p08
done · peak unseen_mismatch 79.6
α schedule
α 2.5183 ×25 α 2.5208 ×25 α 2.5288 ×25 α 2.5449 ×25 α 2.5714 ×25 α 2.6106 ×25 α 2.6645 ×25 α 2.7349 ×25 α 2.824 ×25 α 2.9333 ×25 α 3.0648 ×25 α 3.2201 ×25 α 3.4009 ×25 α 3.6089 ×25 α 3.8457 ×25 α 4.1128 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.1128
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 | 50.7 ± 1.4 | 50.2 ± 1.5 | 50.0 ± 1.4 | 49.4 ± 1.8 |
| 50 | 51.1 ± 1.4 | 50.2 ± 1.7 | 49.7 ± 1.0 | 50.0 ± 1.5 |
| 75 | 51.5 ± 1.8 | 49.8 ± 1.7 | 50.5 ± 1.4 | 49.3 ± 1.6 |
| 100 | 53.2 ± 4.6 | 51.0 ± 1.6 | 51.4 ± 3.3 | 50.0 ± 1.4 |
| 125 | 61.2 ± 12.8 | 55.2 ± 6.9 | 57.2 ± 12.6 | 51.8 ± 5.8 |
| 150 | 71.4 ± 16.2 | 67.1 ± 14.4 | 66.9 ± 15.7 | 62.3 ± 12.8 |
| 175 | 81.7 ± 15.1 | 77.1 ± 15.0 | 74.6 ± 16.9 | 70.1 ± 15.4 |
| 200 | 87.8 ± 10.7 | 84.1 ± 11.0 | 80.5 ± 13.0 | 75.8 ± 12.8 |
| 225 | 92.2 ± 5.7 | 88.7 ± 6.7 | 83.0 ± 9.6 | 79.4 ± 9.8 |
| 250 | 92.8 ± 3.4 | 90.6 ± 3.4 | 83.1 ± 7.4 | 79.6 ± 7.3 |
| 275 | 92.9 ± 2.6 | 90.8 ± 2.6 | 82.2 ± 6.5 | 79.1 ± 6.6 |
| 300 | 92.6 ± 2.4 | 90.0 ± 2.3 | 80.6 ± 5.7 | 77.4 ± 5.7 |
| 325 | 91.7 ± 2.2 | 89.5 ± 1.9 | 78.6 ± 4.7 | 75.4 ± 4.9 |
| 350 | 91.3 ± 1.8 | 89.3 ± 1.5 | 77.7 ± 3.1 | 74.2 ± 4.0 |
| 375 | 90.7 ± 2.1 | 88.3 ± 1.8 | 76.2 ± 3.3 | 72.6 ± 3.3 |
| 400 | 90.0 ± 1.6 | 87.7 ± 1.5 | 75.1 ± 3.6 | 71.6 ± 3.6 |