|
Download: |
Machine learning techniques in biometricsProf. Alessandro Verri In this talk the learning from examples problem is presented within the framework of Regularization Networks. The important notion of Reproducing Kernel Hilbert Space is briefly reviewed. We then show that within this framework several learning methods can be easily obtained. In particular we derive Support Vector Methods and discuss their basic mathematical properties: existence, uniqueness, and consistency. We then illustrate methods for tuning the SVM parameters and in particular for selecting the regularization parameter. The main computational issues behind the implementation of SVMs are presented and, finally, some experimental results in biometric applications are described. |
 |
