p22
done · peak unseen_mismatch 91.1
α schedule
α 0.9069 ×25 α 0.9342 ×25 α 0.9877 ×25 α 1.067 ×25 α 1.1718 ×25 α 1.302 ×25 α 1.4574 ×25 α 1.6378 ×25 α 1.8431 ×25 α 2.0734 ×25 α 2.3284 ×25 α 2.6081 ×25 α 2.9124 ×25 α 3.2414 ×25 α 3.5948 ×25 α 3.9727 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.9727
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 ± 1.6 | 49.7 ± 1.2 | 49.8 ± 1.5 | 49.5 ± 1.7 |
| 50 | 50.4 ± 1.5 | 50.0 ± 1.8 | 50.3 ± 1.5 | 49.5 ± 1.7 |
| 75 | 50.4 ± 1.2 | 49.5 ± 1.7 | 50.2 ± 1.5 | 49.8 ± 1.5 |
| 100 | 50.4 ± 1.9 | 50.2 ± 1.4 | 50.0 ± 1.2 | 49.9 ± 1.5 |
| 125 | 51.4 ± 2.3 | 49.1 ± 1.8 | 50.7 ± 2.6 | 49.2 ± 2.0 |
| 150 | 54.0 ± 9.8 | 48.8 ± 3.6 | 53.3 ± 9.6 | 48.5 ± 3.4 |
| 175 | 58.2 ± 15.2 | 45.5 ± 9.7 | 58.0 ± 15.2 | 46.1 ± 9.3 |
| 200 | 63.5 ± 19.7 | 45.8 ± 11.8 | 63.8 ± 20.1 | 46.4 ± 11.6 |
| 225 | 71.3 ± 22.3 | 47.9 ± 14.6 | 71.3 ± 22.3 | 48.7 ± 14.1 |
| 250 | 78.6 ± 22.3 | 51.1 ± 18.2 | 77.9 ± 22.8 | 51.1 ± 17.5 |
| 275 | 85.8 ± 19.2 | 57.7 ± 22.8 | 84.9 ± 19.7 | 57.3 ± 22.2 |
| 300 | 90.0 ± 18.4 | 64.3 ± 27.4 | 89.8 ± 18.6 | 63.5 ± 27.0 |
| 325 | 93.6 ± 14.2 | 72.3 ± 26.4 | 92.9 ± 15.4 | 71.3 ± 26.4 |
| 350 | 95.7 ± 10.6 | 77.7 ± 24.2 | 94.8 ± 11.8 | 76.7 ± 23.2 |
| 375 | 96.0 ± 9.9 | 86.9 ± 17.4 | 94.7 ± 10.2 | 84.7 ± 16.5 |
| 400 | 96.5 ± 7.8 | 94.4 ± 7.8 | 94.7 ± 9.1 | 91.1 ± 10.2 |