on 22-Jun-2022 (Wed)

In the United States, the leading cause of ESRD is diabetes mellitus, currently accounting for almost 45% of newly diagnosed cases of ESRD.

Commonly accepted criteria for initiating patients on maintenance dialysis include the presence of uremic symptoms, the presence of hyperkalemia unresponsive to conservative measures, persistent extra- cellular volume expansion despite diuretic therapy, acidosis refractory to medical therapy, a bleeding diathesis, and a creatinine clearance or estimated glomerular filtration rate (GFR) <s)

In ESRD, treatment options include hemodialysis (in center or at home); peritoneal dialysis, as either continuous ambulatory peritoneal dialysis (CAPD) or continuous cyclic peritoneal dialysis (CCPD); or transplantation

Hemodialysis relies on the principles of solute diffusion across a semipermeable membrane. Movement of metabolic waste products takes place down a concentration gradient from the circulation into the dialysate

What is the backdoor path criterion?

This is a graphical rule that tells us whether we can identify the causal effect of interest if we know the causal DAG.

Synthetic data is divided into two types, based on whether it is generated from actual datasets or not

As you may have already noticed, the case-control design selects individuals based on their outcome. Women who did develop cancer are much more likely to be included in the study than women who did not develop cancer. Therefore, our causal graph will include a note for selection-- C-- an arrow from the outcome Y to C, and a box around C to indicate that the analysis is conditional on having been selected into the study, which means that we are only one arrow away from selection bias.

Synthetic Reality: Synthetic market data generation at scale using agent based modeling

Why simulations? Trading strategies are evaluated traditionally against historical financial market data. The ability to forecast is largely based on events that occurred in the past. Simulations can explore what might happen outside of historical bounds and provide a powerful mechanism to analyze trading performance against a wide variety of market conditions and unforeseen scenarios

Simulation models

This article outlines mechanisms to generate synthetic market prices. Agents that trade with varying behaviors are used to simulate alternative price paths of assets to create a variety of ‘what-if’ scenarios. The synthetic market prices are then compared to the real market prices using statistical techniques

Traditionally, two modeling methodologies, known as top-down and bottom-up, modeling approaches have been widely applied to model complex systems

In a bottom-up approach, the individual elements of the system are first specified in detail with a predefined rule of behaviors and agent interactions; these elements are then linked in different levels until a sound top-level structure is generated

Proper modeling of individual behaviors and the interactions among individuals are essential for the bottom-up approach, which leads to one of the most important topics in bottom-up modeling methodology—the agent-based modeling (ABM).

The ABM became popular two decades ago. Axelrod [5] contended that ABM is a third way of carrying out science in addition to classical deductive and inductive reasoning.

Developers of ABMs have consistently encountered the difficulty of deriving informed rules for behaviors at the scale of individual agents. This gives them little choice but to employ abstract proxy representation of agent behavior, which is not suitable for quantitative analysis. As demonstrated by Torrens et al. [9], the individual behavior in an agent-based model can be machine-learned from samples collected at the individual-agent level. In addition, modern ABM techniques can help in analysis through their ability to have adaptive agents in different changing environments [10]. The machine- learning-based inference model can thus provide an alterna- tive to coarse models of agents and can extend traditional agent-based transition schemes that hardcode agents’ behavior rules into a model—apriori[9]. Accordingly, integrating ABM and ML in decision-making can combine the inductive and deductive reasoning approaches, enabling us to describe the way that decisions can be made or improved [6

this article uses an “agent-based model” to refer to any model in which a certain number of agents (individuals) interact to comprise the higher macrolevel of a complex system

the review of ML techniques for analyzing ABMs was seldom present. To our best knowledge, there are only two review papers related to this topic. One paper was published in 2017 (by Pereda et al. [12]) and provided a brief introduction to applying classification and regression-based ML techniques in ABMs. The other paper was published in 2019 (by Prasanna et al. [13]) and mainly focused on the review of the integration of reinforcement learning (RL) in ABMs of energy markets. Consequently, a comprehensive review of applying ML techniques in ABMs in terms of analytical frameworks, application scenarios, and procedures of implementation is essential for scientific progress

The motivation of this article is to investigate how different ML techniques can be combined with the ABM for modeling complex, large-scale systems from the perspective of improving the model accuracy or robustness and rendering better decision-making strategies.

ABM is a bottom-up modeling approach in which every agent of the system, theoretically, can be simulated to any level of granularity. Each agent can have corresponding state variables that represent its internal states, and it can also have its unique representation of interaction with other agents and the associated environment. By defining rules and behaviors for individual agents and the environment in which they are present, ABM enables fine granularity of modeling. Further, one can aggregate these rules to study the general behavior of the system. ABM can help in understanding how the macrolevel emergence of a system can be generated from sim- ple microlevel rules of individual behaviors. Grimm et al. [16] and 27 other modelers in the field of ecology proposed a protocol for describing ABM in 2006, which includes three grouped blocks, namely an overview, design concepts, and details (ODD), with a total of seven elements, and the detailed explanation of ODD can be found in Grimm’s paper. This protocol has been widely recognized and used in ecology- related fields, which can help us better understand the concept of the ABM

Although learning ability has always been regarded as one of the main motivations of implementing ABMs, studies regarding applying ML algorithms in ABMs were seldom found before 2000. The first use of ML in ABM may be hard to determine. One candidate appears to be Holland and Miller’s 1991 paper [49], which discussed the application of artificial adaptive agents in economic theory based on the genetic algorithm (GA). Notable work was published by Rand [10] in 2006; in his paper, he first discussed the integration of ML, wherein ML was used for the prediction that was used later for the decision-making of an agent

Basically, ML techniques in ABM are generally triggered by two problem domains and aimed to create smarter agents/models and calibrate/validate complex models in a vast variable space

The key feature of ML is its ability to automatically learn and improve based on data and empirical information without being explicitly programmed, which is known as “self-learning.”

The types of ML algorithms differ depending on the approaches they use, the type of input and output data, and the type of problem to be solved. A common way to classify them is based on their purpose for learning: supervised learning, unsupervised learning, semisupervised learning, and RL

3) Semisupervised Learning: Semisupervised learning aims to learn a better prediction as opposed to labelled data alone. For a supervised learning algorithm, labels are presented for all the observations in the dataset (i.e., with completely labeled training data), whereas for an unsupervised learning algorithm, labels are not required for the observation of the dataset. Semisupervised learning falls in between supervised or unsupervised learning algorithms. It is an approach that combines a small amount of labeled data with a large amount of unlabeled data during training when the cost of labeling work may render large, fully labeled training sets infeasible, whereas the acquisition of unlabeled data is relatively inexpensive. Generally, typical categories of semisupervised learning algorithms include the generative model method, the low-density separation method, the graph-based method, and the heuristic method. Some of the popular algorithms for semisupervised learning are summarized in Table S1 in the Supplementary Material

Augustijn et al. [55] categorized the applications of ML in ABM according to three different stages in the modeling process: preprocessing, agent-behavior prediction, decision-making, and postprocessing of decision analysis. An ML algorithm can be trained to model in the pre-processing state as input for the subsequent ABM to replace the rule-based module; further, the ML algorithm is restored for agent behavior prediction to provide the agent actions in the decision-making stage, which can be done using a pre-trained ML algorithm or can include training via the RL. For the postprocessing, the data are used and mapped back to a trained ML algorithm to calibrate or validate the model after running the ABM [55]. Since the ML can be used in different stages of ABM, the functionalities of ML vary. ML mainly serves two functions: 1) to improve the agents’ situational awareness, which is related to prediction or pattern recognition and 2) to reinforce agents’ behaviors, which is related to behavioral interventions

We have a total of four scenarios to which ML can contribute. 1) Scenario 1: Microagent situational awareness learning. 2) Scenario 2: Microagent behavior interventions. 3) Scenario 3: Macrolevel emergence emulator. 4) Scenario 4: Macro ABMs decision-making, as shown in Fig. 4