p16
done · peak unseen_mismatch 66.6
α schedule
α 3.5668 ×25 α 3.6845 ×25 α 3.7728 ×25 α 3.8482 ×25 α 3.9155 ×25 α 3.9773 ×25 α 4.0348 ×25 α 4.089 ×25 α 4.1405 ×25 α 4.1896 ×25 α 4.2368 ×25 α 4.2824 ×25 α 4.3264 ×25 α 4.369 ×25 α 4.4105 ×25 α 4.4508 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.4508
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 | 50.9 ± 1.3 | 50.3 ± 1.6 | 50.0 ± 1.5 | 49.5 ± 1.8 |
| 50 | 51.2 ± 1.4 | 50.3 ± 1.6 | 49.9 ± 1.1 | 50.1 ± 1.7 |
| 75 | 52.3 ± 2.2 | 50.2 ± 1.5 | 50.5 ± 1.5 | 49.6 ± 1.4 |
| 100 | 57.4 ± 6.7 | 55.2 ± 4.5 | 52.2 ± 3.7 | 50.6 ± 2.1 |
| 125 | 69.1 ± 9.9 | 66.0 ± 9.2 | 58.6 ± 8.8 | 56.7 ± 7.6 |
| 150 | 78.8 ± 6.2 | 75.8 ± 6.3 | 65.2 ± 7.9 | 62.0 ± 7.3 |
| 175 | 82.6 ± 4.2 | 80.4 ± 3.7 | 67.1 ± 8.0 | 64.4 ± 7.8 |
| 200 | 82.6 ± 3.4 | 80.9 ± 3.0 | 65.5 ± 5.5 | 63.1 ± 5.5 |
| 225 | 83.2 ± 2.9 | 81.4 ± 2.9 | 66.0 ± 5.4 | 64.1 ± 5.8 |
| 250 | 83.5 ± 2.4 | 82.3 ± 2.2 | 65.7 ± 4.6 | 64.0 ± 4.6 |
| 275 | 84.1 ± 2.5 | 82.6 ± 2.1 | 66.1 ± 4.6 | 64.5 ± 4.7 |
| 300 | 85.2 ± 2.6 | 83.9 ± 2.5 | 67.4 ± 5.0 | 66.0 ± 4.7 |
| 325 | 85.8 ± 2.6 | 84.5 ± 2.3 | 67.8 ± 5.2 | 66.1 ± 4.6 |
| 350 | 85.8 ± 2.3 | 84.5 ± 2.3 | 67.6 ± 5.0 | 65.8 ± 4.3 |
| 375 | 86.3 ± 2.3 | 84.5 ± 2.3 | 67.5 ± 4.7 | 65.6 ± 4.6 |
| 400 | 87.0 ± 2.4 | 85.5 ± 2.2 | 68.9 ± 5.4 | 66.6 ± 4.8 |