| | |
Summary: PhD thesis presentation
Statistical Learning Theory:
A PAC-Bayesian Approach
Jean-Yves Audibert
Université Pierre et Marie Curie
PhD advisor : Olivier Catoni
Statistical learning theory: a PAC-Bayesian approach J.-Y. AUDIBERT p. 1/3
Introduction
Statistical Learning Theory
how to make predictions about the future based on
past experiences
1. Aggregated estimators in L2 regression
2. A better variance control in classification
PAC-Bayesian complexities
Compression schemes complexities
3. Classification under Tsybakov's type assumptions
Statistical learning theory: a PAC-Bayesian approach J.-Y. AUDIBERT p. 2/3
Setup (1/2)
Training set: ZN
1 = Zi (Xi, Yi) : i = 1, . . . , N ,
|
