Reinforcement Learning project for the Computational Cognitive Modelling class taught by Prof. Brenden Lake at NYU.
The Deep Q-learning algorithm has gained wide attention as one of the first successful combination of deep neural networks and reinforcement learning. Its promise was demonstrated on the Arcade Learning Environment, a challenging framework composed of dozens of Atari 2600 games, which is widely used to evaluate general competency in the field of ArtificialIntelligence (AI). In this project, we train a computer agent to play the popular classic arcade game Pong and transfer the knowledge learned from this game to see if the agent can also play Breakout, without any prior knowledge of this new game. The model is a convolutional neural network, trained using Deep Q-learning, whose input are raw pixels and whose output is a value function estimating future rewards. We investigate the ability of humans to generalize their prior experiences to new unseen environments and see whether we can achieve this element of human cognition through different variants of transfer learning.
Check out the full paper here - Knowldge transfer in Reinforcement Learning