Ent?te

Logo du LIFL

Depuis le 1er janvier 2015 le LIFL et le LAGIS forment le laboratoire CRIStAL

  1. Doctoral studies

Thesis of

Jean Decoster

Thursday 17 July 2014
Amphithéâtre de l'IRCICA

Inductive logic programing for tree classification and transformation

Directeur de Thèse : REMI GILLERON Rapporteurs : CELINE ROUVEIROL ANTOINE CORNUEJOLS Membres : SLAWOMIR STAWORKO FABIEN TORRE

The recent proliferation of XML documents in databases and web applications rises some issues due to the numerous data exchanged and their diversity. To ease their uses, some smart means have been developed such as automatic classification and transformation.

This thesis has two goals:

We have chosen to use Inductive Logic Programming. The expressiveness of logic programs grants flexibility in specifying the learning task and understandability to the induced theories. This flexibility implies a high computational cost, constraining the applicability of ILP systems. However, XML documents being trees, a good concession can be found.

For our first contribution, we define clauses languages that allow encoding xml  trees. The definition of our classification framework follows their studies. It stands on a rewriting of the standard ILP operations such as theta-subsumption and least general generalization [Plotkin1971]. Our algorithms are polynomials in time in the input size whereas the standard ones are exponentials. They grant an identification in the limit [Gold1967] of our languages.

Our second contribution is the building of methods to learn XML documents transformations. It begins by the definition of a clauses class in the way of functional programs [Paulson91]. They are an ILP adaptation of edit scripts and allow a context. Their learning is possible thanks to two A*-like algorithms, a common ILP approach (HOC-Learner [Santos2009]).

Ours

UMR 8022 - Laboratoire d'Informatique Fondamentale de Lille - Copyright © 2012 Sophie TISON - Crédits & Mentions légales

Page respectant XHTML et CSS.

Pour tout commentaire / Comments and remarks : webmaster