Classifiers that approximate functions

  • Authors:
  • Stewart W. Wilson

  • Affiliations:
  • Prediction Dynamics, Concord MA 01742, USA/ Department of General Engineering, The University of Illinois at Urbana-Champaign IL 61801, USA (E-mail: Wilson@prediction-dynamics.com ...

  • Venue:
  • Natural Computing: an international journal
  • Year:
  • 2002

Quantified Score

Hi-index 0.00

Visualization

Abstract

A classifier system, XCSF, is introduced in which the predictionestimation mechanism is used to learn approximations to functions.The addition of weight vectors to the classifiers allowspiecewise-linear approximation, where the classifier'sprediction is calculated instead of being a fixed scalar. The weight vector and the classifier's condition co-adapt.Results on functions of up to six dimensions show high accuracy. The idea of calculating the prediction leads to the concept ofa generalized classifier in which the payoff prediction approximatesthe environmental payoff function over a subspace defined bythe classifier condition and an action restriction specified in theclassifier, permitting continuous-valued actions.