July 24, 2019
CiNet 1F Conference Room
Wenjun Bai, Special Appointed Assistant Professor
Host: Masahiko Haruno (PI)
Learning an optimal affective representation is at the root of affective computing. However, the discussion on priors for affective representation learning receives limited research attention. Here, I propose that an optimal affective representation should be continuous and reward orientated. Through deriving a computational model with constraints on early visual learning, I will argue that it is possible to learn continuous affective representations without human-curated labels. One step further, I will share my thoughts on how to implement other biological constraints under the information-theoretic framework, and to recast affective representation learning under the reward based reinforcement learning.