p19
done · peak unseen_mismatch 86.6
α schedule
α 4.0804 ×25 α 3.9594 ×25 α 3.824 ×25 α 3.6796 ×25 α 3.5286 ×25 α 3.3722 ×25 α 3.2113 ×25 α 3.0465 ×25 α 2.8782 ×25 α 2.7068 ×25 α 2.5325 ×25 α 2.3557 ×25 α 2.1764 ×25 α 1.9949 ×25 α 1.8113 ×25 α 1.6257 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.6257
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 | 50.4 ± 1.7 | 49.8 ± 1.5 | 49.4 ± 1.9 |
| 50 | 51.4 ± 1.5 | 50.3 ± 1.7 | 50.0 ± 1.1 | 49.9 ± 1.6 |
| 75 | 52.6 ± 2.2 | 50.2 ± 1.7 | 50.7 ± 1.3 | 49.4 ± 1.6 |
| 100 | 58.3 ± 7.6 | 56.1 ± 4.5 | 52.8 ± 4.3 | 51.1 ± 2.8 |
| 125 | 70.3 ± 11.1 | 67.0 ± 10.4 | 59.8 ± 9.5 | 57.4 ± 8.3 |
| 150 | 79.6 ± 7.1 | 77.0 ± 6.7 | 66.4 ± 8.6 | 63.4 ± 7.4 |
| 175 | 84.3 ± 4.7 | 81.7 ± 4.3 | 69.8 ± 8.6 | 66.7 ± 8.4 |
| 200 | 86.5 ± 3.1 | 84.2 ± 3.3 | 70.8 ± 6.7 | 68.3 ± 6.6 |
| 225 | 88.4 ± 2.6 | 86.1 ± 2.5 | 72.5 ± 5.6 | 69.9 ± 5.8 |
| 250 | 90.0 ± 2.3 | 87.9 ± 2.1 | 73.6 ± 5.1 | 71.1 ± 4.9 |
| 275 | 91.8 ± 2.0 | 90.0 ± 1.5 | 76.6 ± 5.0 | 74.0 ± 4.5 |
| 300 | 93.3 ± 1.6 | 91.4 ± 2.0 | 79.7 ± 4.7 | 76.4 ± 4.9 |
| 325 | 94.6 ± 1.1 | 93.1 ± 1.3 | 81.5 ± 3.5 | 78.0 ± 4.9 |
| 350 | 96.2 ± 0.9 | 94.5 ± 1.3 | 85.2 ± 3.7 | 81.3 ± 3.8 |
| 375 | 97.3 ± 0.9 | 95.9 ± 1.2 | 87.9 ± 3.1 | 83.3 ± 3.9 |
| 400 | 98.3 ± 0.6 | 97.1 ± 1.0 | 91.1 ± 3.6 | 86.6 ± 4.0 |