r5_c4
done · peak unseen_mismatch 87.9
α schedule
α 1.5731 ×25 α 1.6048 ×25 α 1.6546 ×25 α 1.7191 ×25 α 1.7967 ×25 α 1.8863 ×25 α 1.9869 ×25 α 2.0979 ×25 α 2.2188 ×25 α 2.3492 ×25 α 2.4887 ×25 α 2.6369 ×25 α 2.7936 ×25 α 2.9585 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.9585
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 | 50.4 ± 1.7 | 50.0 ± 1.4 | 50.0 ± 1.7 | 49.4 ± 1.9 |
| 50 | 50.5 ± 1.6 | 50.1 ± 1.8 | 50.2 ± 1.5 | 49.8 ± 1.5 |
| 75 | 50.4 ± 1.4 | 49.6 ± 1.6 | 50.3 ± 1.5 | 49.4 ± 1.5 |
| 100 | 50.5 ± 2.0 | 50.5 ± 1.7 | 50.2 ± 1.5 | 49.9 ± 1.3 |
| 125 | 52.4 ± 3.7 | 49.6 ± 1.6 | 51.4 ± 4.0 | 49.2 ± 2.2 |
| 150 | 56.7 ± 12.4 | 50.3 ± 5.6 | 55.3 ± 12.1 | 49.4 ± 5.9 |
| 175 | 63.5 ± 17.0 | 52.4 ± 9.6 | 62.3 ± 16.8 | 51.4 ± 9.7 |
| 200 | 75.2 ± 20.2 | 62.3 ± 19.8 | 74.2 ± 19.9 | 60.7 ± 19.6 |
| 225 | 82.4 ± 21.8 | 68.4 ± 22.0 | 81.5 ± 21.9 | 67.2 ± 20.8 |
| 250 | 86.1 ± 19.9 | 77.4 ± 20.1 | 84.7 ± 20.6 | 75.3 ± 19.3 |
| 275 | 90.5 ± 16.0 | 84.8 ± 16.8 | 87.9 ± 17.4 | 82.6 ± 17.0 |
| 300 | 92.7 ± 13.2 | 90.5 ± 12.6 | 89.2 ± 14.8 | 86.9 ± 13.8 |
| 325 | 94.1 ± 9.1 | 91.8 ± 9.4 | 90.0 ± 11.6 | 87.2 ± 10.9 |
| 350 | 95.0 ± 6.2 | 93.4 ± 6.1 | 90.3 ± 8.3 | 87.9 ± 8.4 |