May 1st, Wednesday 14:30, Room 303, Jacobs Building

Title: Learning with Confidence

Lecturer: Koby Crammer

Lecturer homepage : http://webee.technion.ac.il/people/koby/

Affiliation : Department of Electrical Engineering, The Technion

 

I will introduce a learning approach that maintain confidence information about parameters. Learned hypotheses are represented both as weights and covariance-like matrix that represents uncertainty and correlations between different weights. Learning in this framework updates parameters by estimating weights and increasing model confidence.

Few online algorithms for binary and multi-class categorization will be described. A loss bound analysis shows that indeed the algorithm performs better under some conditions and also relates between our model and the margin and loss analysis of previous models.

Empirical evaluation on a range of NLP tasks show that our algorithm improves over other state of the art in classification, learns faster on-the-fly, and improves in various settings.

Based on joint work with Mark Dredze, Alex Kulesza, Francesco Orabona, Matan Orbach and Fernando Pereira