In the work led by Wei Chen, we asked whether recurrent neural network can improve the ability of predicting psychiatric symptoms from people’s sequential decision making behavior compared to using parameters extracted by reinforcement learning models. Preprint to come out soon!
In the work led by John Day and Tushar Arora, we illustrated that a deep model driven by the principle of predictive learning can achieve the same abilities of segmenting objects from scenes, localizing them in 3D space and depth perception, all without supervision and only using signals available to the brain! The project page is here.