Accession Number : ADA327588

Title :   Analyzing the Performance of Learning Algorithms.

Descriptive Note : Final rept. 1 Oct 90-31 May 93,

Corporate Author : CALIFORNIA UNIV SANTA CRUZ

Personal Author(s) : Haussler, David ; Warmuth, Manfred K.

PDF Url : ADA327588

Report Date : 14 AUG 1993

Pagination or Media Count : 12

Abstract : We discuss the approach to the analysis of learning algorithms that we have taken in our laboratory and summarize the results we have obtained in the last few years. We have worked on refining and generalizing the PAC learning model introduced by Valiant. Measures of performance for learning algorithms that we have examined include computational complexity, sample complexity, probability of misclassification (learning curves), and worst case total number of misclassifications or hypothesis updates. We have looked for theoretically optimal bounds on these performance measures, and for learning algorithms that achieve these bounds. Learning problems we have examined include those for decision trees, neural networks, finite automata, conjunctive concepts on structural domains, and various classes of Boolean functions. We also worked on clustering data represented as sequences over a finite alphabet. Many of the new learning algorithms that we have developed have been tested empirically as well.

Descriptors :   *ALGORITHMS, *LEARNING, NEURAL NETS, COMPUTATIONS, MODELS, PROBABILITY, CLUSTERING, ERRORS, CLASSIFICATION, HYPOTHESES, SPECIAL FUNCTIONS(MATHEMATICS), AUTOMATA, LEARNING CURVES, ALPHABETS.

Subject Categories : Cybernetics

Distribution Statement : APPROVED FOR PUBLIC RELEASE