Academic Commons

Chapters (Layout Features)

Competitively Evolving Decision Trees Against Fixed Training Cases for Natural Language Processing

Siegel, Eric V.

Competitive fitness functions can generate performance superior to absolute fitness functions [Angelineand Pollack 1993], [Hillis 1992]. This chapter describes a method by which competition can be implemented when training over a fixed (static) set of examples. Since new training cases cannot be generated by mutation or crossover, the probabilistic frequencies by which individual training cases are selected competitively adapt. We evolve decision trees for the problem of word sense disambiguation. The decision trees contain embedded bit strings; bit string crossover is intermingled with subtree-swapping. To approach the problem of overlearning, we have implemented a fitness penalty function specialized for decision trees which is dependent on the partition of the set of training cases implied by a decision tree.

Files

Also Published In

Title
Advances in Genetic Programming
Publisher
The MIT Press

More About This Work

Academic Units
Computer Science
Published Here
April 26, 2013
Academic Commons provides global access to research and scholarship produced at Columbia University, Barnard College, Teachers College, Union Theological Seminary and Jewish Theological Seminary. Academic Commons is managed by the Columbia University Libraries.