Robust bidding in LCS using loan and bid history

Traditional Learning Classifier Systems (LCS) learn syntactically simple string rules in an auction based competitive market economy by continuously interacting with their environment through a reinforcement program. All classifiers participating in an auction issue a bid proportional to their strength and a winner classifier is allowed to fire and receive a reward or punishment from its environment as a consequence of its action. However, in this kind of bidding strategy, optimal classifiers with low strength have to wait until the strength of bad classifiers has come down through continuous taxation. This slows down the convergence rate. In addition, as Genetic Algorithm(GA) has some degree of randomness, offspring classifiers that come from weak parents inherit a small strength as compared to experienced classifiers in the population and hence have to wait for some time till they mature and try their action. In this paper, we introduced bid history and loan concepts to mitigate the above shortcomings of the bidding strategy in traditional LCS. In direct analogy with real auctions, all classifiers matching the current input compare the average bid history with their potential bid based on their current strength. The average bid history parameter gives general information about the bid market (potential of competent classifiers) and determines the amount of loan a classifier should ask. We have also added a debt and due date parameters to the traditional LCS parameter list to keep track of the transaction status and accuracy and experience to grant or deny loan requests. The results obtained show a significant improvement on the performance of the system.12

[1]  D. E. Goldberg,et al.  Genetic Algorithms in Search , 1989 .

[2]  Gunar E. Liepins,et al.  A Classifier Based System for Discovering Scheduling Heuristics , 1987, ICGA.

[3]  John H. Holland,et al.  Induction: Processes of Inference, Learning, and Discovery , 1987, IEEE Expert.

[4]  W. Marsden.  I and J , 2012 .

[5]  G. Robertson,et al.  A Tale of Two Classifier Systems , 2005, Machine Learning.

[6]  Lashon B. Booker,et al.  Intelligent Behavior as an Adaptation to the Task Environment , 1982 .

[7]  Stewart W. Wilson.  Knowledge Growth in an Artificial Animal , 1985, ICGA.

[8]  John H. Holland,et al.  COGNITIVE SYSTEMS BASED ON ADAPTIVE ALGORITHMS1 , 1978 .

[9]  David E. Goldberg,et al.  Reinforcement learning with classifier systems , 1990, Proceedings [1990]. AI, Simulation and Planning in High Autonomy Systems.

[10]  Abdollah Homaifar,et al.  Boolean Function Learning With A Classifier System , 1988, Defense, Security, and Sensing.

[11]  Stephen F. Smith,et al.  A learning system based on genetic adaptive algorithms , 1980 .

[12]  David E. Goldberg,et al.  A Critical Review of Classifier Systems , 1989, ICGA.

[13]  John H. Holland,et al.  Escaping brittleness: the possibilities of general-purpose learning algorithms applied to parallel rule-based systems , 1995 .

[14]  John H. Holland,et al.  Cognitive systems based on adaptive algorithms , 1977, SGAR.

[15]  Stewart W. Wilson.  ZCS: A Zeroth Level Classifier System , 1994, Evolutionary Computation.

[16]  Paul Thagard,et al.  Induction: Processes Of Inference , 1989 .

[17]  David E. Goldberg,et al.  Genetic Algorithms in Search Optimization and Machine Learning , 1988 .

[18]  Stewart W. Wilson.  Classifier Fitness Based on Accuracy , 1995, Evolutionary Computation.