Evolutionary learning of temporal behaviour using discrete and fuzzy classifier systems
Proceedings of Tenth International Symposium on Intelligent Control, 1995
ABSTRACT
A new approach to genetics based machine learning in fuzzy controller design
Proceedings of 1994 9th IEEE International Symposium on Intelligent Control, 1994
ABSTRACT
Evolving Multi-Agent Systems
Sample Paper for I Workshop on Genetic Fuzzy Systems
Brian Carse Faculty of Computing, Engineering and Mathematical Sciences University of the West of... more Brian Carse Faculty of Computing, Engineering and Mathematical Sciences University of the West of England Bristol BS16 1QY, UK E-mail: [email protected] ... Jorge Casillas Department of Computer Science and Artificial Intelligence University of Granada Granada E-18071, ...
A Comparison Between Two Architectures for Searching and Learning in Maze Problems
ABSTRACT We present two architectures, each designed to search 2-Dimensional mazes in order to lo... more ABSTRACT We present two architectures, each designed to search 2-Dimensional mazes in order to locate a goal position, both of which perform on-line learning as the search proceeds. The first architecture is a form of Adaptive Heuristic Critic which uses a Genetic Algorithm to determine the Action Policy and a Radial Basis Function Neural Network to store the acquired knowledge of the Critic. The second is a stimulus-response Classifier System (CS) which uses a Genetic Algorithm, applied Michigan style, for rule generation and the Bucket Brigade algorithm for rule reinforcement. Experiments conducted using agents based upon each architectural model lead us to a comparison of performance, and some observations on the nature and relative levels of abstraction in the acquired knowledge.
Automatic generation of fuzzy sensorimotor rules for mobile robotics
Proceedings of IEEE 5th International Fuzzy Systems, 1996
Abstract We briefly compare world modelling (or map-building) and reactive behavior as techniques... more Abstract We briefly compare world modelling (or map-building) and reactive behavior as techniques for mobile robot controller design. We then describe an approach to automatic generation of behavioral rules for a mobile robot which attempts to combine the best features of these two techniques. It is composed of three parts. In map-building mode the robot forms a “fuzzy image” of a maze in terms of the location of obstacles and goal states. It then acts as a “teacher”, providing examples of trajectories from which may be extracted ...
A comparison between two architectures for searching and learning in maze problems
Lecture Notes in Computer Science, 1994
ABSTRACT We present two architectures, each designed to search 2-Dimensional mazes in order to lo... more ABSTRACT We present two architectures, each designed to search 2-Dimensional mazes in order to locate a goal position, both of which perform on-line learning as the search proceeds. The first architecture is a form of Adaptive Heuristic Critic which uses a Genetic Algorithm to determine the Action Policy and a Radial Basis Function Neural Network to store the acquired knowledge of the Critic. The second is a stimulus-response Classifier System (CS) which uses a Genetic Algorithm, applied Michigan style, for rule generation and the Bucket Brigade algorithm for rule reinforcement. Experiments conducted using agents based upon each architectural model lead us to a comparison of performance, and some observations on the nature and relative levels of abstraction in the acquired knowledge.
Evolving radial basis function neural networks using a genetic algorithm
Proceedings of 1995 IEEE International Conference on Evolutionary Computation, 1995
Abstract Most research to date using genetic algorithms to evolve neural networks has focused on ... more Abstract Most research to date using genetic algorithms to evolve neural networks has focused on the multi-layer perceptron. Alternative neural network approaches such as the radial basis function network, and their representations appear to have received relatively ...
Experiments on a Pittsburgh-style fuzzy classifier system for mobile robotics
Proceedings of the IEEE Internatinal Symposium on Intelligent Control, 2002
ABSTRACT We report on experiments designed to highlight the strengths and weaknesses of an autono... more ABSTRACT We report on experiments designed to highlight the strengths and weaknesses of an autonomous rule acquisition algorithm for the fuzzy controller of a simulated mobile robot. The algorithm is a Pittsburgh-style fuzzy classifier system. The fuzzy classifier system paradigm is an elegant and versatile combination of evolutionary and lifetime reinforcement learning based on an underlying fuzzy logic structure. It possesses a powerful potential to be a general-purpose linguistically interpretable problem-solver for continuous real-valued domains. We have tested the performance and robustness of many controllers using this approach, a sample of which, are presented here. We find that, although the robot controllers can often be quite robust to environmental changes after learning, they also sometimes display critical weaknesses in certain scenarios. We also compare performance of this algorithm with that of a hand-coded fuzzy controller. We find that it is more difficult than one would initially expect to derive a set of hand-coded rules that are as versatile as the best of those acquired by the fuzzy classifier system. It is worth noting that the task to be learned here is itself quite simple, there are far more complex problems within the mobile robot domain, thus providing a forward impetus to the work.
Adaptive Distributed Routing Using Evolutionary Fuzzy Control
First Results from Experiments in Fuzzy Classifier System Architectures for Mobile Robotics
We present first results from a comparison between a Fuzzy Classifier System operating at the lev... more We present first results from a comparison between a Fuzzy Classifier System operating at the level of whole rule-bases, and three variants of one that operates at the level of individual rules. The application domain is mobile robotics, and the problem is autonomous acquisition of an “investigative” obstacle avoidance competency. The Fuzzy Classifier Systems operate on the rules of fuzzy
Abstract We describe two architectures that autonomously acquire fuzzy control rules to provide ,... more Abstract We describe two architectures that autonomously acquire fuzzy control rules to provide ,reactive behavioural competencies ,in a simulated mobile robotics application. One ,architecture is a “Pittsburgh”-style Fuzzy ,Classifier System (Pitt1). The other architecture is a “Michigan”- style Fuzzy Classifier System ,(Mich1). We tested the architectures on their ability to acquire an“investigative” obstacle ,avoidance competency. We found,that Mich1 implemented amore,local
A fuzzy classifier system framework is proposed which employs a tree-based representation for fuz... more A fuzzy classifier system framework is proposed which employs a tree-based representation for fuzzy rule (classifier) antecedents and genetic programming for fuzzy rule discovery. Such a rule representation is employed because of the expressive power and generality it endows to individual rules. The framework proposes accuracy-based fitness for individual fuzzy classifiers and employs evolutionary competition between simultaneously matched classifiers. The evolutionary algorithm (GP) is therefore searching for compact fuzzy rule bases which are simultaneously general, accurate and co-adapted. Additional extensions to the proposed framework are suggested.
The ability of a rule-based system to represent generalisations is of great importance. Generalis... more The ability of a rule-based system to represent generalisations is of great importance. Generalised rules allow more compact rule bases, scalability to higher dimensional spaces, faster inference and better linguistic interpretability. The issue of rule generalisation, and the interplay between general and specific rules in the same evolving population, has received a great deal of attention in the discrete-valued classifier system research community. The same issue does not appear to have received a similar level of attention in the case of fuzzy classifier systems. While it is true that generalised rule representations have been incorporated in a number of reported fuzzy classifier systems, often as an aside to other significant issues, this important feature has not yet been concentrated on in detail. The intention of this contribution is to raise awareness of the issue of generalisation in the fuzzy classifier system so that the issue may be brought under closer scrutiny. Early experimental results using a test-bed specifically designed to test the ability of the fuzzy classifier system to learn an optimal collection of co-existing general and specific fuzzy classifiers are described and discussed.
Learning a robot controller using an adaptive hierarchical fuzzy rule-based system
Soft Computing, 2015
ABSTRACT The majority of machine learning techniques applied to learning a robot controller gener... more ABSTRACT The majority of machine learning techniques applied to learning a robot controller generalise over either a uniform or pre-defined representation that is selected by a human designer. The approach taken in this paper is to reduce the reliance on the human designer by adapting the representation to improve the generalisation during the learning process. An extension of a Hierarchical Fuzzy Rule-Based System (HFRBS) is proposed that identifies and refines inaccurate regions of a fuzzy controller, while interacting with the environment, for both supervised and reinforcement learning problems. The paper shows that a controller using an adaptive HFRBS can learn a suitable control policy using a fewer number of fuzzy rules for both a supervised and reinforcement learning problem and is not sensitive to the layout as with a uniform representation. In supervised learning problems, a small number of extra trials are required to find an effective representation but for reinforcement learning problems, the process of adapting the representation is shown to significantly reduce the time taken to learn a suitable control policy and hence open the door to high-dimensional problems.
International Journal of Pattern Recognition and Artificial Intelligence, 2005
In this article, we describe some of the important currently used methods for solving classificat... more In this article, we describe some of the important currently used methods for solving classification problems, focusing on feature selection and extraction as parts of the overall classification task. We then go on to discuss likely future directions for research in this area, in the context of the other articles from this special issue. We propose that the next major step is the elaboration of a theory of how the methods of selection and extraction interact during the classification process for particular problem domains, along with any learning that may be part of the algorithms. Preferably this theory should be tested on a set of well-established benchmark challenge problems. Using this theory, we will be better able to identify the specific combinations that will achieve best classification performance for new tasks.
Uploads
Papers by Brian Carse