, the WordPiece tokens of the two sentences are separated by another [SEP] token. This input sequence also ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A positional embedding is also added to each token to indicate its position in the sequence. Which Tokenization strategy is used by BERT? BERT uses WordPiece tokenization. The vocabulary is initialized with all the individual characters in the language, and then the most freque




#bert
BERT: For single text sentence tasks, the [CLS] token is followed by the WordPiece tokens and the separator token – [SEP].

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
ication token – [CLS]. BERT: The [CLS] token is used in classification tasks as an aggregate of the entire sequence representation. BERT: The [CLS] token is ignored in non-classification tasks. <span>BERT: For single text sentence tasks, the [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. BERT: For sentence pair tasks, the WordPiece tokens of the two sentences are separated by a [SEP] token. This i For both single and pair sentence inputs in BERT, the sequence ends with

Original toplevel document

BERT Explained – A list of Frequently Asked Questions – Let the Machines Learn
ctives allow it to be used on any single sequence and sequence-pair tasks without substantial task-specific architecture modifications. How is the input text represented before feeding to BERT? <span>The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence is the special classification token – [CLS]. This token is used in classification tasks as an aggregate of the entire sequence representation. It is ignored in non-classification tasks. For single text sentence tasks, this [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. Single Sentence Input For sentence pair tasks, the WordPiece tokens of the two sentences are separated by another [SEP] token. This input sequence also ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A positional embedding is also added to each token to indicate its position in the sequence. Which Tokenization strategy is used by BERT? BERT uses WordPiece tokenization. The vocabulary is initialized with all the individual characters in the language, and then the most freque




#bert
BERT: For sentence pair tasks, the WordPiece tokens of the two sentences are separated by a [SEP] token.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
ntation. BERT: The [CLS] token is ignored in non-classification tasks. BERT: For single text sentence tasks, the [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. <span>BERT: For sentence pair tasks, the WordPiece tokens of the two sentences are separated by a [SEP] token. This i For both single and pair sentence inputs in BERT, the sequence ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to

Original toplevel document

BERT Explained – A list of Frequently Asked Questions – Let the Machines Learn
ctives allow it to be used on any single sequence and sequence-pair tasks without substantial task-specific architecture modifications. How is the input text represented before feeding to BERT? <span>The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence is the special classification token – [CLS]. This token is used in classification tasks as an aggregate of the entire sequence representation. It is ignored in non-classification tasks. For single text sentence tasks, this [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. Single Sentence Input For sentence pair tasks, the WordPiece tokens of the two sentences are separated by another [SEP] token. This input sequence also ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A positional embedding is also added to each token to indicate its position in the sequence. Which Tokenization strategy is used by BERT? BERT uses WordPiece tokenization. The vocabulary is initialized with all the individual characters in the language, and then the most freque




#bert
For both single and pair sentence inputs in BERT, the sequence ends with the [SEP] token.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
[CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. BERT: For sentence pair tasks, the WordPiece tokens of the two sentences are separated by a [SEP] token. This i <span>For both single and pair sentence inputs in BERT, the sequence ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A p

Original toplevel document

BERT Explained – A list of Frequently Asked Questions – Let the Machines Learn
ctives allow it to be used on any single sequence and sequence-pair tasks without substantial task-specific architecture modifications. How is the input text represented before feeding to BERT? <span>The input representation used by BERT is able to represent a single text sentence as well as a pair of sentences (eg., [Question, Answer]) in a single sequence of tokens. The first token of every input sequence is the special classification token – [CLS]. This token is used in classification tasks as an aggregate of the entire sequence representation. It is ignored in non-classification tasks. For single text sentence tasks, this [CLS] token is followed by the WordPiece tokens and the separator token – [SEP]. Single Sentence Input For sentence pair tasks, the WordPiece tokens of the two sentences are separated by another [SEP] token. This input sequence also ends with the [SEP] token. Sentence Pair Input A sentence embedding indicating Sentence A or Sentence B is added to each token. Sentence embeddings are similar to token/word embeddings with a vocabulary of 2. A positional embedding is also added to each token to indicate its position in the sequence. Which Tokenization strategy is used by BERT? BERT uses WordPiece tokenization. The vocabulary is initialized with all the individual characters in the language, and then the most freque




#knowledge-base-construction #machine-learning #unfinished
Fonduer: Our main contributions are (1) we design a new data model that preserves information across different data modalities. (2) We introduce a multimodal LSTM network that combines textual context with universal features derived from other modalities.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
Technical Contributions Our main contributions are as follows: (1) To account for the breadth of signals in richly formatted data, we design a new data model that preserves structural and semantic information across different data modalities. The role of Fonduer ’s data model is twofold: (a) to allow users to specify multimodal domain knowledge that Fonduer leverages to automate the KBC process over richly formatted data, and (b) to provide Fonduer ’s machine-learning model with the necessary representation to reason about document-wide context (see Section 3). (2) We empirically show that existing deep-learning models [ 46 ] tailored for text information extraction (such as long short-term mem- ory (LSTM) networks [ 18 ]) struggle to capture the multimodality of richly formatted data. We introduce a multimodal LSTM network that combines textual context with universal features that correspond to structural and visual properties of the input documents. These features are inherently captured by Fonduer ’s data model and are generated automatically (see Section 4.2). We also introduce a series of data layout optimizations to ensure the scalability of Fonduer to millions of document-wide candidates (see Appendix C).

Original toplevel document (pdf)

cannot see any pdfs




Flashcard 5001401208076

Tags
#machine-learning #software-engineering #unfinished
Question
Managing [...] data-preparation pipelines, detecting errors and recovering from failures are all difficult and costly [1].
Answer
jungle-like


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all difficult and costly [1].

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001402780940

Tags
#machine-learning #software-engineering #unfinished
Question
Managing jungle-like [...], detecting errors and recovering from failures are all difficult and costly [1].
Answer
data-preparation pipelines


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all difficult and costly [1].

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001404353804

Tags
#machine-learning #software-engineering #unfinished
Question
Managing jungle-like data-preparation pipelines, detecting [...] and recovering from failures are all difficult and costly [1].
Answer
errors


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all difficult and costly [1].

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001405926668

Tags
#machine-learning #software-engineering #unfinished
Question
Managing jungle-like data-preparation pipelines, [...] errors and recovering from failures are all difficult and costly [1].
Answer
detecting


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all difficult and costly [1].

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001407499532

Tags
#machine-learning #software-engineering #unfinished
Question
Managing jungle-like data-preparation pipelines, detecting errors and [...] are all difficult and costly [1].
Answer
recovering from failures


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all difficult and costly [1].

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001410645260

Tags
#machine-learning #software-engineering #unfinished
Question
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all [...] and costly [1].
Answer
difficult


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all difficult and costly [1].

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001412218124

Tags
#machine-learning #software-engineering #unfinished
Question
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all difficult and [...] [1].
Answer
costly


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Managing jungle-like data-preparation pipelines, detecting errors and recovering from failures are all difficult and costly [1].

Original toplevel document (pdf)

cannot see any pdfs







Article 5001491385612

Rosario-2011-Hadoop_fatigue-bytemining,com
#frameworks #has-images

CommentsPosts My thoughts on data mining, machine learning, programming languages, open-source software and general nerdery. « SIGKDD 2011 Conference — Day 1 (Graph Mining and David Blei/Topic Models) My Review of Hadoop Summit 2011 #hadoopsummit » Hadoop Fatigue -- Alternatives to Hadoop It’s been a while since I have posted… in the midst of trying to plow through this dissertation while working on papers for submission to some conferences. Hadoop has become the de facto standard in the research and industry uses of small and large-scale MapReduce. Since its inception, an entire ecosystem has been built around it including conferences (Hadoop World, Hadoop Summit), books, training, and commercial distributions (Cloudera, Hortonworks, MapR) with support. Several projects that integrate with Hadoop have been released from the Apache incubator and are designed for certain use cases: Pig, dev



Directive Principles of State Policy are enumerated in Part IV of the Constitution from Articles 36 to 51 1 . The framers of the Constitution borrowed this idea from the Irish Constitution of 1937

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The phrase ‘Directive Principles of State Policy’ denotes the ideals that the State should keep in mind while formulating policies and enacting laws. These are the constitutional instructions or recommendations to the State in legislative, executive and administrative matters. According to Article 36, the term ‘State’ in Part IV has the same meaning as in Part III dealing with Fundamental Rights. Therefore, it includes the legislative and executive organs of the central and state governments, all local authorities and all other public authorities in the country.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The Directive Principles resemble the ‘Instrument of Instructions’ enumerated in the Government of India Act of 1935. In the words of Dr B R Ambedkar, ‘the Directive Principles are like the instrument of instructions, which were issued to the Governor-General and to the Governors of the colonies of India by the British Government under the Government of India Act of 1935

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Flashcard 5001541455116

Tags
#nlp #observations
Question
Where was Christopher Ré's group before Stanford?
Answer
University of Wisconsin Madison


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill






#knowledge-base-construction #unfinished
Our research group’s hypothesis: “The next breakthrough in data analysis may not be in individual algorithms… But may be in the ability to rapidly combine, deploy, and maintain existing algorithms.”

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #unfinished
Data$Acquisi+on$with$Condor$ 7$ Crawl 400K Youtube Videos, and invoke Google’s Speech API to perform video transcription in 3 days

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #unfinished
Deep$NLP$with$Condor$ 8$ We finish deep linguistic processing (Stanford NLP, Coreference, POS) on 500M web pages (2TB text) within 10 days Using 150K machine hours We leverage thousands of OSG nodes to do deep semantic analysis of 2TB of web pages within 24 hours

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #unfinished
High$Throughput$$ Data$Processing$with$Condor$ 9$ We run parallel SQL join (using Python) over 8TB of TSV data with 5X higher throughput than a 100-node parallel database

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #unfinished
100 Nodes 100 TB X 1000 @ UW-Madison X 100K @ US Open Science Grid X 10 High-end Servers Data Acquisition Deep NLP Statistical Inference Web Serving 500M Webpages 500K Videos 50TB Data 14B structured sentences 3M Entites 7B Mentions 100M Relations Magic Happens! Raw Compute Infrastructure Storage Infrastructure Stats. Infrastructure

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The Directive Principles are non-justiciable in nature, that is, they are not legally enforceable by the courts for their violation. Therefore, the government (Central, state and local) cannot be compelled to implement them. Nevertheless, the Constitution (Article 37) itself says that these principles are fundamental in the governance of the country and it shall be the duty of the State to apply these principles in making laws.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #unfinished
Demo: GeoDeepDive ¤   Help Shanan Peters, Assoc. Prof., Geoscience, enhance a rock formation Database 11 Condor: -  Acquire Articles -  Feature Extraction -  Measurement Extraction We Hope to Answer: What is the carbon record of North America?

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The Supreme Court has ruled many a times that in determining the constitutionality of any law, if a court finds that the law in question seeks to give effect to a Directive Principle, it may consider such law to be ‘reasonable’ in relation to Article 14 (equality before law) or Article 19 (six freedoms) and thus save such law from unconstitutionality.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #unfinished
Demo: AncientText ¤   Help Robin Valenza, Assoc. Prof., English understand 140K books from UK 1700-1900 12 Condor Helps: -  Building Topic Models -  Slice and Dice! -  By Year, Author, … -  Advanced OCR -  Challenge how many alternatives to store?

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#knowledge-base-construction #unfinished
Condor is the key enabling tech across a large number of our projects ¤   Crawling, Feature Extraction, and Data Processing, …. and even Statistical Inference ¤   We started with a Hadoop-based infrastructure but are gradually killing it off.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#frameworks #machine-learning

https://stackoverflow.com/questions/4095664/should-i-prefer-hadoop-vs-condor-when-working-with-r

You can use HDFS for your data sets and Condor for your job scheduling. Using Condor to place executors on machines and HDFS + Hadoops Map-Reduce features to process your data (assuming your problem is map-reduce mappable). Then you're using the most appropriate tool for the job: Condor is a job scheduler, and as such does that work better than Hadoop. And Hadoop's HDFS and M-R framework are things Condor doesn't have (but are really helpful for jobs running on Condor to use).

I would personally look at has HDFS to share data among jobs that run discretely as Condor jobs. Especially in a university environment, where shared compute resources are not 100% reliable and can come and go at will, Condor's resilience in this type of set up is going to make getting work done a whole lot easier.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on




#machine-learning #software-engineering #unfinished
If a problem does not have a clear specfication, it is not amenable#to#provable#algorithms.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
If a problem does not have a clear specfication, it is not amenable#to#provable#algorithms. Mac hin e#learnin g#replaces a missing specification by lots of data. As aspects#of#the#da ta#cannot#be#descri bed#by#a#compact# mathematical# statemen, experimentation#is# necessary! <

Original toplevel document (pdf)

cannot see any pdfs




#machine-learning #software-engineering #unfinished
Mac hin e#learnin g#replaces a missing specification by lots of data. As aspects#of#the#da ta#cannot#be#descri bed#by#a#compact# mathematical# statemen, experimentation#is# necessary!

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
If a problem does not have a clear specfication, it is not amenable#to#provable#algorithms. Mac hin e#learnin g#replaces a missing specification by lots of data. As aspects#of#the#da ta#cannot#be#descri bed#by#a#compact# mathematical# statemen, experimentation#is# necessary!

Original toplevel document (pdf)

cannot see any pdfs




Flashcard 5001569242380

Question
I argue that we have to recognize that platforms can have the power to not only [...] content but also to [...] language itself .
Answer
moderate


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself .

Original toplevel document

How Automated Tools Discriminate Against Black Language – MIT Center for Civic Media
vists while doing little about the accounts of white supremacists. There are well-documented problems of content moderation on social media platforms , but as we work to address these problems, <span>I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself. Implicit in Perspective’s bias against AAVE is the notion that the language used by many Black Americans is less acceptable, less intellectual, and less respectable than the language us







Flashcard 5001570815244

Question
I argue that we have to recognize that platforms can have the power to not only moderate [...] but also to moderate language itself .
Answer
content


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself .

Original toplevel document

How Automated Tools Discriminate Against Black Language – MIT Center for Civic Media
vists while doing little about the accounts of white supremacists. There are well-documented problems of content moderation on social media platforms , but as we work to address these problems, <span>I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself. Implicit in Perspective’s bias against AAVE is the notion that the language used by many Black Americans is less acceptable, less intellectual, and less respectable than the language us







Flashcard 5001572388108

Question
I argue that we have to recognize that platforms can have the power to not only [...] content but also to [...] language itself .
Answer
moderate


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself .

Original toplevel document

How Automated Tools Discriminate Against Black Language – MIT Center for Civic Media
vists while doing little about the accounts of white supremacists. There are well-documented problems of content moderation on social media platforms , but as we work to address these problems, <span>I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself. Implicit in Perspective’s bias against AAVE is the notion that the language used by many Black Americans is less acceptable, less intellectual, and less respectable than the language us







Flashcard 5001573960972

Question
I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate [...] .
Answer
language itself


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself .

Original toplevel document

How Automated Tools Discriminate Against Black Language – MIT Center for Civic Media
vists while doing little about the accounts of white supremacists. There are well-documented problems of content moderation on social media platforms , but as we work to address these problems, <span>I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself. Implicit in Perspective’s bias against AAVE is the notion that the language used by many Black Americans is less acceptable, less intellectual, and less respectable than the language us







Flashcard 5001577106700

Question
I argue that we have to recognize that platforms can have [...] to not only moderate content but also to moderate language itself .
Answer
the power


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself .

Original toplevel document

How Automated Tools Discriminate Against Black Language – MIT Center for Civic Media
vists while doing little about the accounts of white supremacists. There are well-documented problems of content moderation on social media platforms , but as we work to address these problems, <span>I argue that we have to recognize that platforms can have the power to not only moderate content but also to moderate language itself. Implicit in Perspective’s bias against AAVE is the notion that the language used by many Black Americans is less acceptable, less intellectual, and less respectable than the language us







Flashcard 5001579465996

Tags
#nlp #reading-group #transformer #unfinished
Question
Transformers: As the model processes each word (each position in the input sequence), [...] allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word.
Answer
self attention


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Transformers: As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word.

Original toplevel document

Alammar-2018-The_Illustrated_Transformer-jalammar,github,io
eet or to the animal? It’s a simple question to a human, but not as simple to an algorithm. When the model is processing the word “it”, self-attention allows it to associate “it” with “animal”. <span>As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word. If you’re familiar with RNNs, think of how maintaining a hidden state allows an RNN to incorporate its representation of previous words/vectors it has processed with the current one it’







Flashcard 5001581038860

Tags
#nlp #reading-group #transformer #unfinished
Question
Transformers: As the model processes each word (each position in the input sequence), self attention allows it to look at [...] for clues that can help lead to a better encoding for this word.
Answer
look at other positions in the input sequence


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Transformers: As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word.

Original toplevel document

Alammar-2018-The_Illustrated_Transformer-jalammar,github,io
eet or to the animal? It’s a simple question to a human, but not as simple to an algorithm. When the model is processing the word “it”, self-attention allows it to associate “it” with “animal”. <span>As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word. If you’re familiar with RNNs, think of how maintaining a hidden state allows an RNN to incorporate its representation of previous words/vectors it has processed with the current one it’







Flashcard 5001584184588

Tags
#nlp #reading-group #transformer #unfinished
Question
Transformers: As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can [...] this word.
Answer
help lead to a better encoding for


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Transformers: As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word.

Original toplevel document

Alammar-2018-The_Illustrated_Transformer-jalammar,github,io
eet or to the animal? It’s a simple question to a human, but not as simple to an algorithm. When the model is processing the word “it”, self-attention allows it to associate “it” with “animal”. <span>As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word. If you’re familiar with RNNs, think of how maintaining a hidden state allows an RNN to incorporate its representation of previous words/vectors it has processed with the current one it’







Flashcard 5001587330316

Tags
#nlp #reading-group #transformer #unfinished
Question
Transformers: As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for [...].
Answer
this word


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for <span>this word. <span>

Original toplevel document

Alammar-2018-The_Illustrated_Transformer-jalammar,github,io
eet or to the animal? It’s a simple question to a human, but not as simple to an algorithm. When the model is processing the word “it”, self-attention allows it to associate “it” with “animal”. <span>As the model processes each word (each position in the input sequence), self attention allows it to look at other positions in the input sequence for clues that can help lead to a better encoding for this word. If you’re familiar with RNNs, think of how maintaining a hidden state allows an RNN to incorporate its representation of previous words/vectors it has processed with the current one it’







#bert #knowledge-base-construction #nlp #unfinished
Suppose vectors H i to H j are the final hidden state vectors from BERT for entity e 1 , and H k to H m are the final hidden state vectors from BERT for entity e 2 . We apply the average opera- tion to get a vector representation for each of the two target entities.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #knowledge-base-construction #nlp #unfinished
Then after an activation opera- tion (i.e. tanh), we add a fully connected layer to each of the two vectors, and the output for e 1 and e 2 are H 0 1 and H 0 2 respectively.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #knowledge-base-construction #nlp #unfinished
This process can be mathematically formalized as Equation (1). H 0 1 = W 1 " tanh 1 j − i + 1 j X t=i H t !# + b 1 H 0 2 = W 2 " tanh 1 m − k + 1 m X t=k H t !# + b 2 (1)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #knowledge-base-construction #nlp #unfinished
For the final hidden state vector of the first token (i.e. ‘[CLS]’), we also add an activa- tion operation and a fully connected layer, which is formally expressed as: H 0 0 = W 0 (tanh(H 0 )) + b 0 (2)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Let x ∈ ℝ m represent the input to a fully connected layer. Let y i ∈ ℝ be the i -th output from the fully connected layer. Then y i ∈ ℝ is computed as follows:

y i = σ ( w 1 x 1 + ⋯ + w m x m )

Here, σ is a nonlinear function (for now, think of σ as the sigmoid function introduced in the previous chapter), and the w i are learnable parameters in the network.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

4. Fully Connected Deep Networks - TensorFlow for Deep Learning [Book]
layer is represented as follows in Figure 4-1. Figure 4-1. A fully connected layer in a deep network. Let’s dig a little deeper into what the mathematical form of a fully connected network is. <span>Let x ∈ ℝ m represent the input to a fully connected layer. Let y i ∈ ℝ be the i -th output from the fully connected layer. Then y i ∈ ℝ is computed as follows: y i = σ ( w 1 x 1 + ⋯ + w m x m ) Here, σ is a nonlinear function (for now, think of σ as the sigmoid function introduced in the previous chapter), and the w i are learnable parameters in the network. The full output y is then y = σ ( w 1,1 x 1 + ⋯ + w 1,m x m ) ⋮ σ ( w n,1 x 1 + ⋯ + w n,m x m ) Note that it’s directly possible to stack fully connected networks. A network with multip




#bert #knowledge-base-construction #nlp #unfinished
Matrices W 0 , W 1 , W 2 have the same dimensions, i.e. W 0 ∈ R d×d , W 1 ∈ R d×d , W 2 ∈ R d×d , where d is the hidden state size from BERT.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #knowledge-base-construction #nlp #unfinished
We concatenate H 0 0 , H 0 1 , H 0 2 and then add a fully connected layer and a softmax layer, which can be expressed as following: h 00 = W 3 concat H 0 0 , H 0 1 , H 0 2 + b 3 p = softmax(h 00 ) (3) where W 3 ∈ R L×3d (L is the number of relation types), and p is the probability output.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #knowledge-base-construction #nlp #unfinished
We use cross entropy as the loss function

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #knowledge-base-construction #nlp #unfinished
We apply dropout before each fully connected layer during training.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Fully connected networks are the workhorses of deep learning, used for thousands of applications.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

4. Fully Connected Deep Networks - TensorFlow for Deep Learning [Book]
g, plus books, videos, and digital content from 200+ publishers. Start your free trial Chapter 4. Fully Connected Deep Networks This chapter will introduce you to fully connected deep networks. <span>Fully connected networks are the workhorses of deep learning, used for thousands of applications. The major advantage of fully connected networks is that they are “structure agnostic.” That is, no special assumptions need to be made about the input (for example, that the input consi




The major advantage of fully connected networks is that they are “structure agnostic.” That is, no special assumptions need to be made about the input (for example, that the input consists of images or videos).

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

4. Fully Connected Deep Networks - TensorFlow for Deep Learning [Book]
Fully Connected Deep Networks This chapter will introduce you to fully connected deep networks. Fully connected networks are the workhorses of deep learning, used for thousands of applications. <span>The major advantage of fully connected networks is that they are “structure agnostic.” That is, no special assumptions need to be made about the input (for example, that the input consists of images or videos). We will make use of this generality to use fully connected deep networks to address a problem in chemical modeling later in this chapter. We delve briefly into the mathematical theory u




fully connected architectures are “universal approximators” capable of learning any function. This concept provides an explanation of the generality of fully connected architectures, but comes with many caveats that we discuss at some depth.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

4. Fully Connected Deep Networks - TensorFlow for Deep Learning [Book]
to address a problem in chemical modeling later in this chapter. We delve briefly into the mathematical theory underpinning fully connected networks. In particular, we explore the concept that <span>fully connected architectures are “universal approximators” capable of learning any function. This concept provides an explanation of the generality of fully connected architectures, but comes with many caveats that we discuss at some depth. While being structure agnostic makes fully connected networks very broadly applicable, such networks do tend to have weaker performance than special-purpose networks tuned to the struct




The fully connected layers are able to very effectively learn non-linear combinations of input features.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

What is the role of fully connected layer in deep learning? - Stack Overflow
lot deep-learning share |improve this question asked Feb 18 '19 at 11:23 [emptylink] [imagelink] ErickYA ErickYA 2144 bronze badges add a comment | 1 Answer 1 active oldest votes 0 [emptylink] <span>The fully connected layers are able to very effectively learn non-linear combinations of input features. Let's take a convolutional neural network for example. The output from the convolutional layers represents high-level features in the data. While that output could be flattened and conn




The output from the convolutional layers represents high-level features in the data. While that output could be flattened and connected to the output layer, adding a fully-connected layer is a (usually) cheap way of learning non-linear combinations of these features.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

What is the role of fully connected layer in deep learning? - Stack Overflow
active oldest votes 0 [emptylink] The fully connected layers are able to very effectively learn non-linear combinations of input features. Let's take a convolutional neural network for example. <span>The output from the convolutional layers represents high-level features in the data. While that output could be flattened and connected to the output layer, adding a fully-connected layer is a (usually) cheap way of learning non-linear combinations of these features. Essentially the convolutional layers are providing a meaningful, low-dimensional, and somewhat invariant feature space, and the fully-connected layer is learning a (possibly non-linear)




But what is it we are doing if we use 2 FC? for example AlexNet has FC (4096) and another FC (4096) – ErickYA Feb 18 '19 at 15:01 It's about abstraction. The more complex (abstract) features you can learn, the more robust model you have. So adding fully connected layer on the fully connected layer allows you just that. The previous FC layer learns some complex features, and the second one learns to combine knowledge from the previous one into some more complex knowledge. It's the basic idea behind the deep learning itself

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

What is the role of fully connected layer in deep learning? - Stack Overflow
a (possibly non-linear) function in that space. share |improve this answer answered Feb 18 '19 at 12:13 [emptylink] [imagelink] Novak Novak 1,77711 gold badge88 silver badges1818 bronze badges <span>But what is it we are doing if we use 2 FC? for example AlexNet has FC (4096) and another FC (4096) – ErickYA Feb 18 '19 at 15:01 It's about abstraction. The more complex (abstract) features you can learn, the more robust model you have. So adding fully connected layer on the fully connected layer allows you just that. The previous FC layer learns some complex features, and the second one learns to combine knowledge from the previous one into some more complex knowledge. It's the basic idea behind the deep learning itself. The same thing happens with Conv layers and all the other layers, but it's easier to see that on conv layers analogy... – Novak Feb 18 '19 at 15:17 Is it right if i say the FC also all




different labeling functions can overlap (label the same data point) and even conflict (assign different labels to the same data point). This is expected, and we demonstrate how we deal with this later.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Unknown title
abel to it (in this case, HAM or SPAM) or abstain (don’t assign any label). Labeling functions can be noisy: they don’t have perfect accuracy and don’t have to label every data point. Moreover, <span>different labeling functions can overlap (label the same data point) and even conflict (assign different labels to the same data point). This is expected, and we demonstrate how we deal with this later. Because their only requirement is that they map a data point a label (or abstain), they can wrap a wide variety of forms of supervision. Examples include, but are not limited to: Keywor




#nlp

Constituency parsing aims to extract a constituency-based parse tree from a sentence that represents its syntactic structure according to a phrase structure grammar.

Example:

             Sentence (S)
                 |
   +-------------+------------+
   |                          |
 Noun (N)                Verb Phrase (VP)
   |                          |
 John                 +-------+--------+
                      |                |
                    Verb (V)         Noun (N)
                      |                |
                    sees              Bill

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on




The 42nd Amendment Act of 1976 added four new Directive Principles to the original list. They require the State: 1. To secure opportunities for healthy development of children (Article 39). 2. To promote equal justice and to provide free legal aid to the poor (Article 39 A). 3. To take steps to secure the participation of workers in the management of industries (Article 43 A). 4. To protect and improve the environment and to safeguard forests and wild life (Article 48 A).

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The 44th Amendment Act of 1978 added one more Directive Principle, which requires the State to minimise inequalities in income, status, facilities and opportunities (Article 38).

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Constitutional Advisor to the Constituent Assembly, recommended that the rights of an individual should be divided into two categories—justiciable and non-justiciable, which was accepted by the Drafting Committee. Consequently, the Fundamental Rights, which are justiciable in nature, are incorporated in Part III and the Directive Principles, which are non-justiciable in nature, are incorporated in Part IV of the Constitution. Though the Directive Principles are non-justiciable, the Constitution (Article 37) makes it clear that ‘these principles are fundamental in the governance of the country and it shall be the duty of the state to apply these

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




According to Sir Ivor Jennings, the Directives are based on the political philosophy of the 19th century England. He remarked: ‘The ghosts of Sydney Webb and Beatrice Webb stalk through the pages of the text.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Flashcard 5001630321932

Question
What do Directive principles of State Policy Intend to Do wrt part lll ?????
Answer
They are intended to fill in the vacuum in Part III by providing for social and economic rights


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

pdf

cannot see any pdfs







the Directive Principles have to conform to and run as subsidiary to the Fundamental Rights. But, it also held that the Fundamental Rights could be amended by the Parliament by enacting constitutional amendment acts.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




The courts are bound to declare a law violative of any of the Fundamental Rights as unconstitutional and invalid. 7. The courts cannot declare a law violative of any of the Directive Principles as unconstitutional and invalid.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




the present position is that the Fundamental Rights enjoy supremacy over the Directive Principles. Yet, this does not mean that the Directive Principles cannot be implemented. The Parliament can amend the Fundamental Rights for implementing the Directive Principles, so long as the amendment does not damage or destroy the basic structure of the Constitution

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Lok adalat is a statutory forum for conciliatory settlement of legal disputes. It has been given the status of a civil court. Its awards are enforceable, binding on the parties and final as no appeal lies before any court against them.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Japanese Constitution is, perhaps, the only democratic Constitution in world which contains a list of duties of citizens.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




the Swaran Singh Committee suggested the incorporation of eight Fundamental Duties in the Constitution, the 42nd Constitutional Amendment Act (1976) included ten Fundamental Duties. Interestingly, certain recommendations of the Committee were not accepted by the Congress Party and hence, not incorporated in the Constitution. These include: 1. The Parliament may provide for the imposition of such penalty or punishment as may be considered appropriate for any non-compliance with or refusal to observe any of the duties. 2. No law imposing such penalty or punishment shall be called in question in any court on the ground of infringement of any of Fundamental Rights or on the ground of repugnancy to any other provision of the Constitution. 3. Duty to pay taxes should also be a Fundamental Duty of the citizens.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Like the Directive Principles, the fundamental duties are also non- justiciable. The Constitution does not provide for their direct enforcement

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Boulimie

  • Il s'agit le plus souvent d'un adolescent plus âgé.
  • Il existe là aussi une prépondérance féminine

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles
La boulimie est définie par la survenue récurrente d'accès hyperphagiques (appelés aussi crises de boulimie ou de gloutonnerie) suivis de comportements compensateurs pour éviter une prise de poids (vomissements provoqués, jeûne, prise de médicaments, exercice physique excessif, oubli volontaire d'insuline chez le diabétique).

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Boulimie

Lorsque les crises se répètent de nombreuses fois par jour, voire la nuit, on parle d'état de mal boulimique

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Boulimie

On retrouve parfois des excoriations à la jonction métacarpophalangienne de la main dominante, signe de vomissements provoqués (signe de Russel), et/ou une parotidomégalie.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Boulimie

Les accès hyperphagiques et les comportements compensatoires inappropriés surviennent tous deux, en moyenne, au moins une fois par semaine pendant 3 mois

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Boulimie

  • La complication métabolique la plus fréquente est une hypokaliémie.
  • En cas de chronicité, la survenue de problèmes dentaires (anomalies de l'émail et des gencives) est fréquente.
  • Une œsophagite (Mallory-Weiss) est à rechercher et à traiter par IPP en cas de douleurs rétrosternales ou de vomissements sanglants

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Boulimie :

  • Un traitement par inhibiteur de la recapture de la sérotonine (antidépresseur) peut être indiqué.
  • On veillera à informer le patient de prendre ses traitements médicamenteux à distance des vomissements (notamment le potassium ou la pilule le cas échéant)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




They are enforceable by law. Hence, the Parliament can provide for the imposition of appropriate penalty or punishment for failure to fulfil any of them. H R Gokhale, the then Law Minister, gave the following reason for incorporating the fundamental duties in the Constitution after twenty-six years of its inauguration: ‘In post-independent India, particularly on the eve of emergency in June 1975, a section of the people showed no anxiety to fulfil their fundamental obligations of respecting the established legal order ..... the provisions of chapter on fundamental duties would have a sobering effect on these restless spirits who have had a host of anti-national subversive and unconstitutional agitations in the past’.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#249 #69 #Alimentaires #Amaigrissement #Conduites #Cours #Facultaires #Médecine #Pédiatrie #Troubles

Hyperphagie boulimique

La principale différence avec la boulimie est qu'il n'existe pas de comportement visant à compenser la prise de poids

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Article 368 in Part XX of the Constitution deals with the powers of Parliament to amend the Constitution and its procedure. It states that the Parliament may, in exercise of its constituent power, amend by way of addition, variation or repeal any provision of the Constitution in accordance with the procedure laid down for the purpose. However, the Parliament cannot amend those provisions which form the ‘basic structure’ of the Constitution.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #unfinished
Synthetic Training Data 1. Use seq2seq model to generate positive questions from context+answer. 2. Heuristically transform positive questions into negatives (i.e., “no answer”/impossible). ● Result: +3.0 F1/EM score, new state-of-the-art. Synthetic Training Data 1. Pre-train seq2seq model on Wikipedia. ○ Encoder trained with BERT, Decoder trained to decode next sentence. 2. Fine-tune model on SQuAD Context+Answer → Question ○ Ceratosaurus was a theropod dinosaur in the Late Jurassic, around 150 million years ago. -> When did the Ceratosaurus live ? 3. Train model to predict answer spans without questions. ○ Ceratosaurus was a theropod dinosaur in the Late Jurassic, around 150 million years ago. -> {150 million years ago, 150 million, theropod dinsoaur, Late Jurassic, in the Late Jurassic} Synthetic Training Data 4. Generate answer spans from a lot of Wikipedia paragraphs using model from (3) 5. Use output of (4) as input to seq2seq model from (2) to generate synthetic questions: ○ Roxy Ann Peak is a 3,576-foot-tall mountain in the Western Cascade Range in the U.S. state of Oregon. → What state is Roxy Ann Peak in? 6. Filter with baseline SQuAD 2.0 system to throw out bad questions. ○ Roxy Ann Peak is a 3,576-foot-tall mountain in the Western Cascade Range in the U.S. state of Oregon. → What state is Roxy Ann Peak in? ( Good) ○ Roxy Ann Peak is a 3,576-foot-tall mountain in the Western Cascade Range in the U.S. state of Oregon. → Where is Oregon? ( Bad) 7. Heuristically generate “strong negatives”: a. Positive questions from other paragraphs of same document. What state is Roxy Ann Peak in? → When was Roxy Ann Peak first summited? b. Replace span of text with other span of same type (based on POS tags). Replacement is usually from paragraph. What state is Roxy Ann Peak in? → What state is Oregon in? What state is Roxy Ann Peak in? → What mountain is Roxy Ann Peak in? 8. Optionally: Two-pass training, where no-answer is modeled as regression second pass (~+0.5 F1)

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Devlin-presentation-BERT
aseline - Translate Train 73.7 67.0 68.8 XNLI Baseline - Translate Test 73.7 68.4 70.7 BERT - Translate Train 81.9 76.6 77.8 BERT - Translate Test 81.9 70.1 74.9 BERT - Zero Shot 81.9 63.8 74.3 <span>Synthetic Training Data 1. Use seq2seq model to generate positive questions from context+answer. 2. Heuristically transform positive questions into negatives (i.e., “no answer”/impossible). ● Result: +3.0 F1/EM score, new state-of-the-art. Synthetic Training Data 1. Pre-train seq2seq model on Wikipedia. ○ Encoder trained with BERT, Decoder trained to decode next sentence. 2. Fine-tune model on SQuAD Context+Answer → Question ○ Ceratosaurus was a theropod dinosaur in the Late Jurassic, around 150 million years ago. -> When did the Ceratosaurus live ? 3. Train model to predict answer spans without questions. ○ Ceratosaurus was a theropod dinosaur in the Late Jurassic, around 150 million years ago. -> {150 million years ago, 150 million, theropod dinsoaur, Late Jurassic, in the Late Jurassic} Synthetic Training Data 4. Generate answer spans from a lot of Wikipedia paragraphs using model from (3) 5. Use output of (4) as input to seq2seq model from (2) to generate synthetic questions: ○ Roxy Ann Peak is a 3,576-foot-tall mountain in the Western Cascade Range in the U.S. state of Oregon. → What state is Roxy Ann Peak in? 6. Filter with baseline SQuAD 2.0 system to throw out bad questions. ○ Roxy Ann Peak is a 3,576-foot-tall mountain in the Western Cascade Range in the U.S. state of Oregon. → What state is Roxy Ann Peak in? ( Good) ○ Roxy Ann Peak is a 3,576-foot-tall mountain in the Western Cascade Range in the U.S. state of Oregon. → Where is Oregon? ( Bad) 7. Heuristically generate “strong negatives”: a. Positive questions from other paragraphs of same document. What state is Roxy Ann Peak in? → When was Roxy Ann Peak first summited? b. Replace span of text with other span of same type (based on POS tags). Replacement is usually from paragraph. What state is Roxy Ann Peak in? → What state is Oregon in? What state is Roxy Ann Peak in? → What mountain is Roxy Ann Peak in? 8. Optionally: Two-pass training, where no-answer is modeled as regression second pass (~+0.5 F1) Common Questions ● Is deep bidirectionality really necessary? What about ELMo-style shallow bidirectionality on bigger model? ● Advantage: Slightly faster training time ● Disadvantages:




Constitution can be amended in three ways: (a) Amendment by simple majority of the Parliament, (b) Amendment by special majority of the Parliament, and (c) Amendment by special majority of the Parliament and the ratification of

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




provisions in the Constitution can be amended by a simple majority of the two Houses of Parliament outside the scope of Article 368. These provisions include: 1. Admission or establishment of new states. 2. Formation of new states and alteration of areas, boundaries or names of existing states. 3. Abolition or creation of legislative councils in states. 4. Second Schedule—emoluments, allowances, privileges and so on of the president, the governors, the Speakers, judges, etc. 5. Quorum in Parliament. 6. Salaries and allowances of the members of Parliament. 7. Rules of procedure in Parliament. 8. Privileges of the Parliament, its members and its committees. 9. Use of English language in Parliament. 10. Number of puisne judges in the Supreme Court. 11. Conferment of more jurisdiction on the Supreme Court. 12. Use of official language. 13. Citizenship—acquisition and termination. 14. Elections to Parliament and state legislatures. 15. Delimitation of constituencies. 16. Union territories. 17. Fifth Schedule—administration of scheduled areas and scheduled tribes. 18. Sixth Schedule—administration of tribal areas.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




(i) Fundamental Rights;(ii) Directive Principles of State Policy; and (iii) All other provisions which are not covered by the first and third categories. By Special Majority of Parliament

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Those provisions of the Constitution which are related to the federal structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all the remaining states take no action on the bill, it does not matter; the moment half of the states give their consent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself).

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Flashcard 5001832434956

Question
Those provisions of the Constitution which are related to the [...] structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all the remaining states take no action on the bill, it does not matter; the moment half of the states give their consent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself).
Answer
federal


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Those provisions of the Constitution which are related to the federal structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all th

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001834007820

Question
Those provisions of the Constitution which are related to the federal structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all the remaining states take no action on the bill, it does not matter; the moment half of the states give their consent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the [...] and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself).
Answer
President


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
onsent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the <span>President and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001835580684

Question
Those provisions of the Constitution which are related to the federal structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all the remaining states take no action on the bill, it does not matter; the moment half of the states give their consent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the [...] power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself).
Answer
executive


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
s no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the <span>executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule.

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001837153548

Question
Those provisions of the Constitution which are related to the federal structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all the remaining states take no action on the bill, it does not matter; the moment half of the states give their consent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of [...] powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself).
Answer
legislative


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
sions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of <span>legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001838726412

Question
Those provisions of the Constitution which are related to the federal structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all the remaining states take no action on the bill, it does not matter; the moment half of the states give their consent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the [...] Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself).
Answer
Seventh


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the <span>Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself). <span>

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001840299276

Question
Those provisions of the Constitution which are related to the federal structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all the remaining states take no action on the bill, it does not matter; the moment half of the states give their consent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of [...] in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself).
Answer
states


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of <span>states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article 368 itself). <span>

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 5001841872140

Question
Those provisions of the Constitution which are related to the federal structure of the polity can be amended by a special majority of the Parliament and also with the consent of half of the state legislatures by a simple majority. If one or some or all the remaining states take no action on the bill, it does not matter; the moment half of the states give their consent, the formality is completed. There is no time limit within which the states should give their consent to the bill. The following provisions can be amended in this way: 1. Election of the President and its manner. 2. Extent of the executive power of the Union and the states. 3. Supreme Court and high courts. 4. Distribution of legislative powers between the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article [...] itself).
Answer
368


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill

Parent (intermediate) annotation

Open it
een the Union and the states. 5. Any of the lists in the Seventh Schedule. 6. Representation of states in Parliament. 7. Power of Parliament to amend the Constitution and its procedure (Article <span>368 itself). <span>

Original toplevel document (pdf)

cannot see any pdfs







The Constitution does not prescribe the time frame within which the state legislatures should ratify or reject an amendment submitted to them. Also, it is silent on the issue whether the states can withdraw their approval after according the same. 5. There is no provision for holding a joint sitting of both the Houses of Parliament if there is a deadlock over the passage of a constitutional amendment bill. On the other hand, a provision for a joint sitting is made in the case of an ordinary bill.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Parliament is empowered to abridge or take away any of the Fundamental Rights. At the same time, it laid down a new doctrine of the ‘basic structure’ (or ‘basic features’) of the Constitution. It ruled that the constituent power of Parliament under Article 368 does not enable it to alter the ‘basic structure’ of the Constitution. This means that the Parliament cannot abridge or take away a Fundamental Right that forms a part of the ‘basic structure’ of the Constitution

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Article 5001850260748

Nielsen-2019-Why_deep_hard_to_train-neuralnetworksanddeeplearning,com
#has-images #machine-learning #theory #unfinished

CHAPTER 5 Why are deep neural networks hard to train? Neural Networks and Deep Learning What this book is about On the exercises and problems Using neural nets to recognize handwritten digits How the backpropagation algorithm works Improving the way neural networks learn A visual proof that neural nets can compute any function Why are deep neural networks hard to train? Deep learning Appendix: Is there a simple algorithm for intelligence? Acknowledgements Frequently Asked Questions If you benefit from the book, please make a small donation. I suggest $5, but you can choose the amount. Alternately, you can make a donation by sending me Bitcoin, at address 1Kd6tXH5SDAmiFb49J9hknG5pqj7KStSAx Sponsors Deep Learning Workstations, Servers, and Laptops Thanks to all the supporters who made the book possible, with especial thanks to Pavel Dudrenov. Thanks also to all the contributors to the Bugfinder Hall



Generally when talking about SVM or tree boostings or most DL models, the model is performing classification. The model being learned is approximating a function that maps n-volumes of a high dimensional space to particular classes. For tabular data those n-volumes are likely to be very regular in nature, bounded on the edges by hyperplanes or manifolds that are approximately hyperplanes. Moving along any particular axis cleanly takes you from one classification to another classification, and small shifts of a few values will not change the classification almost everywhere in the space. This is why it makes sense to put the data in a table in the first place. Since SVM, tree boosting, etc classify a space by slicing it with hyperplanes, they are very suited to the problem of classifying a space where the classifications are delineated by hyperplanes or other manifolds of low curvature.

On the contrary, many classification problems require approximating a function to classify n-volumes whose boundaries have high curvature/concavity/eccentricity/etc. For those sorts of problems segmenting the space with hyperplanes will always perform poorly, which is where the value of the nonlinearity of DL solutions come into play.

level 2 zawerf 74 points · 1 year ago · edited 1 year ago

This lecture calls it the manifold hypothesis.

The tl;dw; is that the layers will sequentially straighten the data manifold until it's more or less linearly separable in the penultimate layer, and can be classified by a final logistic regression(which can only do linear decision boundaries). So rather than carving out volumes with hyperplanes like with traditional techniques, DL massages the manifold into a crazy shape that can be cut with one plane.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

[D] Why is Deep Learning so bad for tabular data? : MachineLearning
by best level 1 Dodobirdlord 87 points · 1 year ago I've found thinking about the high-dimensional space that the data represents to be helpful to get an intuition for these sorts of questions. <span>Generally when talking about SVM or tree boostings or most DL models, the model is performing classification. The model being learned is approximating a function that maps n-volumes of a high dimensional space to particular classes. For tabular data those n-volumes are likely to be very regular in nature, bounded on the edges by hyperplanes or manifolds that are approximately hyperplanes. Moving along any particular axis cleanly takes you from one classification to another classification, and small shifts of a few values will not change the classification almost everywhere in the space. This is why it makes sense to put the data in a table in the first place. Since SVM, tree boosting, etc classify a space by slicing it with hyperplanes, they are very suited to the problem of classifying a space where the classifications are delineated by hyperplanes or other manifolds of low curvature. On the contrary, many classification problems require approximating a function to classify n-volumes whose boundaries have high curvature/concavity/eccentricity/etc. For those sorts of problems segmenting the space with hyperplanes will always perform poorly, which is where the value of the nonlinearity of DL solutions come into play. level 2 zawerf 74 points · 1 year ago · edited 1 year ago This lecture calls it the manifold hypothesis. The tl;dw; is that the layers will sequentially straighten the data manifold until it's more or less linearly separable in the penultimate layer, and can be classified by a final logistic regression(which can only do linear decision boundaries). So rather than carving out volumes with hyperplanes like with traditional techniques, DL massages the manifold into a crazy shape that can be cut with one plane. Continue this thread level 2 maltin Original Poster9 points · 1 year ago I like this vision, and I share it with you. This explains why traditional methods are good for tabular data, an




At Google about 70-80% of "deep learning" is done on tabular data. The stuff people typically think of as deep learning (image, speech, translation, etc.) only makes up a small percentage.

I suspect the reason for your question is that most tabular datasets are tiny (you did say Excel) and neural networks can easily overfit, so you'd have to fiddle with regularization and other hyperparameters to get comparable results to xgboost. If you have a large dataset with billions (even trillions) of rows, a simple network with embeddings and a few hidden layers can work very well without much tuning.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

[D] Why is Deep Learning so bad for tabular data? : MachineLearning
a good model, so I do not understand which part of DL makes it so troublesome when attacking tabular data. Continue this thread level 1 rongou 48 points · 1 year ago This is actually not true. <span>At Google about 70-80% of "deep learning" is done on tabular data. The stuff people typically think of as deep learning (image, speech, translation, etc.) only makes up a small percentage. I suspect the reason for your question is that most tabular datasets are tiny (you did say Excel) and neural networks can easily overfit, so you'd have to fiddle with regularization and other hyperparameters to get comparable results to xgboost. If you have a large dataset with billions (even trillions) of rows, a simple network with embeddings and a few hidden layers can work very well without much tuning. level 2 strojax 16 points · 1 year ago · edited 1 year ago I am working over billions of data points for a private company over tabular data only. After years of experiments, we have fo




At Google about 70-80% of "deep learning" is done on tabular data.

This would be news to me. Source?

level 2 rongou 31 points · 1 year ago

When I worked there (I left about a year ago) this number was thrown around a lot. I think it was originally based on survey data, crawling the code base, or a combination of both. Internally they are called sparse data (and sparse models) because many features/columns have very high dimensions (think referral urls), but only a few values are ever present in each row, thus the sparsity (imagine if you one-hot encode each feature). But they're the same as what people typically consider as tabular data.

The reason these are not talked more outside is, one, the models are pretty simple (embeddings + fully-connected layers + logistic regression/softmax/regressor), so there is not much to write papers about; two, a lot of it is business critical, so best kept under wraps (ads prediction, search ranking).

If you look carefully, there are a few hints about this in the published information:

  • The TFX platform is mostly about dealing with tabular data (at least initially).

  • The Estimator API also started its life supporting tabular data (DNNClassifier).

  • TPUs have specialized hardware to accelerate embedding lookup, which is mostly used for categorical features in tabular data.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

[D] Why is Deep Learning so bad for tabular data? : MachineLearning
s share save hide report This thread is archived New comments cannot be posted and votes cannot be cast Sort by best View all commentsShow parent comments level 1 MohKohn 34 points · 1 year ago <span>At Google about 70-80% of "deep learning" is done on tabular data. This would be news to me. Source? level 2 rongou 31 points · 1 year ago When I worked there (I left about a year ago) this number was thrown around a lot. I think it was originally based on survey data, crawling the code base, or a combination of both. Internally they are called sparse data (and sparse models) because many features/columns have very high dimensions (think referral urls), but only a few values are ever present in each row, thus the sparsity (imagine if you one-hot encode each feature). But they're the same as what people typically consider as tabular data. The reason these are not talked more outside is, one, the models are pretty simple (embeddings + fully-connected layers + logistic regression/softmax/regressor), so there is not much to write papers about; two, a lot of it is business critical, so best kept under wraps (ads prediction, search ranking). If you look carefully, there are a few hints about this in the published information: The TFX platform is mostly about dealing with tabular data (at least initially). The Estimator API also started its life supporting tabular data (DNNClassifier). TPUs have specialized hardware to accelerate embedding lookup, which is mostly used for categorical features in tabular data. level 3 MohKohn 5 points · 1 year ago Thanks for the description, hopefully we're outside your NDA :P. The term deep learning has always bothered me as a catch-all for some pretty diffe




Escape Analysis allows a compiler to determine whether an object is accessible outside the allocating method or thread. This information is used to perform optimizations such as Scalar Replacement, Stack Allocation and Lock Elision, al- lowing modern dynamic compilers to remove some of the abstractions introduced by advanced programming models.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Effectiveness of Self Normalizing Neural Networks for Text Classification Authors: Avinash Madasu, Vijjini Anvesh Rao (Submitted on 3 May 2019)
Abstract: Self Normalizing Neural Networks(SNN) proposed on Feed Forward Neural Networks(FNN) outperform regular FNN architectures in various machine learning tasks. Particularly in the domain of Computer Vision, the activation function Scaled Exponential Linear Units (SELU) proposed for SNNs, perform better than other non linear activations such as ReLU. The goal of SNN is to produce a normalized output for a normalized input. Established neural network architectures like feed forward networks and Convolutional Neural Networks(CNN) lack the intrinsic nature of normalizing outputs. Hence, requiring additional layers such as Batch Normalization. Despite the success of SNNs, their characteristic features on other network architectures like CNN haven't been explored, especially in the domain of Natural Language Processing. In this paper we aim to show the effectiveness of proposed, Self Normalizing Convolutional Neural Networks(SCNN) on text classification. We analyze their performance with the standard CNN architecture used on several text classification datasets. Our experiments demonstrate that SCNN achieves comparable results to standard CNN model with significantly fewer parameters. Furthermore it also outperforms CNN with equal number of parameters.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

[1905.01338] Effectiveness of Self Normalizing Neural Networks for Text Classification
ence ACM classification MSC classification Report number arXiv identifier DOI ORCID arXiv author ID Help pages Full text Search Download PDF Computer Science > Computation and Language Title:<span>Effectiveness of Self Normalizing Neural Networks for Text Classification Authors:Avinash Madasu, Vijjini Anvesh Rao (Submitted on 3 May 2019) Abstract: Self Normalizing Neural Networks(SNN) proposed on Feed Forward Neural Networks(FNN) outperform regular FNN architectures in various machine learning tasks. Particularly in the domain of Computer Vision, the activation function Scaled Exponential Linear Units (SELU) proposed for SNNs, perform better than other non linear activations such as ReLU. The goal of SNN is to produce a normalized output for a normalized input. Established neural network architectures like feed forward networks and Convolutional Neural Networks(CNN) lack the intrinsic nature of normalizing outputs. Hence, requiring additional layers such as Batch Normalization. Despite the success of SNNs, their characteristic features on other network architectures like CNN haven't been explored, especially in the domain of Natural Language Processing. In this paper we aim to show the effectiveness of proposed, Self Normalizing Convolutional Neural Networks(SCNN) on text classification. We analyze their performance with the standard CNN architecture used on several text classification datasets. Our experiments demonstrate that SCNN achieves comparable results to standard CNN model with significantly fewer parameters. Furthermore it also outperforms CNN with equal number of parameters. Comments: Accepted Long Paper at 20th International Conference on Computational Linguistics and Intelligent Text Processing, April 2019, La Rochelle, France Subjects: Computation and La




#machine-learning #theory #unfinished
On 75 small datasets with less than 1000 data points, random forests and SVMs outperform SNNs and other FNNs. On 46 larger datasets with at least 1000 data points, SNNs show the highest performance followed by SVMs and random forests (see right panel of Table 1, for complete results see Supplementary Tables S9 and S10). Overall, SNNs have outperformed state of the art machine learning methods on UCI datasets with more than 1,000 data points

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#machine-learning #theory #unfinished
To summarize, self-normalizing networks work well with the following configuration: • SELU activation with parameters λ ≈ 1.0507 and α ≈ 1.6733, • inputs normalized to zero mean and unit variance, • network weights initialized with variance 1/n, and • regularization with “alpha-dropout”.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#machine-learning #theory #unfinished
We have introduced self-normalizing neural networks for which we have proved that neuron ac- tivations are pushed towards zero mean and unit variance when propagated through the network. Additionally, for activations not close to unit variance, we have proved an upper and lower bound on the variance mapping. Consequently, SNNs do not face vanishing and exploding gradient prob- lems. Therefore, SNNs work well for architectures with many layers, allowed us to introduce a novel regularization scheme and learn very robustly.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks Authors: Nils Reimers, Iryna Gurevych (Submitted on 27 Aug 2019)
Abstract: BERT (Devlin et al., 2018) and RoBERTa (Liu et al., 2019) has set a new state-of-the-art performance on sentence-pair regression tasks like semantic textual similarity (STS). However, it requires that both sentences are fed into the network, which causes a massive computational overhead: Finding the most similar pair in a collection of 10,000 sentences requires about 50 million inference computations (~65 hours) with BERT. The construction of BERT makes it unsuitable for semantic similarity search as well as for unsupervised tasks like clustering.
In this publication, we present Sentence-BERT (SBERT), a modification of the pretrained BERT network that use siamese and triplet network structures to derive semantically meaningful sentence embeddings that can be compared using cosine-similarity. This reduces the effort for finding the most similar pair from 65 hours with BERT / RoBERTa to about 5 seconds with SBERT, while maintaining the accuracy from BERT.
We evaluate SBERT and SRoBERTa on common STS tasks and transfer learning tasks, where it outperforms other state-of-the-art sentence embeddings methods.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

[1908.10084] Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks
ence ACM classification MSC classification Report number arXiv identifier DOI ORCID arXiv author ID Help pages Full text Search Download PDF Computer Science > Computation and Language Title:<span>Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks Authors:Nils Reimers, Iryna Gurevych (Submitted on 27 Aug 2019) Abstract: BERT (Devlin et al., 2018) and RoBERTa (Liu et al., 2019) has set a new state-of-the-art performance on sentence-pair regression tasks like semantic textual similarity (STS). However, it requires that both sentences are fed into the network, which causes a massive computational overhead: Finding the most similar pair in a collection of 10,000 sentences requires about 50 million inference computations (~65 hours) with BERT. The construction of BERT makes it unsuitable for semantic similarity search as well as for unsupervised tasks like clustering. In this publication, we present Sentence-BERT (SBERT), a modification of the pretrained BERT network that use siamese and triplet network structures to derive semantically meaningful sentence embeddings that can be compared using cosine-similarity. This reduces the effort for finding the most similar pair from 65 hours with BERT / RoBERTa to about 5 seconds with SBERT, while maintaining the accuracy from BERT. We evaluate SBERT and SRoBERTa on common STS tasks and transfer learning tasks, where it outperforms other state-of-the-art sentence embeddings methods. Comments: Published at EMNLP 2019 Subjects: Computation and Language (cs.CL) Cite as: arXiv:1908.10084 [cs.CL] (or arXiv:1908.10084v1 [cs.CL] for this version) Bibliographic data [Enabl




#bert #unfinished
informa- tion retrieval via semantic search

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #unfinished
BERT uses a cross-encoder: Two sentences are passed to the transformer network and the target value is predicted. However, this setup is unsuitable for various pair regression tasks due to too many possible combinations. Finding in a collection of n = 10 000 sentences the pair with the highest similarity requires with BERT n·(n−1)/2 = 49 995 000 inference computations. On a modern V100 GPU, this requires about 65 hours. Similar, finding which of the over 40 mil- lion existent questions of Quora is the most similar for a new question could be modeled as a pair-wise comparison with BERT, however, answering a sin- gle query would require over 50 hours

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#bert #unfinished
A common method to address clustering and se- mantic search is to map each sentence to a vec- tor space such that semantically similar sentences are close. Researchers have started to input indi- vidual sentences into BERT and to derive fixed- size sentence embeddings. The most commonly used approach is to average the BERT output layer (known as BERT embeddings) or by using the out- put of the first token (the [CLS] token). As we will show, this common practice yields rather bad sentence embeddings, often worse than averaging GloVe embeddings (Pennington et al., 2014). To alleviate this issue, we developed SBERT.

statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs