Improving transfer using augmented feedback in Progressive Neural Networks

Deepika Bablani, Parth Chadha

Abstract: Learning faster on a task by utilizing learned representations from previous similar tasks is an active area of research in reinforcement learning. Recently proposed progressive neural networks demonstrate this effectively. We use motivations from reciprocal feedback connections in the visual cortex to augment lateral connections in the progressive neural network architecture. We evaluate our modified architecture on Pong-v0 and its variants and show that it improves transfer over the progressive baseline.