r4_c3
done · peak unseen_mismatch 87.8
α schedule
α 0.7126 ×25 α 0.8033 ×25 α 0.9098 ×25 α 1.0265 ×25 α 1.1509 ×25 α 1.2818 ×25 α 1.418 ×25 α 1.559 ×25 α 1.7043 ×25 α 1.8535 ×25 α 2.0062 ×25 α 2.1622 ×25 α 2.3212 ×25 α 2.4831 ×25 α 2.6477 ×25 α 2.8148 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.8148
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 ± 1.3 | 50.1 ± 1.4 | 49.8 ± 1.5 | 49.3 ± 1.7 |
| 50 | 50.4 ± 1.7 | 50.0 ± 1.6 | 50.2 ± 1.4 | 49.7 ± 1.7 |
| 75 | 50.5 ± 1.4 | 49.4 ± 1.9 | 50.2 ± 1.4 | 49.4 ± 1.8 |
| 100 | 50.2 ± 1.8 | 50.0 ± 1.6 | 50.1 ± 1.3 | 49.8 ± 1.5 |
| 125 | 51.2 ± 1.5 | 49.5 ± 1.5 | 50.3 ± 2.0 | 49.0 ± 1.9 |
| 150 | 52.6 ± 7.6 | 49.5 ± 1.4 | 51.5 ± 7.1 | 49.3 ± 1.4 |
| 175 | 57.6 ± 13.0 | 46.0 ± 6.0 | 57.7 ± 12.7 | 46.9 ± 6.2 |
| 200 | 65.2 ± 20.2 | 43.4 ± 12.3 | 65.2 ± 20.1 | 44.2 ± 12.7 |
| 225 | 71.1 ± 22.8 | 42.8 ± 14.8 | 71.4 ± 22.5 | 43.9 ± 15.3 |
| 250 | 76.0 ± 21.9 | 43.3 ± 14.1 | 75.1 ± 22.5 | 43.5 ± 14.0 |
| 275 | 83.9 ± 20.0 | 51.7 ± 19.9 | 83.2 ± 20.5 | 51.4 ± 19.4 |
| 300 | 91.0 ± 17.8 | 50.9 ± 26.8 | 90.8 ± 17.8 | 51.0 ± 26.4 |
| 325 | 93.2 ± 15.5 | 62.2 ± 26.0 | 92.8 ± 16.0 | 62.3 ± 25.3 |
| 350 | 95.7 ± 13.1 | 71.6 ± 24.8 | 95.5 ± 13.3 | 71.0 ± 24.1 |
| 375 | 96.4 ± 12.0 | 81.9 ± 23.2 | 96.1 ± 11.9 | 81.3 ± 22.6 |
| 400 | 97.3 ± 8.9 | 88.5 ± 16.0 | 96.9 ± 9.4 | 87.8 ± 15.9 |