Invited Speaker: Kyunghyun Cho, New York University Curiosity-driven Journey into Neural Sequence Models

摘要

In this talk, I take the audience on a tour of my earlier and recent experiences in building neural sequence models. I start from the earlier experience of using a recurrent net for sequence-to-sequence learning and talk about the attention mechanism. I discuss factors behind the success of these earlier approaches, and how these were embraced by the community even before they sota'd. 1 then move on to a more recent research direction in unconventional neural sequence models that automatically learn to decide on the order of generation.