Large Margin Learning Algorithms

Machine learning algorithms solve a range of problems concerning inference from finite data sets, including, for example, pattern classification. Large margin algorithms (which include support vector machines and boosting) have recently been shown to both perform well in practice and have rigorous theoretical performance guarantees. The present proposal aims to systematically develop and extend the theory and practice of these algorithms. Expected outcomes include a unified and tighter theory of generalization performance and highly efficient implementations on commonly available computer hardware. This will lead to more effective methods of inference (with performance guarantees) for both large and small data sets.