Learning I: Induction

Advances in Artifical Intelligence

Volume 1081 of the series Lecture Notes in Computer Science pp 227-239

Date:

LPMEME: A statistical method for inductive logic programming

  • Karan BhatiaAffiliated withDepartment of Computer Science and Engineering 0114, University of California, San Diego
  • , Charles ElkanAffiliated withDepartment of Computer Science and Engineering 0114, University of California, San Diego

* Final gross prices may vary according to local VAT.

Get Access

Abstract

This paper describes LPMEME, a new learning algorithm for inductive logic programming that uses statistical techniques to find first-order patterns. LPMEME takes as input examples in the form of logical facts and outputs a first-order theory that is represented to some degree in all of the examples. LPMEME uses an underlying statistical model whose parameters are learned using expectation maximization, an iterative gradient descent method for maximum likelihood parameter estimation. The underlying statistical model is described and the EM algorithm developed. Experimental tests show that LPMEME can learn first-order concepts and can be used to find approximate solutions to the subgraph isomorphism problem.

Keywords

learning inductive logic programming maximum likelihood parameter estimation expectation maximization