*********************************
There is now a CONTENT FREEZE for Mercury while we switch to a new platform. It began on Friday, March 10 at 6pm and will end on Wednesday, March 15 at noon. No new content can be created during this time, but all material in the system as of the beginning of the freeze will be migrated to the new platform, including users and groups. Functionally the new site is identical to the old one. webteam@gatech.edu
*********************************
"Learning Theory 2.0: New Theoretical Insights for Modern Machine Learning Problems"
Abstract:
Over the past twenty years, applications of machine learning have grown more and more varied ranging from spam detection to computational biology to astronomy. Moreover, many of these application areas have faced a huge increase in the volume of available data of various kinds. In order to better use all this data a number of powerful new learning approaches have been proposed and explored. In particular, a major direction in machine learning research nowadays is incorporating unlabeled data together with labeled data in the learning process, which is known as Semi-Supervised Learning. Another increasingly important research direction is bringing interaction into the learning process; this is generically called Active Learning. These approaches have been intensely explored in the machine learning community, with many heuristics and specific algorithms, as well as various successful experimental results reported. Unfortunately, however, the standard theoretical models do not capture the key issues involved in these learning techniques, and it has become clear that for developing robust, versatile, and general algorithms in these settings a more fundamental understanding is necessary. In this talk we discuss new theoretical frameworks as well as new and general algorithms for both Active Learning and Semi-Supervised Learning.
In the context of Kernel methods (another flourishing area of machine learning research), we discuss a way of analyzing them that matches the standard intuition that a good kernel function is one that acts as a good measure of similarity. Building on insights and techniques we develop for all these learning problems, we also propose a new approach to analyzing the classic problem of Clustering, which has not been satisfactorily captured by existing models.
