Skip to Main content Skip to Navigation
New interface
Theses

Apprentissage Supervisé Relationnel par Algorithmes d'Évolution

Sébastien Augier 1 
1 I&A
LRI - Laboratoire de Recherche en Informatique
Abstract : This thesis investigates the application of evolutionary-based algorithms to the problem of inducing relational rules from positive and negative examples. We first study a language bias that allows enough expressivity to cover at the same time relational learning from interpretations, and classical propositional languages. Even though the induction cost for these languages is characterized by NP-completeness on the subsumption test, a practical solution suitable for real complex problems is proposed. The SIAO1 system, which uses this language bias to learn relational rules, is then presented. It is based on an evolutionary search strategy characterized by: - mutation and crossover operators driven by background knowledge and learning examples; - a bottom-up search direction that respects the ordering relation defined over the language. SIAO1 is then compared to other systems on various classical machine learning bases, proving its polyvalency. These tests also show that this system compares well to the other approaches and that few learning and evaluation biases are demanded from the user. The third part of this work concerns two generic parallel architectures derived from asynchronous master-slave and pipeline models. These architectures are studied from the point of view of their scalability (ie. dependance on the dataset size) and expected speed-up. A simple but accurate prediction model to forecast the performance of these architectures is also proposed.
Complete list of metadata

Cited literature [148 references]  Display  Hide  Download

https://theses.hal.science/tel-00947322
Contributor : Sébastien Augier Connect in order to contact the contributor
Submitted on : Saturday, February 15, 2014 - 12:14:56 PM
Last modification on : Sunday, June 26, 2022 - 12:00:50 PM
Long-term archiving on: : Thursday, May 15, 2014 - 11:20:36 AM

Identifiers

  • HAL Id : tel-00947322, version 1

Collections

Citation

Sébastien Augier. Apprentissage Supervisé Relationnel par Algorithmes d'Évolution. Apprentissage [cs.LG]. Université Paris Sud - Paris XI, 2000. Français. ⟨NNT : ⟩. ⟨tel-00947322⟩

Share

Metrics

Record views

315

Files downloads

415