fixed_16
done · peak unseen_mismatch 86.6
α schedule
α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.6
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 ± 1.1 | 50.1 ± 1.2 | 49.8 ± 1.9 | 50.4 ± 1.7 |
| 50 | 50.5 ± 1.3 | 50.2 ± 1.5 | 50.1 ± 1.6 | 49.4 ± 1.4 |
| 75 | 50.8 ± 1.5 | 49.4 ± 1.3 | 50.3 ± 1.7 | 49.9 ± 1.6 |
| 100 | 51.5 ± 1.9 | 49.9 ± 1.5 | 50.7 ± 2.0 | 49.9 ± 1.2 |
| 125 | 54.2 ± 8.8 | 49.8 ± 1.9 | 53.0 ± 8.4 | 49.1 ± 1.9 |
| 150 | 58.9 ± 14.5 | 50.0 ± 6.5 | 57.3 ± 14.9 | 50.0 ± 6.3 |
| 175 | 67.7 ± 18.7 | 52.1 ± 8.0 | 66.1 ± 19.0 | 51.7 ± 6.6 |
| 200 | 80.4 ± 19.8 | 60.1 ± 12.4 | 79.8 ± 20.1 | 58.1 ± 12.0 |
| 225 | 88.6 ± 19.1 | 63.5 ± 17.4 | 88.3 ± 19.3 | 62.5 ± 17.6 |
| 250 | 91.3 ± 17.9 | 68.3 ± 20.4 | 91.4 ± 17.5 | 67.4 ± 20.0 |
| 275 | 94.6 ± 13.6 | 77.5 ± 20.2 | 94.0 ± 14.5 | 77.0 ± 19.5 |
| 300 | 96.1 ± 13.3 | 82.1 ± 19.6 | 95.8 ± 13.4 | 81.7 ± 18.8 |
| 325 | 96.8 ± 10.7 | 81.9 ± 20.1 | 96.6 ± 11.2 | 81.2 ± 20.3 |
| 350 | 98.1 ± 7.8 | 85.7 ± 18.8 | 97.6 ± 8.8 | 84.7 ± 18.7 |
| 375 | 99.6 ± 1.3 | 86.2 ± 22.7 | 99.0 ± 2.4 | 84.9 ± 22.2 |
| 400 | 99.8 ± 0.4 | 87.8 ± 25.2 | 99.6 ± 0.7 | 86.6 ± 24.7 |