r4_c5
done · peak unseen_mismatch 91.8
α schedule
α 1.4165 ×25 α 1.4214 ×25 α 1.4346 ×25 α 1.4585 ×25 α 1.4948 ×25 α 1.545 ×25 α 1.6106 ×25 α 1.6927 ×25 α 1.7926 ×25 α 1.9112 ×25 α 2.0496 ×25 α 2.2087 ×25 α 2.3894 ×25 α 2.5925 ×25 α 2.8189 ×25 α 3.0694 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.0694
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 | 50.7 ± 1.4 | 49.8 ± 1.6 | 49.7 ± 1.5 | 49.6 ± 1.6 |
| 50 | 50.8 ± 1.9 | 49.9 ± 1.5 | 49.7 ± 1.4 | 49.9 ± 1.5 |
| 75 | 50.5 ± 1.4 | 49.7 ± 1.6 | 50.2 ± 1.5 | 49.4 ± 1.8 |
| 100 | 50.7 ± 2.5 | 50.1 ± 1.3 | 50.3 ± 1.5 | 50.0 ± 1.7 |
| 125 | 52.1 ± 2.4 | 49.3 ± 1.9 | 51.1 ± 3.0 | 49.2 ± 2.3 |
| 150 | 57.2 ± 13.4 | 49.6 ± 5.8 | 56.1 ± 13.1 | 48.8 ± 6.1 |
| 175 | 62.3 ± 17.6 | 46.7 ± 8.6 | 62.1 ± 17.6 | 46.7 ± 8.2 |
| 200 | 73.1 ± 22.3 | 47.7 ± 14.2 | 72.7 ± 22.6 | 47.1 ± 14.2 |
| 225 | 80.2 ± 21.5 | 51.8 ± 19.8 | 79.9 ± 21.7 | 51.8 ± 19.0 |
| 250 | 88.2 ± 19.1 | 58.8 ± 21.8 | 87.4 ± 19.6 | 57.8 ± 21.4 |
| 275 | 93.4 ± 14.4 | 74.1 ± 20.6 | 92.8 ± 15.4 | 73.3 ± 20.5 |
| 300 | 95.9 ± 12.9 | 82.9 ± 19.4 | 95.6 ± 13.4 | 82.5 ± 19.0 |
| 325 | 96.9 ± 10.2 | 88.7 ± 15.1 | 96.1 ± 11.8 | 87.6 ± 15.3 |
| 350 | 97.6 ± 8.1 | 87.2 ± 18.0 | 96.7 ± 8.6 | 85.6 ± 17.2 |
| 375 | 98.3 ± 3.9 | 92.4 ± 13.6 | 96.7 ± 5.1 | 90.0 ± 12.8 |
| 400 | 98.2 ± 2.1 | 95.0 ± 7.9 | 96.1 ± 3.8 | 91.8 ± 7.5 |