arxivst stuff from arxiv that you should probably bookmark

Beating Atari with Natural Language Guided Reinforcement Learning

Abstract · Apr 18, 2017 21:31 ·

montezuma completing revenge agent score openai atari gym instructions cs-ai

Arxiv Abstract

  • Russell Kaplan
  • Christopher Sauer
  • Alexander Sosa

We introduce the first deep reinforcement learning agent that learns to beat Atari games with the aid of natural language instructions. The agent uses a multimodal embedding between environment observations and natural language to self-monitor progress through a list of English instructions, granting itself reward for completing instructions in addition to increasing the game score. Our agent significantly outperforms Deep Q-Networks (DQNs), Asynchronous Advantage Actor-Critic (A3C) agents, and the best agents posted to OpenAI Gym on what is often considered the hardest Atari 2600 environment: Montezuma’s Revenge.

Read the paper (pdf) »