p10
done · peak unseen_mismatch 93.1
α schedule
α 1.4187 ×25 α 1.4833 ×25 α 1.5508 ×25 α 1.6201 ×25 α 1.6905 ×25 α 1.762 ×25 α 1.8342 ×25 α 1.9071 ×25 α 1.9806 ×25 α 2.0547 ×25 α 2.1292 ×25 α 2.2041 ×25 α 2.2795 ×25 α 2.3552 ×25 α 2.4313 ×25 α 2.5076 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.5076
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 | 50.8 ± 1.4 | 49.8 ± 1.5 | 49.7 ± 1.4 | 49.5 ± 1.5 |
| 50 | 50.5 ± 1.9 | 50.0 ± 1.6 | 49.9 ± 1.2 | 49.9 ± 1.4 |
| 75 | 50.7 ± 1.5 | 49.6 ± 1.7 | 50.1 ± 1.4 | 49.4 ± 1.6 |
| 100 | 50.7 ± 2.3 | 50.1 ± 1.5 | 50.4 ± 1.7 | 50.1 ± 1.6 |
| 125 | 52.4 ± 3.9 | 49.2 ± 2.4 | 51.6 ± 3.9 | 49.0 ± 2.5 |
| 150 | 56.8 ± 13.9 | 48.4 ± 7.1 | 55.6 ± 14.2 | 47.9 ± 7.0 |
| 175 | 62.1 ± 16.5 | 47.6 ± 9.2 | 61.6 ± 16.6 | 47.2 ± 8.7 |
| 200 | 74.4 ± 21.0 | 56.4 ± 13.9 | 73.2 ± 21.0 | 55.6 ± 13.7 |
| 225 | 81.0 ± 21.3 | 61.1 ± 18.6 | 80.1 ± 21.6 | 60.9 ± 17.9 |
| 250 | 86.3 ± 18.7 | 73.7 ± 18.6 | 85.0 ± 19.7 | 72.0 ± 18.0 |
| 275 | 92.8 ± 14.3 | 80.9 ± 18.2 | 91.7 ± 15.3 | 79.8 ± 18.0 |
| 300 | 95.1 ± 12.1 | 87.5 ± 15.1 | 94.0 ± 12.8 | 85.1 ± 15.6 |
| 325 | 97.9 ± 5.3 | 92.0 ± 10.8 | 96.9 ± 6.3 | 90.5 ± 10.6 |
| 350 | 99.0 ± 1.2 | 93.3 ± 12.1 | 97.8 ± 2.8 | 91.2 ± 11.7 |
| 375 | 98.6 ± 1.4 | 95.3 ± 8.7 | 96.4 ± 3.7 | 92.0 ± 8.7 |
| 400 | 98.2 ± 1.4 | 96.8 ± 1.7 | 95.7 ± 3.8 | 93.1 ± 3.8 |