Home Seminars Low dimensional coding and multi-task learning
 
Home Seminars Low dimensional coding and multi-task learning
Low dimensional coding and multi-task learning

Speaker: Massimilano Pontil (UCL, London, UK)

Date: 10/02/2010, h. 11.45

Location: DISI - room 326

We discuss a general coding method where data in a Hilbert space are represented by finite dimensional coding vectors. The method is based on empirical risk minimization within a certain class of linear operators, which map the set of coding vectors to the Hilbert space. We derive a bound on the expected reconstruction error of the method, which highlights the role played by the codebook and the class of linear operators. This result is specialized to some cases of practical importance, including K-means clustering, non-negative matrix factorization and other sparse coding methods. We also discuss extensions of these ideas to an incomplete data setting, drawing a connection to the problem of multi-task learning. (This is a joint work with Andreas Maurer)