r1_c3
done · peak unseen_mismatch 90.4
α schedule
α 1.4079 ×25 α 1.417 ×25 α 1.4407 ×25 α 1.4825 ×25 α 1.5453 ×25 α 1.6312 ×25 α 1.7424 ×25 α 1.8806 ×25 α 2.0475 ×25 α 2.2445 ×25 α 2.4732 ×25 α 2.7348 ×25 α 3.0305 ×25 α 3.3617 ×25 α 3.7295 ×25 α 4.1348 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.1348
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 |
|---|---|---|---|---|
| 25 | 50.6 ± 1.3 | 49.8 ± 1.5 | 49.6 ± 1.4 | 49.6 ± 1.8 |
| 50 | 50.6 ± 1.7 | 50.1 ± 1.7 | 49.6 ± 1.3 | 49.8 ± 1.6 |
| 75 | 50.7 ± 1.7 | 49.5 ± 1.7 | 50.3 ± 1.5 | 49.5 ± 1.8 |
| 100 | 50.7 ± 2.6 | 49.9 ± 1.7 | 50.5 ± 1.7 | 49.7 ± 1.7 |
| 125 | 53.4 ± 5.7 | 48.8 ± 3.0 | 52.2 ± 5.5 | 48.6 ± 2.6 |
| 150 | 56.8 ± 12.5 | 48.5 ± 4.5 | 55.8 ± 12.7 | 48.2 ± 4.7 |
| 175 | 62.9 ± 17.3 | 47.2 ± 9.8 | 62.8 ± 17.3 | 47.5 ± 9.3 |
| 200 | 73.6 ± 22.0 | 50.2 ± 11.7 | 73.1 ± 22.1 | 50.0 ± 12.1 |
| 225 | 80.2 ± 21.9 | 59.7 ± 17.0 | 79.3 ± 22.1 | 58.9 ± 16.8 |
| 250 | 87.4 ± 18.6 | 69.0 ± 20.8 | 86.6 ± 19.3 | 68.2 ± 20.7 |
| 275 | 92.3 ± 14.1 | 84.0 ± 16.5 | 90.9 ± 15.8 | 82.1 ± 16.9 |
| 300 | 95.6 ± 8.7 | 89.2 ± 13.3 | 94.0 ± 11.1 | 86.7 ± 14.6 |
| 325 | 98.0 ± 3.5 | 93.4 ± 10.0 | 96.1 ± 6.5 | 90.4 ± 11.1 |
| 350 | 97.8 ± 2.7 | 93.8 ± 9.0 | 95.4 ± 5.4 | 89.9 ± 9.9 |
| 375 | 96.6 ± 2.5 | 94.5 ± 3.1 | 92.8 ± 5.3 | 89.7 ± 5.5 |
| 400 | 95.6 ± 2.6 | 94.1 ± 2.9 | 90.4 ± 5.7 | 87.5 ± 6.4 |