Showing 1–1 of 1 results for author: Sobol, D

Visual Analogies between Atari Games for Studying Transfer Learning in RL

Authors: Doron Sobol, Lior Wolf, Yaniv Taigman

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 targ… ▽ More 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 map** between four pairs of games (eight map**s), which are used to evaluate three transfer learning approaches. △ Less

Submitted 29 July, 2018; originally announced July 2018.