Activities
|
Seminars
| Seminar on Theoretical and Computational Physics:
| | Title | Bayesian learning of kernels for large margin classifiers with extended Monte Carlo method
|
|---|
| Speaker | Dr. Hiroshi Tsukahara
|
|---|
| Affiliation | Department of research planning Denso IT Laboratory, Inc.Tokyo, Japan
|
|---|
| Date | Tuesday, 06-05-2008
|
|---|
| Time | Thứ năm 10g00
|
|---|
| Location | Meeting room, 1st floor, Institute of Physics & Electronics, 10 Daotan, Thule, Badinh, Hanoi
|
|---|
| Abstract | I will introduce typical problems in pattern recognition and application of learning machines for them. Then, I will introduce the concept of large margin classifiers, especially the support vector machines (SVMs) and the kernel trick. I will propose an approach to extend the usual kernel SVMs and to learn the kernel by bayesian inference. The extended Monte Carlo method is applied to estimate the posterior distribution for the bayesian inference. The effectiveness of the proposed approach is shown in specific problems and the priority of the exchange Monte Carlo method for evaluating stochastic complexity is emphasized. Another application of the present approach to the semi-supervised learning is also discussed.
|
|---|
|
