Learning linearly separable languages

Leonid Kontorovich, Corinna Cortes, Mehryar Mohri

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

This paper presents a novel paradigm for learning languages that consists of mapping strings to an appropriate high-dimensional feature space and learning a separating hyperplane in that space. It initiates the study of the linear separability of automata and languages by examining the rich class of piecewise-testable languages. It introduces a high-dimensional feature map and proves piecewise-testable languages to be linearly separable in that space. The proof makes use of word combinatorial results relating to subsequences. It also shows that the positive definite kernel associated to this embedding can be computed in quadratic time. It examines the use of support vector machines in combination with this kernel to determine a separating hyperplane and the corresponding learning guarantees. It also proves that all languages linearly separable under a regular finite cover embedding, a generalization of the embedding we used, are regular.

Original languageEnglish (US)
Title of host publicationAlgorithmic Learning Theory - 17th International Conference, ALT 2006, Proceedings
PublisherSpringer Verlag
Pages288-303
Number of pages16
ISBN (Print)3540466495, 9783540466499
DOIs
StatePublished - 2006
Event17th International Conference on Algorithmic Learning Theory, ALT 2006 - Barcelona, Spain
Duration: Oct 7 2006Oct 10 2006

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume4264 LNAI
ISSN (Print)0302-9743
ISSN (Electronic)1611-3349

Other

Other17th International Conference on Algorithmic Learning Theory, ALT 2006
Country/TerritorySpain
CityBarcelona
Period10/7/0610/10/06

ASJC Scopus subject areas

  • Theoretical Computer Science
  • General Computer Science

Fingerprint

Dive into the research topics of 'Learning linearly separable languages'. Together they form a unique fingerprint.

Cite this