Workshop on Machine Learning in Speech and Language Processing

August 11, 2017
Sydney, Australia
Speaker: Jason Weston (Facebook)

Title: (Towards) Learning from Conversing

Abstract:
An (end-to-end) dialogue agent might be far from knowing and understanding everything, but _if_ it can learn _while_ it is conversing with humans, maybe it can move towards that goal? We look at some key ingredients we will need: (i) ability to learn from textual feedback, i.e. the things said to it, (ii) ability to ask (useful) questions and learn from the replies, (iii) bootstrapping the model so that it doesn't take 6 months - 2 years to say something (like a human baby), (iv) making this a elegant unified system, and not a bunch of hacks. This talk describes joint work with Jiwei Li, Alexander H. Miller, Sumit Chopra and Marc'Aurelio Ranzato.