FoCM 2014 conference
Workshop C3 - Learning Theory - Semi-plenary talk
December 20, 14:35 ~ 15:25 - Room B23
Kernel-based learning methods
Ingo Steinwart
University of Stuttgart, Germany - ingo.steinwart@mathematik.uni-stuttgart.de
The last decade has witnessed an explosion of data collected from various sources. Since in many cases these sources do not obey the assumptions of classical statistical approaches, new automated methods for interpreting such data have been developed in the machine learning community. Statistical learning theory tries to understand the statistical principles and mechanisms these methods are based on.
This talk begins by introducing some central questions considered in statistical learning. Then various theoretical aspects of a popular class of learning algorithms, which include support vector machines, are discussed. In particular, I will describe how classical concepts from approximation theory such as interpolation spaces and entropy numbers are used in the analysis of these methods. The last part of the talk considers more practical aspects including the choice of the involved loss function and some implementation strategies. In addition, I will present a data splitting strategy that enjoys the same theoretical guarantees as the standard approach but reduces the training time significantly.