Transfer Learning in Deep Reinforcement Learning

In this project, we proposed a new nested neural network architecture to be used for deep reinforcement learning. To do this, we combined two recently proposed architectures, Actor-Mimic and Progressive Neural Networks, to speed-up the learning process, to save space, and to improve transfer learning. We have tested our approach on a set of Atari Games and our results show some interesting trends and improvement in comparison to the state-of-the-art models. More details maybe found in our final paper.