Deep and Self-organizing Neural Networks for Affective Modeling

Modeling affective mechanisms from humans in computer systems is a very difficult task, even with the recent advances of??deep neural networks. This project focus on two different mechanisms of affective computing: perception and intrinsic modeling.?? On the perception side, one of the crucial problems is the subjectivity of emotion description, as different persons can express and perceive emotions differently, depending on several contextual factors. On the intrinsic modeling side, the constraints on emotion representation and its modulation by perception and action must be taken into consideration. This project proposes a neural framework for dealing with emotion perception, modeling, and modulation over different affective mechanisms with the use of deep and self-organizing networks.

One of the outcome of this project was the Facechannel software, which aimed to create light-weighted neural networks for facial expression description, relying on innovative implementation and training mechanisms. The outcome of this project is a ready-to-use model built in python for the description of facial expressions on images and videos.

To know more about it, please refer to the project page .