Probabilistic Inductive Logic Programming
(invited lecture for ALT 2004)
Author: Luc De Raedt
Affiliation: Institute for Computer Science,
Universität Freiburg, Freiburg, Germany
Probabilistic inductive logic programming (sometimes also called
statistical relational learning) addresses
one of the central questions of artificial intelligence:
the integration of Probabilistic reasoning with
first order Logic representations and machine Learning.
In this talk, I shall start from an inductive logic programming
perspective and sketch how it can be extended with probabilistic methods.
More specifically, I shall outline three settings for inductive logic
learning from entailment, learning from interpretations and learning from
proofs or traces
and show how they can be used to learn different types of probabilistic
The learning from entailment setting is natural when learning stochastic
context free grammars and their upgrade, stochastic logic programs,
the learning from interpretations settings is the method of choice when
learning bayesian networks or bayesian logic programs, and learning from
proofs or traces correspond to learning (hidden) markov models and their
first order upgrades.
The resulting settings will also be illustrated using various real-life
examples from the field of bio-informatics.
This is joint work with Kristian Kersting and part of the EU project
APRIL II (Application of Probabilistic Inductive Logic Programming II).
The talk will partly be based on
De Raedt, L., Kersting, K., Probabilistic Logic Learning,
SIGKDD Explorations, Vol. 5(1), 2003.
©Copyright 2004 Author