Continual Learning on Atari Normalized continual learning
2.38Max PerformanceDV3
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| DV3Task Order=Default Task Order, 85% Threshold=2.02, Frame at Middle=4,423,6802026.03 | 2.38 | 5,652,480 | 80 | |
| DV3Task Order=Two-Cycle Training, 85% Threshold=1.17, Frame at Middle=4,423,6802026.03 | 1.38 | 3,112,960 | 60 | |
| ARROWTask Order=Two-Cycle Training, 85% Threshold=1.17, Frame at Middle=4,423,6802026.03 | 1.04 | 3,440,640 | 60 | |
| ARROWTask Order=Reversed Task Order, 85% Threshold=0.82, Frame at Middle=4,423,6802026.03 | 0.96 | 5,079,040 | 80 | |
| ARROWTask Order=Default Task Order, 85% Threshold=2.02, Frame at Middle=4,423,6802026.03 | 0.88 | — | 0 | |
| DV3Task Order=Reversed Task Order, 85% Threshold=0.82, Frame at Middle=4,423,6802026.03 | 0.31 | — | 0 | |
| TES-SACTask Order=Two-Cycle Training, 85% Threshold=1.17, Frame at Middle=4,423,6802026.03 | 0.18 | — | 0 | |
| TES-SACTask Order=Default Task Order, 85% Threshold=2.02, Frame at Middle=4,423,6802026.03 | 0.16 | — | 0 | |
| TES-SACTask Order=Reversed Task Order, 85% Threshold=0.82, Frame at Middle=4,423,6802026.03 | 0.15 | — | 0 |