Efficient first-order probabilistic models for inference and learningA probabilistic model is any formalism to specify a complex probability distribution. Such formalisms facilitate uncertainty handling and evidential reasoning in artificial intelligence. Current probabilistic models restrict their variables to simple boolean propositions, discrete attributes, or numbers. The goal of this project was to enhance these models with the power of first-order logic. This enables the variables to range over complex structured objects, be they molecules or websites. The project has proposed several new methods for specifying such models, reasoning with them, and learning them from data. The approach uses the individual-centred representations that are a central topic of study in recent work in machine learning and inductive logic programming. Possible domains of application include molecular biology, drug design, information retrieval on the web, and user modelling. Experimental validation of the utility of first-order probabilistic models has been carried on several of these domains.
Staff and StudentsPeter Flach, Elias Gyftodimos.
- Final report
- Peter Flach, Elias Gyftodimos and Nicolas Lachiche. Probabilistic reasoning with terms. To appear in Electronic Transactions in Artificial Intelligence. [Submitted version, PDF]
- P. Flach and N. Lachiche. Naive Bayesian classification of structured data. Machine Learning 57(3): 233--269, 2004. [Pre-publication version, PDF]
- N. Lachiche and P. Flach. 1BC2: a true first-order Bayesian classifier. In: Proceedings of the 12th International Conference on Inductive Logic Programming, pages 133--148. Springer-Verlag, July 2002. [PDF]
- Elias Gyftodimos and Peter Flach. Hierarchical Bayesian Networks: an Approach to Classification and Learning for Structured Data. In: Proceedings of the ECML/PKDD - 2003 Workshop on Probablistic Graphical Models for Classification, pages 25--36. Ruder Boskovic Institute, Zagreb, Croatia, September 2003. [PDF]
- Elias Gyftodimos and Peter Flach. Hierarchical Bayesian Networks: A Probabilistic Reasoning Model for Structured Domains. In: Proceedings of the ICML-2002 Workshop on Development of Representations, Edwin de Jong and Tim Oates, editors, pages 23--30. The University of New South Wales, July 2002. [PDF]