s1_e0
done · peak unseen_mismatch 51.2
α schedule
α 1.0 ×50 α 0.9565 ×50 α 0.913 ×50 α 0.8696 ×50 α 0.8261 ×50 α 0.7826 ×50 α 0.7391 ×50 α 0.6957 ×50 α 0.6522 ×50 α 0.6087 ×50 α 0.5652 ×50 α 0.5217 ×50 α 0.4783 ×50 α 0.4348 ×50 α 0.3913 ×50 α 0.3478 ×50 α 0.3043 ×50 α 0.2609 ×50 α 0.2174 ×50 α 0.1739 ×50 α 0.1304 ×50 α 0.087 ×50 α 0.0435 ×50 α 0.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.0
seen_* probed at reference α = 1.4 · unseen_* = held-out pairings, uniform (α-independent) · 1000 pairs per condition
| iteration | seen_match | seen_mismatch | unseen_match | unseen_mismatch |
|---|---|---|---|---|
| 50 | 50.2 ± 1.0 | 49.1 ± 1.5 | 49.8 ± 1.6 | 51.2 ± 1.6 |
| 100 | 51.0 ± 3.0 | 49.5 ± 1.8 | 51.0 ± 2.1 | 48.8 ± 2.4 |
| 150 | 56.0 ± 12.7 | 48.1 ± 2.8 | 54.4 ± 13.5 | 49.2 ± 3.1 |
| 200 | 72.3 ± 18.1 | 39.7 ± 8.5 | 70.6 ± 19.1 | 41.1 ± 8.7 |
| 250 | 94.8 ± 11.0 | 19.3 ± 17.5 | 94.3 ± 11.8 | 19.7 ± 17.9 |
| 300 | 100.0 ± 0.0 | 11.7 ± 14.8 | 100.0 ± 0.0 | 11.9 ± 15.3 |
| 350 | 100.0 ± 0.0 | 14.4 ± 13.1 | 100.0 ± 0.0 | 15.1 ± 14.1 |
| 400 | 100.0 ± 0.0 | 16.7 ± 15.7 | 100.0 ± 0.0 | 17.2 ± 15.9 |
| 450 | 100.0 ± 0.0 | 31.4 ± 18.0 | 100.0 ± 0.0 | 31.6 ± 18.4 |
| 500 | 100.0 ± 0.0 | 21.3 ± 22.0 | 100.0 ± 0.0 | 21.2 ± 22.0 |
| 550 | 100.0 ± 0.0 | 12.8 ± 18.7 | 100.0 ± 0.0 | 13.2 ± 19.3 |
| 600 | 100.0 ± 0.0 | 28.3 ± 18.9 | 100.0 ± 0.0 | 28.1 ± 19.8 |
| 650 | 100.0 ± 0.0 | 32.5 ± 30.4 | 100.0 ± 0.0 | 32.7 ± 30.2 |
| 700 | 100.0 ± 0.0 | 23.6 ± 25.4 | 100.0 ± 0.0 | 23.8 ± 25.7 |
| 750 | 100.0 ± 0.0 | 24.0 ± 23.6 | 100.0 ± 0.0 | 24.4 ± 24.2 |
| 800 | 100.0 ± 0.0 | 11.7 ± 16.7 | 100.0 ± 0.0 | 11.7 ± 17.0 |
| 850 | 100.0 ± 0.0 | 23.8 ± 18.8 | 100.0 ± 0.1 | 24.0 ± 19.3 |
| 900 | 100.0 ± 0.0 | 13.5 ± 17.4 | 100.0 ± 0.0 | 13.2 ± 17.2 |
| 950 | 100.0 ± 0.0 | 36.3 ± 28.9 | 100.0 ± 0.0 | 36.0 ± 29.6 |
| 1000 | 100.0 ± 0.0 | 23.9 ± 19.6 | 100.0 ± 0.0 | 24.6 ± 20.2 |
| 1050 | 100.0 ± 0.0 | 20.6 ± 18.7 | 100.0 ± 0.0 | 20.6 ± 18.9 |
| 1100 | 100.0 ± 0.0 | 17.0 ± 18.7 | 100.0 ± 0.0 | 16.7 ± 18.3 |
| 1150 | 100.0 ± 0.0 | 19.1 ± 23.5 | 100.0 ± 0.0 | 18.4 ± 23.6 |
| 1200 | 100.0 ± 0.0 | 26.1 ± 21.0 | 100.0 ± 0.0 | 25.5 ± 20.5 |