SOTAVerified

Prioritized Experience Replay

2015-11-18Code Available1· sign in to hype

Tom Schaul, John Quan, Ioannis Antonoglou, David Silver

Code Available — Be the first to reproduce this paper.

Reproduce

Code

Abstract

Experience replay lets online reinforcement learning agents remember and reuse experiences from the past. In prior work, experience transitions were uniformly sampled from a replay memory. However, this approach simply replays transitions at the same frequency that they were originally experienced, regardless of their significance. In this paper we develop a framework for prioritizing experience, so as to replay important transitions more frequently, and therefore learn more efficiently. We use prioritized experience replay in Deep Q-Networks (DQN), a reinforcement learning algorithm that achieved human-level performance across many Atari games. DQN with prioritized experience replay achieves a new state-of-the-art, outperforming DQN with uniform replay on 41 out of 49 games.

Tasks

Benchmark Results

DatasetModelMetricClaimedVerifiedStatus
Atari 2600 AlienPrior hsScore1,334.7Unverified
Atari 2600 AlienPrior noopScore4,203.8Unverified
Atari 2600 AmidarPrior noopScore1,838.9Unverified
Atari 2600 AmidarPrior hsScore129.1Unverified
Atari 2600 AssaultPrior hsScore6,548.9Unverified
Atari 2600 AssaultPrior noopScore7,672.1Unverified
Atari 2600 AsterixPrior hsScore22,484.5Unverified
Atari 2600 AsterixPrior noopScore31,527Unverified
Atari 2600 AsteroidsPrior hsScore1,745.1Unverified
Atari 2600 AsteroidsPrior noopScore2,654.3Unverified
Atari 2600 AtlantisPrior noopScore357,324Unverified
Atari 2600 AtlantisPrior hsScore330,647Unverified
Atari 2600 Bank HeistPrior hsScore876.6Unverified
Atari 2600 Bank HeistPrior noopScore1,054.6Unverified
Atari 2600 Battle ZonePrior hsScore25,520Unverified
Atari 2600 Battle ZonePrior noopScore31,530Unverified
Atari 2600 Beam RiderPrior noopScore23,384.2Unverified
Atari 2600 Beam RiderPrior hsScore31,181.3Unverified
Atari 2600 BerzerkPrior hsScore865.9Unverified
Atari 2600 BerzerkPrior noopScore1,305.6Unverified
Atari 2600 BowlingPrior noopScore47.9Unverified
Atari 2600 BowlingPrior hsScore52Unverified
Atari 2600 BoxingPrior hsScore72.3Unverified
Atari 2600 BoxingPrior noopScore95.6Unverified
Atari 2600 BreakoutPrior hsScore343Unverified
Atari 2600 BreakoutPrior noopScore373.9Unverified
Atari 2600 CentipedePrior noopScore4,463.2Unverified
Atari 2600 CentipedePrior hsScore3,489.1Unverified
Atari 2600 Chopper CommandPrior hsScore4,635Unverified
Atari 2600 Chopper CommandPrior noopScore8,600Unverified
Atari 2600 Crazy ClimberPrior hsScore127,512Unverified
Atari 2600 Crazy ClimberPrior noopScore141,161Unverified
Atari 2600 Demon AttackPrior noopScore71,846.4Unverified
Atari 2600 Demon AttackPrior hsScore61,277.5Unverified
Atari 2600 Double DunkPrior hsScore16Unverified
Atari 2600 Double DunkPrior noopScore18.5Unverified
Atari 2600 EnduroPrior noopScore2,093Unverified
Atari 2600 EnduroPrior hsScore1,831Unverified
Atari 2600 Fishing DerbyPrior noopScore39.5Unverified
Atari 2600 Fishing DerbyPrior hsScore9.8Unverified
Atari 2600 FreewayPrior noopScore33.7Unverified
Atari 2600 FreewayPrior hsScore28.9Unverified
Atari 2600 FrostbitePrior noopScore4,380.1Unverified
Atari 2600 FrostbitePrior hsScore3,510Unverified
Atari 2600 GopherPrior hsScore34,858.8Unverified
Atari 2600 GopherPrior noopScore32,487.2Unverified
Atari 2600 GravitarPrior hsScore269.5Unverified
Atari 2600 GravitarPrior noopScore548.5Unverified
Atari 2600 HEROPrior noopScore23,037.7Unverified
Atari 2600 HEROPrior hsScore20,889.9Unverified

Reproductions