On the Use of Pairwise Comparison of Hypotheses in Evolutionary Learning Applied to Learning from Visual Examples

  • Authors:
  • Krzysztof Krawiec

  • Affiliations:
  • -

  • Venue:
  • MLDM '01 Proceedings of the Second International Workshop on Machine Learning and Data Mining in Pattern Recognition
  • Year:
  • 2001

Quantified Score

Hi-index 0.00

Visualization

Abstract

This paper is devoted to the use of genetic programming for the search of hypothesis space in visual learning tasks. The long-term goal of our research is to synthesize human-competitive procedures for pattern discrimination by means of learning process based directly on the training set of images. In particular, we introduce a novel concept of evolutionary learning employing, instead of scalar evaluation function, pairwise comparison of hypotheses, which allows the solutions to remain incomparable in some cases. That extension increases the diversification of the population and improves the exploration of the hypothesis space search in comparison with "plain" evolutionary computation using scalar evaluation. This supposition is verified experimentally in this study in an extensive comparative experiment of visual learning concerning the recognition of handwritten characters.