r6_c2
done · peak unseen_mismatch 87.3
α schedule
α 1.6201 ×25 α 2.4551 ×25 α 3.0472 ×25 α 1.2272 ×25 α 1.3757 ×25 α 4.3496 ×25 α 4.704 ×25 α 5.0 ×25 α 1.8547 ×25 α 1.9566 ×25 α 2.0537 ×25 α 5.0 ×25 α 5.0 ×25 α 2.322 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.322
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 | 50.5 ± 1.5 | 50.2 ± 1.5 | 50.0 ± 1.6 | 49.3 ± 1.9 |
| 50 | 51.1 ± 1.6 | 50.1 ± 1.6 | 50.4 ± 1.7 | 49.7 ± 1.5 |
| 75 | 51.1 ± 1.6 | 49.4 ± 1.5 | 50.5 ± 1.5 | 49.7 ± 1.8 |
| 100 | 50.9 ± 2.2 | 50.3 ± 1.3 | 50.3 ± 1.3 | 50.0 ± 1.5 |
| 125 | 53.2 ± 3.7 | 49.5 ± 2.0 | 52.0 ± 4.1 | 48.9 ± 2.2 |
| 150 | 60.5 ± 14.1 | 52.9 ± 7.5 | 59.0 ± 13.6 | 51.5 ± 7.3 |
| 175 | 70.3 ± 16.9 | 61.9 ± 11.5 | 67.3 ± 16.7 | 58.9 ± 10.3 |
| 200 | 79.3 ± 16.2 | 75.2 ± 14.5 | 75.4 ± 16.4 | 69.8 ± 15.2 |
| 225 | 87.6 ± 15.8 | 80.3 ± 15.8 | 84.6 ± 16.6 | 76.2 ± 16.0 |
| 250 | 92.5 ± 12.7 | 87.0 ± 12.4 | 89.3 ± 14.7 | 82.7 ± 12.9 |
| 275 | 95.7 ± 7.3 | 92.3 ± 8.7 | 92.4 ± 9.5 | 87.3 ± 9.5 |
| 300 | 95.1 ± 3.3 | 92.7 ± 3.6 | 89.7 ± 6.3 | 85.7 ± 6.5 |
| 325 | 93.6 ± 2.7 | 91.4 ± 2.7 | 86.4 ± 5.7 | 82.5 ± 5.8 |
| 350 | 95.7 ± 1.7 | 93.7 ± 2.1 | 88.3 ± 5.1 | 84.8 ± 5.7 |