报告题目: |
Novel Kernel-Based Supervised Machine Learning Techniques |
报告人: |
S.Y. Kung |
|
Professor Princeton University
|
报告时间: |
2009-07-20 15:00 |
报告地点: |
FIT楼1区315 |
主办单位: |
电子工程系 |
简介: |
Machine learning techniques are useful for data mining and pattern recognition when only pairwise relationships of the training objects are known as opposed to the training objects themselves. Such pairwise learning approach has a special appeal to many practical applications such as multimedia and bioinformatics, where a variety of heterogeneous sources of information are available. The primary information used in the kernel approach is either (1) the kernel matrix (K) associated with either vectorial data or (2) the similarity matrix (S) associated with nonvectorial objects.
Both the FDA and SVM classifiers have natural kernel extensions. Furthermore, there exists an intimate interplay between FDA and SVM methods in the kernel approach.
Three subjects will be presented:
. First, there are (non-unique) perfect KFD solutions for generic training datasets. Their characterization can be mathematically established.
. Second, in order to enhance the robustness of the trained classifiers, a perturbation analysis is introduced. This leads to a perturbation adjusted discriminant analysis which facilitates the regularization of classifiers.
.Finally, a hybrid FisherSVM classifier is proposed which combines the merits of two representation spaces: E and K. On one hand, the weaker spectral components in the E space can be regularized by the perturbed Fisher discriminant analysis. On the other hand, the constraints adopted by the SVM classifiers are formulated in terms of solutions in the K space. In this sense, the two regularization techniques used in the hybrid classifiers are complementary. |
|
