r2_c7
done · peak unseen_mismatch 93.4
α schedule
α 1.5059 ×25 α 1.5148 ×25 α 1.5321 ×25 α 1.5575 ×25 α 1.5911 ×25 α 1.6327 ×25 α 1.6822 ×25 α 1.7396 ×25 α 1.8048 ×25 α 1.8779 ×25 α 1.9587 ×25 α 2.0472 ×25 α 2.1435 ×25 α 2.2474 ×25 α 2.359 ×25 α 2.4782 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.4782
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 | 50.6 ± 1.4 | 50.2 ± 1.7 | 49.8 ± 1.2 | 49.4 ± 1.8 |
| 50 | 50.5 ± 1.9 | 50.1 ± 1.6 | 50.0 ± 1.6 | 49.7 ± 1.7 |
| 75 | 50.7 ± 1.3 | 49.5 ± 1.6 | 50.3 ± 1.3 | 49.5 ± 1.7 |
| 100 | 50.4 ± 2.0 | 50.1 ± 1.6 | 50.1 ± 1.3 | 50.3 ± 1.4 |
| 125 | 51.7 ± 2.1 | 49.4 ± 1.5 | 50.8 ± 2.5 | 49.2 ± 1.9 |
| 150 | 56.4 ± 12.0 | 48.2 ± 6.3 | 55.5 ± 11.8 | 47.3 ± 6.9 |
| 175 | 63.6 ± 18.8 | 47.6 ± 8.5 | 63.0 ± 18.5 | 47.4 ± 8.4 |
| 200 | 72.9 ± 21.8 | 52.2 ± 12.4 | 72.5 ± 21.8 | 51.2 ± 11.9 |
| 225 | 78.5 ± 23.7 | 58.3 ± 14.7 | 78.2 ± 23.8 | 58.1 ± 14.3 |
| 250 | 82.3 ± 23.0 | 60.2 ± 19.9 | 82.3 ± 22.8 | 59.5 ± 20.4 |
| 275 | 87.5 ± 19.4 | 72.5 ± 21.1 | 86.7 ± 20.0 | 71.5 ± 20.8 |
| 300 | 91.3 ± 16.0 | 79.6 ± 18.9 | 90.8 ± 16.2 | 78.5 ± 18.3 |
| 325 | 96.1 ± 10.2 | 88.6 ± 13.7 | 95.4 ± 10.9 | 87.3 ± 13.9 |
| 350 | 97.9 ± 7.4 | 91.1 ± 13.0 | 96.9 ± 8.3 | 89.1 ± 13.3 |
| 375 | 98.8 ± 1.8 | 95.1 ± 8.3 | 97.7 ± 2.9 | 93.0 ± 8.3 |
| 400 | 98.7 ± 1.3 | 96.3 ± 4.5 | 97.2 ± 2.8 | 93.4 ± 5.2 |