'Recurrent Attention Model and its Applications,' by Yiming Zhao

Department(s):

Data Science

Abstract

Recurrent attention model is a kind of human-like deep learning structure which can dramatically 

decrease the input dimension by focusing on a small area. In this paper, we will extend this model 

to deal with two real world problems. The first problem we want to do is to teach the computer 

automatically to notice the events by looking through videos. We have the labeled traffic videos, 

and we want our trained AI to find the traffic events by looking the videos itself. Secondly, we 

will combine the attention model with deep reinforcement learning method to teach AI to play video 

games. We think the intelligent attention model can handle those tasks better.

Keywords:   deep learning, reinforcement learning, attention model

Adviser: Xiangnan Kong, Ph.D.