Reinforcement Learning on Atari 2600 (Specific Game Scores and Win Rates)
242Asterix ScoreDQN with Plasticity (rational)
Evaluation Results
| Method | Links | |||||||||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| DQN with Plasticity (rational)Algorithm=DQN with Plasticity, Activation=rational2021.02 | 242 | 70.1 | 1,134 | 141 | 308 | 107 | 107 | 120 | 16.3 | -59.5 | 42.3 | 257.8 | 341 | 130 | 1,616 | — | — | |
| DQN with Plasticity (joint-rational)Algorithm=DQN with Plasticity, Activation=joint-rational2021.02 | 168 | 77.4 | 1,210 | 129 | 312 | 193 | 107.3 | 117 | 18.4 | -60.2 | 95.1 | 258.3 | 253 | 134 | 906 | — | — | |
| DDQNAlgorithm=DDQN, Activation=LReLU2021.02 | 48.9 | 68.2 | 286 | 47.7 | 10.7 | 17.2 | 91.3 | 74 | 2.17 | -86.9 | 31 | 32.1 | 6.61 | 24.4 | 626 | — | — | |
| DQN with PlasticityAlgorithm=DQN with Plasticity, Activation=PELU2021.02 | 25.8 | 46.6 | 788 | 24.5 | 74.2 | 57.7 | 106.4 | 101 | 9.21 | -111 | 50.1 | 106 | 124 | 91.6 | 299 | — | — | |
| DQNAlgorithm=DQN, Activation=d+SiLU2021.02 | 2.14 | 11.3 | 11.7 | 0.37 | 5.28 | 13.9 | 104 | 2.74 | 0.18 | -85.5 | 32.4 | 78.5 | 18.3 | 2.89 | -4.03 | — | — | |
| DQNAlgorithm=DQN, Activation=LReLU2021.02 | 1.85 | 11.4 | 558 | 16.3 | 8.62 | 11.8 | 101 | 55.4 | 0.57 | -90.7 | 33.9 | 8.94 | 14.9 | 0.03 | 440 | — | — | |
| DQNAlgorithm=DQN, Activation=SiLU2021.02 | 0.52 | 21.2 | 93.9 | 37 | 6.08 | 128 | 96.1 | 14.2 | 3.67 | -111 | 33.1 | 26.3 | 19.3 | 58.2 | 55.8 | — | — |