| Home | Project | I.Q. Program |
| Project Home | Introduction | Methods | Results | Submissions | Links | References |
Title: MML, inverse learning, and medical data-sets
Supervisors: A./Prof. D. L. Dowe, Dr P. E. Tischer
Abstract:Bayesian Networks (BNs) model data to infer the probability of a certain outcome. Conditional Probability Distribution (CPDs) specifies the frequency distributions for every possible value an attribute can take. Large dimensionality of data result in very complex CPDs which makes it difficult to create the CPDs and to infer some property of the model once created. This project extends the work of Comley and Dowe(2003, 2004) based on ideas from Dowe and Wallace(1998) on improving BNs. Techniques for estimating complex CPDs are investigated. In particular, factor analysis, logistic regression and distance preserving projections are explored.