Janusz Wojtusiak

The AQ21 Natural Induction Program for Pattern Discovery: Initial Version and its Novel Features

The AQ21 program aims to perform natural induction, a process of generating inductive hypotheses in human-oriented forms that are easy to interpret and understand. This is achieved by employing a highly expressive representation language, attributional calculus, whose statements resemble natural language descriptions. This paper focuses on the pattern discovery mode of AQ21, which produces attributional rules that capture strong regularities in the data, but may not be fully consistent or complete with regard to the training data. AQ21 integrates several novel features, such as optimizing patterns according to multiple criteria, learning attributional rules with exceptions, generating optimized sets of alternative hypotheses, and handling data with unknown, irrelevant and/or non-applicable meta-values

Traffic Prediction for Agent Route Planning

This paper describes a methodology and initial results of predicting traffic by autonomous agents within a vehicle route planning system. The traffic predictions are made using AQ21, a natural induction system that learns and applies attributional rules. The presented methodology is implemented and experimentally evaluated within a multiagent-based simulation system. Initial results obtained by simulation indicate advantage of agents using AQ21 predictions when compared to naive agents that make no predictions and agents that use only weather-related information.

The LEM3 implementation of learnable evolution model and its testing on complex function optimization problems

Learnable Evolution Model (LEM) is a form of non-Darwinian evolutionary computation that employs machine learning to guide evolutionary processes. Its main novelty are new type of operators for creating new individuals, specifically, hypothesis generation, which learns rules indicating subareas in the search space that likely contain the optimum, and hypothesis instantiation, which populates these subspaces with new individuals. This paper briefly describes the newest and most advanced implementation of learnable evolution, LEM3, its novel features, and results from its comparison with a conventional, Darwinian-type evolutionary computation program (EA), a cultural evolution algorithm (CA), and the estimation of distribution algorithm (EDA) on selected function optimization problems (with the number of variables varying up to 1000). In every experiment, LEM3 outperformed the compared programs in terms of the evolution length (the number of fitness evaluations needed to achieved a desired solution), sometimes more than by one order of magnitude.

Towards application of rule learning to the meta-analysis of clinical data: an example of the metabolic syndrome.

PURPOSE Systematic reviews and meta-analysis of published clinical datasets are important part of medical research. By combining results of multiple studies, meta-analysis is able to increase confidence in its conclusions, validate particular study results, and sometimes lead to new findings. Extensive theory has been built on how to aggregate results from multiple studies and arrive to the statistically valid conclusions. Surprisingly, very little has been done to adopt advanced machine learning methods to support meta-analysis. METHODS In this paper we describe a novel machine learning methodology that is capable of inducing accurate and easy to understand attributional rules from aggregated data. Thus, the methodology can be used to support traditional meta-analysis in systematic reviews. Most machine learning applications give primary attention to predictive accuracy of the learned knowledge, and lesser attention to its understandability. Here we employed attributional rules, the special form of rules that are relatively easy to interpret for medical experts who are not necessarily trained in statistics and meta-analysis. RESULTS The methodology has been implemented and initially tested on a set of publicly available clinical data describing patients with metabolic syndrome (MS). The objective of this application was to determine rules describing combinations of clinical parameters used for metabolic syndrome diagnosis, and to develop rules for predicting whether particular patients are likely to develop secondary complications of MS. The aggregated clinical data was retrieved from 20 separate hospital cohorts that included 12 groups of patients with present liver disease symptoms and 8 control groups of healthy subjects. The total of 152 attributes were used, most of which were measured, however, in different studies. Twenty most common attributes were selected for the rule learning process. By applying the developed rule learning methodology we arrived at several different possible rulesets that can be used to predict three considered complications of MS, namely nonalcoholic fatty liver disease (NAFLD), simple steatosis (SS), and nonalcoholic steatohepatitis (NASH).

Sequence of Functional Loss and Recovery in Nursing Homes

PURPOSE OF THE STUDY This study provides benchmarks for likelihood, number of days until, and sequence of functional decline and recovery. DESIGN AND METHODS We analyzed activities of daily living (ADLs) of 296,051 residents in Veteran Affairs nursing homes between January 1, 2000 and October 9, 2012. ADLs were extracted from standard minimum data set assessments. Because of significant overlap between short- and long-stay residents, we did not distinguish between these populations. Twenty-five combinations of ADL deficits described the experience of 84.3% of all residents. A network model described transitions among these 25 combinations. The network was used to calculate the shortest, longest, and maximum likelihood paths using backward induction methodology. Longitudinal data were used to derive a Bayesian network that preserved the sequence of occurrence of 9 ADL deficits. RESULTS The majority of residents (57%) followed 4 pathways in loss of function. The most likely sequence, in order of occurrence, was bathing, grooming, walking, dressing, toileting, bowel continence, urinary continence, transferring, and feeding. The other three paths occurred with reversals in the order of dressing/toileting and bowel/urinary continence. ADL impairments persisted without any change for an average of 164 days (SD = 62). Residents recovered partially or completely from a single impairment in 57% of cases over an average of 119 days (SD = 41). Recovery rates declined as residents developed more than 4 impairments. IMPLICATIONS Recovery of deficits among those studied followed a relatively predictable path, and although more than half recovered from a single functional deficit, recovery exceeded 100 days suggesting time to recover often occurs over many months.

Shared Homes as an Alternative to Nursing Home Care: Impact of VA’s Medical Foster Home Program on Hospitalization

PURPOSE OF THE STUDY This study compares hospitalization rates for common conditions in the Veteran Affairs (VA) Medical Foster Home (MFH) program to VA nursing homes, known as Community Living Centers (CLCs). DESIGN AND METHODS We used a nested, matched, case control design. We examined 817 MFH residents and matched each to 3 CLC residents selected from a pool of 325,031. CLC and MFH cases were matched on (a) baseline time period, (b) follow-up time period, (c) age, (d) gender, (e) race, (f) risk of mortality calculated from comorbidities, and (g) history of hospitalization for the selected condition during the baseline period. Odds ratio (OR) and related confidence interval (CI) were calculated to contrast MFH cases and matched CLC controls. RESULTS Compared with matched CLC cases, MFH residents were less likely to be hospitalized for adverse care events, (OR = 0.13, 95% CI = 0.03-0.53), anxiety disorders (OR = 0.52, 95% CI = 0.33-0.80), mood disorders (OR = 0.57, 95% CI = 0.42-0.79), skin infections (OR = 0.22, 95% CI = 0.10-0.51), pressure ulcers (OR = 0.22, 95% CI = 0.09-0.50) and bacterial infections other than tuberculosis or septicemia (OR = 0.54, 95% CI = 0.31-0.92). MFH cases and matched CLC controls did not differ in rates of urinary tract infections, pneumonia, septicemia, suicide/self-injury, falls, other injury besides falls, history of injury, delirium/dementia/cognitive impairments, or adverse drug events. Hospitalization rates were not higher for any conditions studied in the MFH cohort compared with the CLC cohort. IMPLICATIONS MFH participants had the same or lower rates of hospitalizations for conditions examined compared with CLC controls suggesting that noninstitutional care by a nonfamilial caregiver does not increase hospitalization rates for common medical conditions.

Extreme logistic regression

Kernel logistic regression (KLR) is a very powerful algorithm that has been shown to be very competitive with many state-of the art machine learning algorithms such as support vector machines (SVM). Unlike SVM, KLR can be easily extended to multi-class problems and produces class posterior probability estimates making it very useful for many real world applications. However, the training of KLR using gradient based methods or iterative re-weighted least squares can be unbearably slow for large datasets. Coupled with poor conditioning and parameter tuning, training KLR can quickly design matrix become infeasible for some real datasets. The goal of this paper is to present simple, fast, scalable, and efficient algorithms for learning KLR. First, based on a simple approximation of the logistic function, a least square algorithm for KLR is derived that avoids the iterative tuning of gradient based methods. Second, inspired by the extreme learning machine (ELM) theory, an explicit feature space is constructed through a generalized single hidden layer feedforward network and used for training iterative re-weighted least squares KLR (IRLS-KLR) and the newly proposed least squares KLR (LS-KLR). Finally, for large-scale and/or poorly conditioned problems, a robust and efficient preconditioned learning technique is proposed for learning the algorithms presented in the paper. Numerical results on a series of artificial and 12 real bench-mark datasets show first that LS-KLR compares favorable with SVM and traditional IRLS-KLR in terms of accuracy and learning speed. Second, the extension of ELM to KLR results in simple, scalable and very fast algorithms with comparable generalization performance to their original versions. Finally, the introduced preconditioned learning method can significantly increase the learning speed of IRLS-KLR.

Machine learning in agent-based stochastic simulation: Inferential theory and evaluation in transportation logistics

Multiagent-based simulation is an approach to realize stochastic simulation where both the behavior of the modeled multiagent system and dynamic aspects of its environment are implemented with autonomous agents. Such simulation provides an ideal environment for intelligent agents to learn to perform their tasks before being deployed in a real-world environment. The presented research investigates theoretical and practical aspects of learning by autonomous agents within stochastic agent-based simulation. The theoretical work is based on the Inferential Theory of Learning, which describes learning processes from the perspective of a learners goal as a search through knowledge space. The theory is extended for approximate and probabilistic learning to account for the situations encountered when learning in stochastic environments. Practical aspects are exemplified by two use cases in autonomous logistics: learning predictive models for environment conditions in the future, and learning in the context of evolutionary plan optimization.

Combining Rule Induction and Reinforcement Learning: An Agent-based Vehicle Routing

Reinforcement learning suffers from inefficiency when the number of potential solutions to be searched is large. This paper describes a method of improving reinforcement learning by applying rule induction in multi-agent systems. Knowledge captured by learned rules is used to reduce search space in reinforcement learning, allowing it to shorten learning time. The method is particularly suitable for agents operating in dynamically changing environments, in which fast response to changes is required. The method has been tested in transportation logistics domain in which agents represent vehicles being routed in a simple road network. Experimental results indicate that in this domain the method performs better than traditional Q-learning, as indicated by statistical comparison.

Computational intelligence virtual community: Framework and implementation issues

This paper discusses the framework for virtual collaborative environment for researchers, practitioners, users and learners in the areas of computational intelligence and machine learning (CIML) that is currently developed by our group. It also outlines main features of the community portal under construction that will support communication and sharing of computational resources. In particular, selected aspects of structure of the portal such as common formats of data, models, software, publications and software documentation are discussed.