carlb15/Humans_and_RL_in_Atari

In this project, we attempt to equip the agent with the recognition of basic components of an Atari game environment through curriculum learning—gleaned from human developmental psychology—and evaluate its performance. Our best agent was pre-trained on a carefully designed curriculum to learn to complete a new game 5x faster than regular agents. By analyzing our agent’s behavior, we can examine how it acquires knowledge and skills differently from humans.

Python