Edited, memorised or added to reading list

on 04-Feb-2020 (Tue)

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

Flashcard 4887003401484

Question
[...] is a classification of scientific research projects that seek fundamental understanding of scientific problems, while also having immediate use for society. The term was introduced by Donald E. Stokes.
Answer
Pasteur's quadrant

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
Pasteur's quadrant is a classification of scientific research projects that seek fundamental understanding of scientific problems, while also having immediate use for society. Louis Pasteur 's research is

Original toplevel document

Pasteur's quadrant - Wikipedia
Pasteur's quadrant - Wikipedia Pasteur's quadrant From Wikipedia, the free encyclopedia Jump to navigation Jump to search Pasteur's quadrant is a classification of scientific research projects that seek fundamental understanding of scientific problems, while also having immediate use for society. Louis Pasteur 's research is thought to exemplify this type of method, which bridges the gap between "basic " and "applied " research.[1] The term was introduced by Donald E. Stokes in his book, Pasteur's Quadrant.[2] Applied and basic research[edit ] As shown in the following table, scientific research can be classified by whether it advances human knowledge by seeking a fundamental understanding







Flashcard 4907364388108

Tags
#knowledge-base-construction #machine-learning #nlp #unfinished
Question
In recent years, there has been much interest in [...] as a powerful tool for defining rich probabilistic models [Defense Advanced Research Projects Agency, 2013, Daniel Roy et al., 2017]
Answer
probabilistic programming

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 recent years, there has been much interest in probabilistic programming as a powerful tool for defining rich probabilistic models [Defense Advanced Research Projects Agency, 2013, Daniel Roy et al., 2017]

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 4916568526092

Tags
#knowledge-base-construction #machine-learning #unfinished
Question
Fonduer obviates the need for feature engineering by [...]to obtain a representation needed to automate relation extraction from richly formatted data.
Answer
extending a bidirectional LSTM

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
Fonduer obviates the need for feature engineering by extending a bidirectional LSTM—the de facto deep-learning standard in natu- ral language processing [ 24 ]—to obtain a representation needed to automate relation extraction from richly formatted data.

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 4921143725324

Question
How to draw a Domain Class Diagram ?
Answer

1. Identify Responsibilities on Domain Class Diagrams.

2. Indicate Visibility Only On Design Models.

3. Indicate Language-Dependent Visibility With Property Strings.

4. Indicate Types Only On Design Models.

5. Indicate Types On Analysis Models Only When The Type is an Actual Requirement.

6. Design Class Diagrams Should Reflect Language Naming Conventions. 1. In Figure 1 you see that the design version of the Order class uses names that conform to common Java programming conventions such as placementDate and calculateTaxes() .

7. Model Association Classes On Analysis Diagrams. Figure 2 shows that association classes are depicted as class attached via a dashed line to an association - the association line, the class, and the dashed line are considered one symbol in the UML.

8. Do Not Name Associations That Have Association Classes.

9. Center The Dashed Line of an Association Class.


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill
UML 2 Class Diagram Guidelines
circumstance. This section describes style guidelines pertaining to different types of class diagrams. Figure 1. Analysis and design versions of a class. Figure 2. Modeling association classes. <span>Identify Responsibilities on Domain Class Diagrams. Indicate Visibility Only On Design Models. Indicate Language-Dependent Visibility With Property Strings. Indicate Types Only On Design Models. Indicate Types On Analysis Models Only When The Type is an Actual Requirement. Design Class Diagrams Should Reflect Language Naming Conventions. In Figure 1 you see that the design version of the Order class uses names that conform to common Java programming conventions such as placementDate and calculateTaxes(). Model Association Classes On Analysis Diagrams. Figure 2 shows that association classes are depicted as class attached via a dashed line to an association - the association line, the class, and the dashed line are considered one symbol in the UML. Do Not Name Associations That Have Association Classes. Center The Dashed Line of an Association Class. 2. Class Style Guidelines A class is effectively a template from which objects are created (instantiated).Although in the real world Doug, Wayne, John, and Bill are all student objects







Flashcard 4921146084620

Question
How to conform to styleguide for class diagram ?
Answer

1. Use Common Terminology for Names

2. Prefer Complete Singular Nouns for Class Names

3. Name Operations with a Strong Verb

4. Name Attributes With a Domain-Based Noun

5. Do Not Model Scaffolding Code. Scaffolding code refers to the attributes and operations required to implement basic functionality within your classes, such as the code required to implement relationships with other classes. Figure 3 depicts the difference between the OrderItem class without scaffolding code and with it.

6. Never Show Classes With Just Two Compartments

7. Label Uncommon Class Compartments

8. Include an Ellipsis ( . . . ) At The End of Incomplete Lists

9. List Static Operations/Attributes Before Instance Operations/Attributes

10. List Operations/Attributes in Decreasing Visibility

11. For Parameters That Are Objects, Only List Their Type

12. Develop Consistent Method Signatures

13. Avoid Stereotypes Implied By Language Naming Conventions

14. Indicate Exceptions In An Operation's Property String. Exceptions can be indicated with a UML property string, an example of which is shown in Figure 4.


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill
UML 2 Class Diagram Guidelines
ure 3. The OrderItem class with and without scaffolding code. Figure 4. Indicating the exceptions thrown by an operation. + findAllInstances(): Vector {exceptions=NetworkFailure, DatabaseError} <span>Use Common Terminology for Names Prefer Complete Singular Nouns for Class Names Name Operations with a Strong Verb Name Attributes With a Domain-Based Noun Do Not Model Scaffolding Code. Scaffolding code refers to the attributes and operations required to implement basic functionality within your classes, such as the code required to implement relationships with other classes. Figure 3 depicts the difference between the OrderItem class without scaffolding code and with it. Never Show Classes With Just Two Compartments Label Uncommon Class Compartments Include an Ellipsis ( . . . ) At The End of Incomplete Lists List Static Operations/Attributes Before Instance Operations/Attributes List Operations/Attributes in Decreasing Visibility For Parameters That Are Objects, Only List Their Type Develop Consistent Method Signatures Avoid Stereotypes Implied By Language Naming Conventions Indicate Exceptions In An Operation's Property String. Exceptions can be indicated with a UML property string, an example of which is shown in Figure 4. 3. Interfaces An interface is a collection of operation signature and/or attribute definitions that ideally defines a cohesive set of behaviors. Interfaces are implemented, "realized" i







Flashcard 4921147919628

Question
How to model an interface in a class diagram ?
Answer

1. Interface Definitions Must Reflect Implementation Language Constraints. In Figure 5 you see that a standard class box has been used to define the interface PersistentObject (note the use of the <<interface>> stereotype).

2. Name Interfaces According To Language Naming Conventions

3. Apply "Lollipop" Notation To Indicate That A Class Realizes an Interface

4. Define Interfaces Separately From Your Classes

5. Do Not Model the Operations and Attributes of an Interface in Your Classes. In Figure 5 you'll notice that the Shipment class does not include the attributes or operations defined by the two interfaces that it realizes.

6. Consider an Interface to Be a Contract


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill
UML 2 Class Diagram Guidelines
interface. Any given class or component may implement zero or more interfaces and one or more classes or components can implement the same interface. Figure 5. Interfaces on UML class diagrams. <span>Interface Definitions Must Reflect Implementation Language Constraints. In Figure 5 you see that a standard class box has been used to define the interface PersistentObject (note the use of the <<interface>> stereotype). Name Interfaces According To Language Naming Conventions Apply "Lollipop" Notation To Indicate That A Class Realizes an Interface Define Interfaces Separately From Your Classes Do Not Model the Operations and Attributes of an Interface in Your Classes. In Figure 5 you'll notice that the Shipment class does not include the attributes or operations defined by the two interfaces that it realizes. Consider an Interface to Be a Contract 4. Relationship Guidelines For ease of discussion the term relationships shall include all UML concepts such as associations, aggregation, composition, dependencies, inheritance, and re







Flashcard 4921149754636

Question
How to model relationships ?
Answer

1. Model Relationships Horizontally

2. Collaboration Indicates Need for a Relationship

3. Model a Dependency When The Relationship is Transitory

4. Depict Similar Relationships Involving A Common Class As A Tree. In Figure 6 you see that both Delivery and Order have a dependency on OIDGenerator . Note how the two dependencies are drawn in combination in "tree configuration", instead of as two separate lines, to reduce clutter in the diagram.

5. Always Indicate the Multiplicity

6. Avoid a Multiplicity of "*"

7. Replace Relationships By Indicating Attribute Types. In Figure 7 you see that Customer has a shippingAddress attribute of type Address - part of the scaffolding code to maintain the association between customer objects and address objects.

8. Do Not Model Implied Relationships

9. Do Not Model Every Single Dependency

10. Center Names on Associations

11. Write Concise Association Names In Active Voice

12. Indicate Directionality To Clarify An Association Name

13. Name Unidirectional Associations In The Same Direction

14. Word Association Names Left-To-Right

15. Indicate Role Names When Multiple Associations Between Two Classes Exist

16. Indicate Role Names on Recursive Associations

17. Make Associations Bi-Directional Only When Collaboration Occurs In Both Directions. The lives at association of Figure 9 is uni-directional.

18. Redraw Inherited Associations Only When Something Changes

19. Question Multiplicities Involving Minimums And Maximums


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill
UML 2 Class Diagram Guidelines
on a UML class diagram we'll consider it a relationship. Figure 6. Shipping an order. Figure 7. Modeling an order. Figure 8. Professors and seminars. Figure 9. Modeling people at a university. <span>Model Relationships Horizontally Collaboration Indicates Need for a Relationship Model a Dependency When The Relationship is Transitory Depict Similar Relationships Involving A Common Class As A Tree. In Figure 6 you see that both Delivery and Order have a dependency on OIDGenerator. Note how the two dependencies are drawn in combination in "tree configuration", instead of as two separate lines, to reduce clutter in the diagram. Always Indicate the Multiplicity Avoid a Multiplicity of "*" Replace Relationships By Indicating Attribute Types. In Figure 7 you see that Customer has a shippingAddress attribute of type Address - part of the scaffolding code to maintain the association between customer objects and address objects. Do Not Model Implied Relationships Do Not Model Every Single Dependency Center Names on Associations Write Concise Association Names In Active Voice Indicate Directionality To Clarify An Association Name Name Unidirectional Associations In The Same Direction Word Association Names Left-To-Right Indicate Role Names When Multiple Associations Between Two Classes Exist Indicate Role Names on Recursive Associations Make Associations Bi-Directional Only When Collaboration Occurs In Both Directions. The lives at association of Figure 9 is uni-directional. Redraw Inherited Associations Only When Something Changes Question Multiplicities Involving Minimums And Maximums 6. Inheritance Guidelines Inheritance models "is a" and "is like" relationships, enabling you to easily reuse existing data and code. When "A" inherits from "B" we say that "A" is the s







Flashcard 4921154211084

Tags
#Case #System #Use
Question
How to identify System Use Case ?
Answer

Identifying Use Cases

How do you go about identifying potential use cases? Constantine and Lockwood (1999) suggest one way to identify essential use cases, or simply to identify use cases, is to identify potential services by asking your stakeholders the following questions from the point of view of the actors:

  • What are users in this role trying to accomplish?
  • To fulfill this role, what do users need to be able to do?
  • What are the main tasks of users in this role?
  • What information do users in this role need to examine, create, or change?
  • What do users in this role need to be informed of by the system?
  • What do users in this role need to inform the system about?

statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill
System Use Cases: An Agile Introduction
ust be handled. The use case text references several alternate courses, think of them simply as the use case way of doing if/then logic, one of which is described at the bottom of the use case. <span>Identifying Use Cases How do you go about identifying potential use cases? Constantine and Lockwood (1999) suggest one way to identify essential use cases, or simply to identify use cases, is to identify potential services by asking your stakeholders the following questions from the point of view of the actors: What are users in this role trying to accomplish? To fulfill this role, what do users need to be able to do? What are the main tasks of users in this role? What information do users in this role need to examine, create, or change? What do users in this role need to be informed of by the system? What do users in this role need to inform the system about? For example, from the point-of-view of the Student actor, you may discover that students: Enroll in, attend, drop, fail, and pass seminars. Need a list of available seminars. Need to de







Flashcard 4921157618956

Question
List the AMDD lifecycle: Modeling activities throughout the lifecycle of a project.
Answer

1. Identitfy the high-level scope

2. Identify initial "requirement stack"

3. Identitfy an architectural vision

4. Modeling is part of iteration planning effort

5. Need to model enough to give good estimates

6. Need to plan the work for the iteration

7. Work through specific issues on a JIT manner

8. Stakeholder actively participate

9. Requirement evolve throughout project

10. Model just enough for now, you can always come back later

11. Develop working software via test-first appraoch

12. Details captured in the form of executable specification


statusnot learnedmeasured difficulty37% [default]last interval [days]               
repetition number in this series0memorised on               scheduled repetition               
scheduled repetition interval               last repetition or drill
Iteration Modeling: An Agile Core Practice
ng in the right direction, you do iteration modeling, and you do just in time (JIT) model storming to explore the detailed requirements and design throughout an iteration. [imagelink] Figure 5. <span>The AMDD lifecycle: Modeling activities throughout the lifecycle of a project. Recommended Reading [imagelink] This book, Choose Your WoW! A Disciplined Agile Delivery Handbook for Optimizing Your Way of Working , is an indispensable guide for agile coaches and pr







Flashcard 4921159978252

Question
Fraksioni i ejeksionit ventrikular të majtë nuk
Answer
[default - edit me]

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







Flashcard 4921161551116

Question
[default - edit me]
Answer
është i barabartë me fraksionin e ejeksionit ventrikular të djathtë në kardiopatinë ishemike.

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







Fraksioni i ejeksionit ventrikular të majtë nuk A) është indeksi më i mirë i cili i vetëm mund te vlerësoje performancen globale të ventrikulit të majtë. B) është i vetmi indeks, më i miri prognostik, i kardiopatisë ishemike. C) rritet me aktivitetin fizik në subjektet normal. D) është i barabartë me fraksionin e ejeksionit ventrikular të djathtë në kardiopatinë ishemike. E) duhet te vlerësohet në insufiçencën kardiake kongjestive për të dalluar disfunksionin sistolik nga ai diastolik.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Flashcard 4921165221132

Question
Valvulopatia e shpeshtë e shoqëruar me sindromën e Marfan është
Answer
prolapsi i valvules mitrale

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







Flashcard 4921167056140

Question
Nuk është efekt tipik kolateral i regjimit javore të metotreksatit me doze të ulët në trajtimin e artritit reumatoid
Answer
nefropatia

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







Tek pacienti uremik ka një alterim të metabolizmit kalço/fosfor që konsiston në A) një reduktimi të sintezës së kalçitriolit. B) një prodhim të reduktuar të kalçitoninës. C) një prodhim të reduktuar tëPTH ( ParaT-hormonit). D) një absorbim të reduktuar intestinal të vitaminës D. E) një hiperkalçemi.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Flashcard 4921170464012

Question
Tek pacienti uremik ka një alterim të metabolizmit kalço/fosfor që konsiston në
Answer
[default - edit me]

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
Tek pacienti uremik ka një alterim të metabolizmit kalço/fosfor që konsiston në A) një reduktimi të sintezës së kalçitriolit. B) një prodhim të reduktuar të kalçitoninës. C) një prodhim të reduktuar tëPTH ( ParaT-hormonit). D) një absorbim të reduktuar intestinal t

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 4921171512588

Question
[default - edit me]
Answer
një reduktimi të sintezës së kalçitriolit

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
Tek pacienti uremik ka një alterim të metabolizmit kalço/fosfor që konsiston në A) një reduktimi të sintezës së kalçitriolit. B) një prodhim të reduktuar të kalçitoninës. C) një prodhim të reduktuar tëPTH ( ParaT-hormonit). D) një absorbim të reduktuar intestinal të vitaminës D. E) një hiperkalçemi. </

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 4921172561164

Question
Tek pacienti uremik ka një alterim të metabolizmit kalço/fosfor që konsiston në A) [...]. B) një prodhim të reduktuar të kalçitoninës. C) një prodhim të reduktuar tëPTH ( ParaT-hormonit). D) një absorbim të reduktuar intestinal të vitaminës D. E) një hiperkalçemi.
Answer
një reduktimi të sintezës së kalçitriolit

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
Tek pacienti uremik ka një alterim të metabolizmit kalço/fosfor që konsiston në A) një reduktimi të sintezës së kalçitriolit. B) një prodhim të reduktuar të kalçitoninës. C) një prodhim të reduktuar tëPTH ( ParaT-hormonit). D) një absorbim të reduktuar intestinal të vitaminës D. E) një hiperkalçemi. </

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 4921176231180

Tags
#knowledge-base-construction #machine-learning #unfinished
Question
As of 2018, what is the de facto deep-learning standard in natu- ral language processing?
Answer
A bidirectional LSTM

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
Fonduer obviates the need for feature engineering by extending a bidirectional LSTM—the de facto deep-learning standard in natu- ral language processing [ 24 ]—to obtain a representation needed to automate relation extraction from richly formatted data.

Original toplevel document (pdf)

cannot see any pdfs







A workflow engine is a software application that manages business processes. Workflow engines typically make use of a database server.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Workflow engine - Wikipedia
Workflow engine - Wikipedia Workflow engine From Wikipedia, the free encyclopedia Jump to navigation Jump to search A workflow engine is a software application that manages business processes. It is a key component in workflow technology and typically makes use of a database server . A workflow engine manages and monitors the state of activities in a workflow , such as the processing and approval of a loan application form, and determines which new activity to trans




A workflow engine manages and monitors the state of activities in a workflow, such as the processing and approval of a loan application form, and determines which new activity to transition to according to defined processes (workflows).[1] The actions may be anything from saving an application form in a document management system to sending a reminder e-mail to users or escalating overdue items to management. A workflow engine facilitates the flow of information, tasks, and events.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

Workflow engine - Wikipedia
avigation Jump to search A workflow engine is a software application that manages business processes. It is a key component in workflow technology and typically makes use of a database server . <span>A workflow engine manages and monitors the state of activities in a workflow , such as the processing and approval of a loan application form, and determines which new activity to transition to according to defined processes (workflows).[1] The actions may be anything from saving an application form in a document management system to sending a reminder e-mail to users or escalating overdue items to management. A workflow engine facilitates the flow of information, tasks, and events. Workflow engines may also be referred to as Workflow Orchestration Engines.[2] Workflow engines mainly have three functions: Verification of the current process status: Check whether it




Sketching NLP-powered experiences, however, presents unique challenges. It can be hard, for example, to visualize abstract language interaction, or to ideate a broad range of technically feasible intelligent functionalities.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




From conversational agents at home to opinion mining on social media, intelligent language interactions are hitting the mainstream.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




HCI research, while o ering many valuable and creative NLP applications, typically does not report the ideation or prototyping process
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




we provide a rst-person account of our experience adding intelligent language functionality into a Word document editor in collaboration with a group of NLP researchers. We started by attempting to rapidly exper- iment with many tentative NLP design ideas and broadly explore how NLP might improve the authoring experience. We wanted to use storyboards, UI wireframes, paper proto- types, and other simple, tangible instruments to sketch out early design ideas and probe users’ reactions [ 7 , 14 ].
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Common sketch- ing tools and techniques deal with tangible interactions and are ine ective at abstracting the experience of language or a conversation. A number of technical aspects of language intelligence further complicate its UX design.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




A number of technical aspects of language intelligence further complicate its UX design. For example, data-driven interactions vary across users, adapt to di erent contexts, and evolve over time, and it can be di cult for designers to envision such divergent courses of interactions or to visualize using traditional wireframes and prototypes [2, 13, 20, 43].
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




explore how to sketch NLP- powered user experiences, and what sketching actually means in the context of intelligent language interactions.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




We took a Research through Design approach [ 45 ]
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




#nlp #unfinished #ux
We also describe a set of instruments that became e ective for our sketching: a new form of wireframes that illustrated abstract language-interaction design ideas and became an e ective boundary object; a set of NLP technical properties that are closely relevant to UX design; and a new prototyping method that enabled us to rapidly simulate various kinds of NLP errors. We discuss how these ndings reveal under- explored research questions and new insights in innovating the UX of NLP systems.
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
The experimental paradigm in machine learning is reaching its limits. This is challenging the speed of scientific progress in the area.
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
Abstraction$in$engineering Thinking#with# the#specif ications#instead# of#the# details. ! part# specifications#instead#of#part# detail s. ! assembly#specifications#instead#of#assembly#details. Abstract i ons#lea k! Sometimes#one#needs#to#look#more#closely. ! Example:# en gin e#co oli n g#assemb ly# can # have#co mpl i cated#h eat#i nertia. ! Example:#global#resource#allocation#(cost,#weight,#etc.) Ab straction #leaks#limits#the#complexity#of#what#we#can#b uild .
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

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 learning

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

pdf

cannot see any pdfs




Article 4922021383436

Bottou-2019-Learning_Representations_Causal-leon,bottou,org
#has-images #machine-learning #theory #unfinished

leon.bottou.org User Tools Site Tools Search Home Research Publications Talks Projects Vitae Learning Representation with Causal Invariance (ICLR 2019) This talk was first given as a ICLR 2019 keynote. Joint work with Martin Arjovsky (NYU), Ishaan Gulrajani (Google), and David Lopez-Paz (Facebook) Learning algorithms often capture spurious correlations present in the training data distribution instead of addressing the task of interest. Such spurious correlations occur because the data collection process is subject to uncontrolled confounding biases. Suppose however that we have access to multiple datasets exemplifying the same concept but whose distributions exhibit different biases. Can we learn something that is common across all these distributions, while ignoring the spurious ways in which they differ? This can be achieved by projecting the data into a representation space that satisfy a causal invariance criterion. This idea di



#machine-learning #software-engineering #unfinished
Compu ter$pr o gr am s The#closer#to#logic,#th e#more#complex#the#artifacts ! Au to mob il e:# ~#30,000#parts## ! Airp lane:# ~#3,000,000#parts ! Processor:# ~#3,000,000,000#transistors (note#:# counts#include#many#identical#pa rts#and#as semblies.) ! MS#Office:# ~#40,000,000#LOCS ! Facebook: ~#60,000,000#LOCS ! Debi an :# ~#400,000,000#LOCS (note:#different#metric:#lines#of#codes#ar e#ra rely#identical.)
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
Programming$versus$learning Programming#has#real#advantages Minsky#and#Papert,#Perceptrons,# 1968
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
Is#a#shape#made#of#a#single#connected#component? M&P#prove#that#a sma ll\order#perceptron#cannot#say, but#a#simple#algorithm#can. Su ch#an#algorithm#fulfils#a#clear#contract:#verifying#co nn ectedness
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
Why$is$learning$gaining$momentum? ! Since#“connectedness”# has#a#clea r# m athema tica l#specification, we#can#envision#provable#a lgorithms . ! Since#“mousiness”#and#“cheesiness”# do#not#have#such#a#speci fi cation, we#ca nnot#envision#provable#algorithms. We#must#rely#on \ heuri stic#speci fications \ or#learning#techniques
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
When#the#volume#of#data#increases \ defining#heuristic#specifications#becomes#harder. \ training#learning#systems#becomes#more#effective
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
Integr ating$mac hine$learning$ in $large$so f tware$system s Machin e#learning#systems#must*#work#with#digital#software because#digital#computers#are#everywhere. ! Use#trained#mo dels#as#so ftware#modules. ! Use#learning#algorithms#as#software#modules.
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
Many#web# sites#offer#lists#of#items (#search#res ults,#recommendations,#ads#…) and#select#them#b y#modeling#click#probabilities
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

Norbert# Wiener,#

Cybernetics,#1948

“Information#(signal)#feedback#loops#a re#everywhere.” “They#are#central#to# adaptation#and#learning…” !See#(Bottou et#al.,#JMLR#2013)#for#a#possible#treatment#of#causal#loops.

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
The#red#team#knows#the# bandit#literature !There#are#l ots#of#potenti al#recommendation s#to#select#from. !On e#can#on ly#mo del #the#click#prob abiliti es#of#the#recommen datio ns# that#have#been#shown#often#enough#in#each#context. !Therefore#th e#red#team# code#p erforms#ε\greedy#exploration,# showing#a#small#proportion#of#ra ndom#recommendations. !Although#exploration#has#a#cost,#such#the#system#cannot# discover#the#most#relevant#recommendations#wi thout#it.
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
The#cli ck#p ro bab il ity#mo del#easi l y#di sco vers#that#some# of#the# recommendations#generated#by#the#red#team#are#not#very#good. ! Therefore#these#recommend ati ons#wi ll #be#sh own #with#a# smal ler# fo nt.
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
“A2manager2of2the2two2teams2should2hav e2paid2more2attention.” “The2 software2archi tect2shoul d2have2paid2 more2attenti on.” Micro\management#does#not#wor k#beca us e#it#does # not#s cale…
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
! Use#trained#models#as#software#modules. " problematic#because#trained# models#offer#weak#contracts. ! Use#lear ning#algorith ms#as#software#modules. " problematic#because#the#output#of#the#learning# algorithm# depends#on#the#training#data#which#itself # depends# on#every#other#module. Working*around*these*difficulties*could*save*billions…
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
Essential$epistemology Exact# sciences Experimental sciences Engineering Dea l s#wit h Axioms#& Theorems Facts#& Theori es Artifacts Truth is Forever Temporary It#works. Examples Mathematics C.S. theory Physics Biology Lots…
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
xperimen tal#sciences ! Deal#with#facts and#theories. ! Facts#are#true#when#they#can#be#replicated.# Theories #a re#true#wh en#they#predict#the#facts.# New#facts#can#invalidate#theories# that#were#previously#considered#true# (K.# Popper) ! Ex a mp les o Physics,# Biology,… o Social#sciences, #Experimental#psychology,…
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
Engineering ! Deal s# wi th#artifacts. ! A#claim#i s#tru e#when#the#arti fact#works, (the#a r tifact#fulfils #a#contract.) ! Examp les o Mechanical#engineering,#… o Nuclear#enginee ring,#… o Computer#sci ence.
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
What#is#machin e#lear nin g? ! An #engineerin g#d iscip li ne? Yes:#applications#abound. ! An #exact#scien ce? Yes:#statistical/algorithmic#learning#theory. ! An #experimental#science? Yes:#papers#often#report#on#experiments.
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
What#is#machin e#lear nin g? ! An #engineerin g#d iscip li ne? Yes:#applications#abound. ! An #exact#scien ce? Yes:#statistical/algorithmic#learning#theory. ! An #experimental#science? Yes:#most#papers#include#experiments. This#is #all#good#…#as#long#as#we#don’t#mix#the#ge nres. ! “My2software2 works2really 2well,2therefore2my2theory2is2true.” ! “This2is2NPKcomplete,2therefore2 thi s2experiment2is2wrong.”
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




[unknown IMAGE 4922057297164] #has-images #machine-learning #software-engineering #unfinished

Progress#during#the#

las t#few#dec ades#

has#been#driven#by

a#single# experimental#

paradigm!

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
How#to#improve#the#performance#of#a#ML#application? ! Get# better# algor ithm?# + ! Get# better# features? +++ ! Get# better# da ta? ++++++ Dat a #coll ection#is#hard#work ! The#data#distribu tion #must#match#the#operationa l#conditions. ! Rely#on#ma nua l#data#curation to#avoid#dataset#bias.
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
Training/testing#on#biased#datasets#gives#unrealistic#results. !E.g.#:#Torralba and#Efros,#Unbiased2 look2at2dataset2bias,#CVPR#2011.
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
Too#much#data#for#manual#curation ! Big#data#exists#because#data#col lection#is#automated. ! Nobody#checks#that#the#data#distribution#matches# the#operational#c onditions#of#the#s ystem. " All#large#scale#d atasets#are#bi ased. Training/testing#on#a#big#dataset#can#give#unrealistic#results.
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
Statistical#machine#learn ing !Building #a#classifier#that#works#well#under#a#certain#distribution. Machine#learnin g#for#artificial#in telligence !Building#a# classifier#that#recognizes#a#“concept ”. !Con cepts#exi st#in d epend ently#of#the#pattern #d istri b u tio n. !Training# data#will# never#cover#the#whole# range#of#possible# inputs. In2fact,2a2system 2that2recognizes2 a2“concept”2fulfils2a2stronger2contrac t than2a2classifier2th at2works2well2under2a2certain2distribution.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




[unknown IMAGE 4922066996492] #has-images #machine-learning #software-engineering #unfinished

Comp uter#v ision#is#n ot#a#statistical#prob lem

Car# examp l es#i n#Imag eNet

Is#this#less#of#a#car

because#the#context#is#wro ng?

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
Training/testing#is#an#unusually#convenient#experimental#paradigm… " it#makes#experimentation#easier#than#in#other#sciences " it#has#played#a#role#in#the#fast#progress#of#ML. …but#may#not#fulfil#our#increased#ambition#(big#data,#AI.) " the#end#of#an#anomaly?
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
Challenges#of#a #new#kind !We#are#used#to#face#technical#challenges. !These#are#“process2challenges.”
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
Scientific#process !We#cannot#solely rely#on#training/testing#experiments. !We#can#understand#more#th ings#with#ad#hoc#exp eriments. Exa m ple#in#qu estion#an swering#:# the#“BABI”#tasks##(Bordes et#al. ,#2015) o questions#that#require#combining#one/two/ more#factoids. o questions#that#require#spatial/temporal#reason ing. o etc
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
A#constructive#propositi on ! Machine#learning#papers#rarely#show#the#limits#of#their#a pproa ch. \ For#fear#that#reviewers#will#use#them#against#their#work. ! Reviewers#should#instead#demand#that#authors#di scuss# these#limits. \ Additional#credits#if#they#illustrate#these#li mi ts#with#experiments. \ Thi s#woul d#make#the#papers# more#in formative… \ and#would#therefore#facilitate#scientific#progress.
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
How#to#better#resist#distribution#shifts? ! Op timize#measu re#of#coverage instead#of#average#accuracy. \ Selective#classification#(El#Yaniv ’s,#…) \ KWIK#(Li#&#Littman) \ Con fo rmal #learn i n g#(Vovk,#…) !Very#interesting# properties# when#the#data#is#Zipf distributed…
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
Digital#computers#:#“software” ! Learning#mach i nes#:# ◦ Trained#module#as#a#software#component? Trained#components#only#offer#“weak#con tracts”. ◦ Training#software? If#you #can#get#th e#t raining#d ata#an d #replicate#the#rig. Recent#example#:#AlexNet. ◦ Task#specific#trained#features Nearly#as#good.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on

pdf

cannot see any pdfs




Flashcard 4922079579404

Tags
#bert #knowledge-base-construction #nlp #unfinished
Question
Before BERT, the state-of-the-art methods for relation classification were primarily based on Convolutional Neural Networks or [...] .
Answer
Recurrent Neural Net- works

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 state-of-the-art methods for rela- tion classification are primarily based on Convolutional or Recurrent Neural Net- works.... In this paper, we pro- pose a model that both leverages the pre- trained BERT language model and incor- porates information from the target entities to tackle the relation classificatio

Original toplevel document (pdf)

cannot see any pdfs







Flashcard 4922081414412

Tags
#bert #knowledge-base-construction #nlp #unfinished
Question
Before BERT, the state-of-the-art methods for relation classification were primarily based on [...] or Recurrent Neural Net- works.
Answer
Convolutional Neural Networks

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 state-of-the-art methods for rela- tion classification are primarily based on Convolutional or Recurrent Neural Net- works.... In this paper, we pro- pose a model that both leverages the pre- trained BERT language model and incor- porates information from the target entities to tackle the relation classificatio

Original toplevel document (pdf)

cannot see any pdfs







#bert #knowledge-base-construction #nlp #unfinished
We achieve significant improvement over the state-of-the-art method on the SemEval- 2010 task 8 relational dataset.
statusnot read reprioritisations
last reprioritisation on reading queue position [%]
started reading on finished reading on


Parent (intermediate) annotation

Open it
this paper, we pro- pose a model that both leverages the pre- trained BERT language model and incor- porates information from the target entities to tackle the relation classification task. ... <span>We achieve significant improvement over the state-of-the-art method on the SemEval- 2010 task 8 relational dataset. <span>

Original toplevel document (pdf)

cannot see any pdfs




Flashcard 4922088230156

Tags
#knowledge-base-construction #machine-learning #nlp #unfinished
Question
In recent years, there has been much interest in probabilistic programming as [...] [Defense Advanced Research Projects Agency, 2013, Daniel Roy et al., 2017]
Answer
a powerful tool for defining rich probabilistic models

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 recent years, there has been much interest in probabilistic programming as a powerful tool for defining rich probabilistic models [Defense Advanced Research Projects Agency, 2013, Daniel Roy et al., 2017]

Original toplevel document (pdf)

cannot see any pdfs