Visual Analogies between Atari Games for Studying Transfer Learning in RL

TL;DR

This paper explores using unsupervised visual analogies to transfer knowledge between Atari games, enabling agents trained on one game to play another by mapping game states through learned visual correspondences.

Contribution

It introduces a method for creating visual analogies between game pairs and evaluates transfer learning approaches using these mappings in Atari games.

Findings

Convincing visual mappings between game pairs

Effective transfer of policies across games

Potential for unsupervised transfer learning in RL

Abstract

In this work, we ask the following question: Can visual analogies, learned in an unsupervised way, be used in order to transfer knowledge between pairs of games and even play one game using an agent trained for another game? We attempt to answer this research question by creating visual analogies between a pair of games: a source game and a target game. For example, given a video frame in the target game, we map it to an analogous state in the source game and then attempt to play using a trained policy learned for the source game. We demonstrate convincing visual mapping between four pairs of games (eight mappings), which are used to evaluate three transfer learning approaches.