p11
done · peak unseen_mismatch 80.6
α schedule
α 3.5303 ×25 α 3.2284 ×25 α 3.0561 ×25 α 2.9296 ×25 α 2.8276 ×25 α 2.7413 ×25 α 2.666 ×25 α 2.5988 ×25 α 2.538 ×25 α 2.4822 ×25 α 2.4307 ×25 α 2.3827 ×25 α 2.3377 ×25 α 2.2953 ×25 α 2.2552 ×25 α 2.2171 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.2171
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 ± 1.4 | 49.5 ± 1.7 |
| 50 | 51.4 ± 1.3 | 50.3 ± 1.6 | 49.8 ± 1.1 | 50.0 ± 1.9 |
| 75 | 52.2 ± 1.9 | 50.0 ± 1.6 | 50.7 ± 1.6 | 49.2 ± 1.5 |
| 100 | 56.0 ± 6.1 | 53.5 ± 3.2 | 52.4 ± 3.7 | 50.3 ± 1.8 |
| 125 | 67.2 ± 12.5 | 62.2 ± 10.1 | 58.8 ± 11.6 | 55.3 ± 8.0 |
| 150 | 78.2 ± 12.7 | 74.9 ± 12.2 | 68.8 ± 12.9 | 65.2 ± 12.2 |
| 175 | 86.7 ± 6.5 | 84.0 ± 6.7 | 75.3 ± 10.9 | 71.7 ± 10.2 |
| 200 | 90.7 ± 3.9 | 88.2 ± 3.6 | 78.8 ± 8.2 | 75.6 ± 7.9 |
| 225 | 92.1 ± 2.8 | 89.7 ± 2.8 | 80.8 ± 6.3 | 77.1 ± 6.5 |
| 250 | 92.6 ± 1.8 | 90.4 ± 2.1 | 80.7 ± 4.8 | 77.4 ± 4.8 |
| 275 | 93.5 ± 1.8 | 91.6 ± 1.7 | 81.3 ± 4.7 | 78.0 ± 4.6 |
| 300 | 93.8 ± 1.8 | 91.9 ± 1.9 | 81.2 ± 4.8 | 77.8 ± 5.7 |
| 325 | 94.5 ± 1.6 | 92.6 ± 1.6 | 82.6 ± 3.9 | 78.5 ± 4.9 |
| 350 | 95.4 ± 1.0 | 93.6 ± 1.3 | 84.2 ± 3.9 | 80.2 ± 4.2 |
| 375 | 95.3 ± 1.2 | 93.2 ± 1.6 | 83.0 ± 4.5 | 78.8 ± 4.7 |
| 400 | 96.1 ± 1.0 | 94.5 ± 1.3 | 85.5 ± 3.4 | 80.6 ± 3.9 |