New Models for Relational Learning

Ricardo Silva (Cambridge University)

Monday, March 3rd, 2008
10 am - SENSQ 5317

Refreshments at 9:30 a.m

Hosted by

Abstract

The field of machine learning is becoming increasingly more important across different disciplines, a natural consequence oF the availability of more complex data sources. This is particularly true when the raw information comes as a relational database: we want our models to predict features of objects, but such objects are not independent of each other. One might be interested in classifying text documents which cite one another, or predicting features of proteins that interact with each other by different biological criteria, or predicting behavior of users that share common interests in a social network. The most common machine learning methods ignore this link information.

However, when predicting class labels for objects within a relational database, it is often helpful to consider a model for relationships: this allows for information between class labels to be shared and to improve prediction performance. There are different ways by which objects can be related. One traditional way corresponds to a Markov network structure: each existing relation is represented by an undirected edge. This encodes that, conditioned on input features, each object label is independent of other object labels given its neighbors in the relational graph. However, there is no reason why Markov networks should be the only representation of choice for symmetric dependence structures. Here we discuss the case when relationships are postulated to exist due to hidden common causes. We discuss how the resulting graphical model differs from Markov networks, and how it describes different types of real-world relational processes. A Bayesian nonparametric classification model is built upon this graphical representation and evaluated with several empirical studies.

Biography of Speaker

Ricardo Silva is a postdoctoral researcher at the Statistical Laboratory, University of Cambridge. Prior to that he was a Senior Research Fellow in the Gatsby Computational Neuroscience Unit, University College London. He was in the very first batch of students to graduate from the new Machine Learning PhD program organized by the School of Computer Science, Carnegie Mellon University, where he won a Microsoft Fellowship and a Siebel Scholarship. His research in machine learning focuses on graphical models, Bayesian inference, relational and causal learning. None of his models, however, ever predicted the Steelers were going to win the SuperBowl just after he decided to graduate and spend some time in the UK.