 |
|
Date: Thursday January the 10th, 2008
Time and location: 14:00-15:00, Room 3021 in Building 13. Speaker: Tijl De Bie (Department of Engineering Mathematics, University Of Bristol) (Joint work with Elisa Ricci and Nello Cristianini) In this talk I will describe a new approach to learning to predict over structured output spaces. The approach is conceptually simple and natural, and it has strong ties with established statistical tools such as Fisher's linear discriminant analysis. It involves a computationally benign convex optimisation problem that is extremely easy to implement. In terms of predictive performance, we will show that it is competetive with existing approaches. Furthermore, we provide a bound for the performance as quantified by the ranking loss. |
