Algorithms and applications for probabilistic relational models
The vast majority of real-world data is stored using relational representations. Unfortunately, many machine learning techniques are unable to handle rich relational models. Probabilistic Relational Models (PRMs) are an extension of the Bayesian network frame work which allows relational structure t...
| Main Author: | |
|---|---|
| Format: | Others |
| Language: | en |
| Published: |
McGill University
2005
|
| Subjects: | |
| Online Access: | http://digitool.Library.McGill.CA:80/R/?func=dbin-jump-full&object_id=98786 |
| Summary: | The vast majority of real-world data is stored using relational representations. Unfortunately, many machine learning techniques are unable to handle rich relational models. Probabilistic Relational Models (PRMs) are an extension of the Bayesian network frame work which allows relational structure to be fully exploited. They are a formalism based on relational logic for describing probabilistic models of structured data which also allow us to model uncertainty in the relationship between objects and the attributes of those objects. In this thesis we present an implementation of PRMs which allows defining conditional probability distributions over mixtures of discrete and continuous variables. This is an important new feature. We provide experimental results using our package on both synthetic and real data sets. |
|---|