Statistical Learning Theory and Applications to Microarray Data
Speaker: Sayan Mukherjee , Center for Biological and Computational Learning & Department of Brain and Cognitive Sciences, MIT
Date: November 22 2000
This talk will be comprised of three parts: (a) Classification and gene selection of DNA Microarray data using Support Vector Machine (SVM) classification, (b) Generalization bounds based on algorithmic stability for Minimum Relative Entropy Discrimination (MRED) and SVM classification, and (c) the Support Vector Method for multivariate density estimation.
The first part will address the problem of classifying molecular patterns and inferring which genes are relevant in this classification. We look at problems of predicting cancer morphologies and treatment outcomes and infer which genes are relevant in these predictions. Specifically we look at morphology data for Leukemia, Lymphoma, and Brain tumors. For treatment outcome we look at Diffuse Large B Cell Lymphoma and Medullablastomas. We state error rates for both types of problems and survival statistics for outcome prediction, all of these are statistically significant. Some details will be given about SVM classification with feature selection.
In the second part, generalization bounds for both Minimum Relative Entropy Discrimination (MRED) and Support Vector Machine (SVM) classification are stated. These bounds are derived using a new idea of algorithmic stability rather than finite VC or V-gamma dimension.
The third part will describe the Support Vector Method for multivariate density estimation. A theoretical motivation for this algorithm will be given and experimental results will be stated for some toy problems.
See other events that are part of Brains & Machines Seminar Series Fall 2000
See other events happening in November 2000