early_fast
done · peak unseen_mismatch 90.7
α schedule
α 3.6919 ×25 α 3.1812 ×25 α 2.8799 ×25 α 2.6548 ×25 α 2.4711 ×25 α 2.3142 ×25 α 2.1762 ×25 α 2.0523 ×25 α 1.9395 ×25 α 1.8357 ×25 α 1.7392 ×25 α 1.649 ×25 α 1.5641 ×25 α 1.4839 ×25 α 1.4077 ×25 α 1.3351 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.3351
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 | 51.2 ± 2.0 | 49.9 ± 1.5 | 50.1 ± 2.1 | 50.4 ± 1.5 |
| 50 | 51.9 ± 2.0 | 50.8 ± 1.3 | 50.3 ± 1.8 | 49.9 ± 1.9 |
| 75 | 52.5 ± 2.2 | 50.5 ± 2.1 | 50.6 ± 2.1 | 49.1 ± 1.4 |
| 100 | 58.4 ± 7.6 | 54.0 ± 4.5 | 54.3 ± 6.1 | 51.1 ± 2.2 |
| 125 | 70.9 ± 13.2 | 64.7 ± 10.8 | 64.4 ± 13.2 | 57.7 ± 9.6 |
| 150 | 85.6 ± 10.8 | 78.0 ± 12.6 | 78.3 ± 13.9 | 70.0 ± 12.6 |
| 175 | 93.1 ± 5.4 | 88.2 ± 7.3 | 85.9 ± 9.4 | 79.2 ± 9.4 |
| 200 | 95.9 ± 2.6 | 92.6 ± 4.9 | 89.5 ± 6.7 | 84.0 ± 7.6 |
| 225 | 97.2 ± 1.9 | 94.9 ± 2.2 | 90.9 ± 6.2 | 86.1 ± 5.8 |
| 250 | 97.5 ± 1.4 | 95.3 ± 1.8 | 91.6 ± 4.5 | 86.7 ± 4.9 |
| 275 | 98.0 ± 1.0 | 96.1 ± 1.4 | 91.7 ± 4.1 | 86.7 ± 5.2 |
| 300 | 98.2 ± 0.9 | 96.3 ± 1.2 | 92.1 ± 3.7 | 86.7 ± 4.8 |
| 325 | 98.7 ± 0.7 | 97.1 ± 1.0 | 93.9 ± 3.1 | 88.3 ± 4.9 |
| 350 | 99.0 ± 0.6 | 97.7 ± 1.0 | 94.2 ± 3.1 | 88.5 ± 4.3 |
| 375 | 99.1 ± 0.5 | 97.8 ± 1.1 | 94.7 ± 2.6 | 89.0 ± 4.7 |
| 400 | 99.5 ± 0.3 | 98.3 ± 1.1 | 96.5 ± 1.8 | 90.7 ± 4.2 |