I've redone my treatment of this material, and I encourage you to come to the talk if you have the time because I'm interested in your feedback.
Taking into account previous comments from the lab about not understanding the basic ideas, I will focus more on the high-level approach to the problem, and the general ideas. I will spend less time on certain details and the experiments, which is too bad, but perhaps necessary given the time constraints.
Joseph
On Thursday 10 January 2008 14:26:02 Yoshua Bengio wrote:
Begin forwarded message:
From: Guy Lapalme lapalme@iro.umontreal.ca Date: January 9, 2008 9:07:44 PM EST To: rali-olst@iro.umontreal.ca Subject: Séminaire RALI-OLST: JEUDI le 17 janvier
Large-Scale Structure Prediction by Classification Joseph Turian Laboratoire LISA, DIROJEUDI 17 janvier, 11:30, Salle 3195, Pavillon André-Aisenstadt ^^^^^ Jour de la semaine inhabituel
We present an approach to structure prediction based upon a reduction to binary classification. This binary classification task typically involves many examples and many features. We propose a learning method for these large-scale classification tasks that is simple, accurate, and efficient. We apply this approach large-scale structure prediction problems in NLP: parsing and machine translation.
Guy Lapalme 514-343-2145