Larry Bull

For real! XCS with continuous-valued inputs

Many real-world problems are not conveniently expressed using the ternary representation typically used by Learning Classifier Systems and for such problems an interval-based representation is preferable. We analyse two interval-based representations recently proposed for XCS, together with their associated operators and find evidence of considerable representational and operator bias. We propose a new interval-based representation that is more straightforward than the previous ones and analyse its bias. The representations presented and their analysis are also applicable to other Learning Classifier System architectures. We discuss limitations of the real multiplexer problem, a benchmark problem used for Learning Classifier Systems that have a continuous-valued representation, and propose a new test problem, the checkerboard problem, that matches many classes of real-world problem more closely than the real multiplexer. Representations and operators are compared using both the real multiplexer and checkerboard problems and we find that representational, operator and sampling bias all affect the performance of XCS in continuous-valued environments.

Applications of Learning Classifier Systems

Learning Classifier Systems: A Brief Introduction.- Section 1 - Data Mining.- Data Mining using Learning Classifier Systems.- NXCS Experts for Financial Time Series Forecasting.- Encouraging Compact Rulesets from XCS for Enhanced Data Mining.- Section 2 - Modelling and Optimization.- The Fighter Aircraft LCS: A Real-World, Machine Innovation Application.- Traffic Balance using Learning Classifier Systems in an Agent-based Simulation.- A Multi-Agent Model of the UK Market in Electricity Generation.- Exploring Organizational-Learning Oriented Classifier Systems in Real-World Problems.- Section 3 - Control.- Distributed Routing in Communication Networks using the Temporal Fuzzy Classifier System - a Study on Evolutionary Multi-Agent Control.- The Development of an Industrial Learning Classifier System for Data-Mining in a Steel Hop Strip Mill.- Application of Learning Classifier Systems to the On-Line Reconfiguration of Electric Power Distribution Networks.- Towards Distributed Adaptive Control for Road Traffic Junction Signals using Learning Classifier Systems.- Bibliography of Real-World Classifier Systems Applications.

Genetic Programming with a Genetic Algorithm for Feature Construction and Selection

The use of machine learning techniques to automatically analyse data for information is becoming increasingly widespread. In this paper we primarily examine the use of Genetic Programming and a Genetic Algorithm to pre-process data before it is classified using the C4.5 decision tree learning algorithm. Genetic Programming is used to construct new features from those available in the data, a potentially significant process for data mining since it gives consideration to hidden relationships between features. A Genetic Algorithm is used to determine which such features are the most predictive. Using ten well-known datasets we show that our approach, in comparison to C4.5 alone, provides marked improvement in a number of cases. We then examine its use with other well-known machine learning techniques.

Fuzzy-XCS: A Michigan Genetic Fuzzy System

The issue of finding fuzzy models with an interpretability as good as possible without decreasing the accuracy is one of the main research topics on genetic fuzzy systems. When they are used to perform online reinforcement learning by means of Michigan-style fuzzy rule systems, this issue becomes even more difficult. Indeed, rule generalization (description of state-action relationships with rules as compact as possible) has received a great attention in the nonfuzzy evolutionary learning field (e.g., XCS is the subject of extensive ongoing research). However, the same issue does not appear to have received a similar level of attention in the case of Michigan-style fuzzy rule systems. This may be due to the difficulty in extending the discrete-valued system operation to the continuous case. The intention of this contribution is to propose an approach to properly develop a fuzzy XCS system for single-step reinforcement problems.

ZCS redux

Learning classifier systems traditionally use genetic algorithms to facilitate rule discovery, where rule fitness is payoff based. Current research has shifted to the use of accuracy-based fitness. This paper re-examines the use of a particular payoff-based learning classifier systemZCS. By using simple difference equation models of ZCS, we show that this system is capable of optimal performance subject to appropriate parameter settings. This is demonstrated for both single- and multistep tasks. Optimal performance of ZCS in well-known, multistep maze tasks is then presented to support the findings from the models.

Evolution of Plastic Learning in Spiking Networks via Memristive Connections

This paper presents a spiking neuroevolutionary system which implements memristors as plastic connections, i.e., whose weights can vary during a trial. The evolutionary design process exploits parameter self-adaptation and variable topologies, allowing the number of neurons, connection weights, and interneural connectivity pattern to emerge. By comparing two phenomenological real-world memristor implementations with networks comprised of: 1) linear resistors, and 2) constant-valued connections, we demonstrate that this approach allows the evolution of networks of appropriate complexity to emerge whilst exploiting the memristive properties of the connections to reduce learning time. We extend this approach to allow for heterogeneous mixtures of memristors within the networks; our approach provides an in-depth analysis of network structure. Our networks are evaluated on simulated robotic navigation tasks; results demonstrate that memristive plasticity enables higher performance than constant-weighted connections in both static and dynamic reward scenarios, and that mixtures of memristive elements provide performance advantages when compared to homogeneous memristive networks.

On Model-Based Evolutionary Computation

Abstract Traditional evolutionary computing techniques use an explicit fitness function – mathematical or simulated – to derive a solution to a problem from a population of individuals, over a number of generations. In this paper an approach which allows such techniques to be used on problems in which evaluations are costly, which cannot be expressed formally, or which are difficult to simulate, is examined. A neural network is trained using example individuals with the explicit fitness and the resulting model of the fitness function is then used by the evolutionary algorithm to find a solution. It is shown that the approach is effective over a small range of function types in comparison to the traditional approach when limited training data is available. An iterative step is then added whereby after a number of generations the current best individual in a population is evaluated directly on the explicit fitness function. The individual and its “real” fitness are then added to the training data and the neural network is re-trained to improve its approximation of the fitness function. It is shown that in this way the performance of the model-based architecture is greatly improved on more rugged/complex landscapes without a large increase in the amount of training data required.

Learning Classifier System Ensembles With Rule-Sharing

This paper presents an investigation into exploiting the population-based nature of learning classifier systems (LCSs) for their use within highly parallel systems. In particular, the use of simple payoff and accuracy-based LCSs within the ensemble machine approach is examined. Results indicate that inclusion of a rule migration mechanism inspired by parallel genetic algorithms is an effective way to improve learning speed in comparison to equivalent single systems. Presentation of a mechanism which exploits the underlying niche-based generalization mechanism of accuracy-based systems is then shown to further improve their performance, particularly, as task complexity increases. This is not found to be the case for payoff-based systems. Finally, considerably better than linear speedup is demonstrated with the accuracy-based systems on a version of the well-known Boolean logic benchmark task used throughout.

Learning classifier systems: A brief introduction

Machine learning is synonymous with advanced computing and a growing body of work exists on the use of such techniques to solve real-world problems [e.g., Tsoukalas & Uhrig, 1997]. The complex and/or ill-understood nature of many problem domains, such as data mining or process control, has led to the need for technologies which can adapt to the task they face. Learning Classifier Systems (LCS) [Holland, 1976] are a machine learning technique which combines reinforcement learning, evolutionary computing and other heuristics to produce adaptive systems. The subject of this book is the use of LCS for real-world applications.

A Neural Learning Classifier System with Self-Adaptive Constructivism for Mobile Robot Control

For artificial entities to achieve true autonomy and display complex lifelike behavior, they will need to exploit appropriate adaptable learning algorithms. In this context adaptability implies flexibility guided by the environment at any given time and an open-ended ability to learn appropriate behaviors. This article examines the use of constructivism-inspired mechanisms within a neural learning classifier system architecture that exploits parameter self-adaptation as an approach to realize such behavior. The system uses a rule structure in which each rule is represented by an artificial neural network. It is shown that appropriate internal rule complexity emerges during learning at a rate controlled by the learner and that the structure indicates underlying features of the task. Results are presented in simulated mazes before moving to a mobile robot platform.