p02
done · peak unseen_mismatch 78.2
α schedule
α 2.133 ×25 α 2.2502 ×25 α 2.367 ×25 α 2.4835 ×25 α 2.5999 ×25 α 2.7162 ×25 α 2.8324 ×25 α 2.9484 ×25 α 3.0644 ×25 α 3.1804 ×25 α 3.2963 ×25 α 3.4121 ×25 α 3.5278 ×25 α 3.6436 ×25 α 3.7593 ×25 α 3.8749 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.8749
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 | 50.9 ± 1.4 | 50.1 ± 1.8 | 50.0 ± 1.6 | 49.5 ± 1.8 |
| 50 | 51.0 ± 1.5 | 50.3 ± 1.6 | 49.8 ± 1.4 | 50.0 ± 1.6 |
| 75 | 51.1 ± 1.6 | 49.6 ± 1.6 | 50.3 ± 1.4 | 49.6 ± 1.5 |
| 100 | 52.1 ± 3.4 | 50.6 ± 1.4 | 50.9 ± 2.1 | 49.9 ± 1.6 |
| 125 | 58.4 ± 10.0 | 52.5 ± 5.2 | 54.7 ± 8.7 | 50.2 ± 3.6 |
| 150 | 66.8 ± 16.3 | 61.8 ± 12.5 | 63.5 ± 16.0 | 57.2 ± 11.0 |
| 175 | 78.0 ± 17.0 | 72.6 ± 16.0 | 72.5 ± 18.4 | 67.0 ± 15.1 |
| 200 | 83.5 ± 13.8 | 80.1 ± 13.7 | 75.7 ± 16.3 | 71.7 ± 14.9 |
| 225 | 89.6 ± 7.6 | 86.5 ± 7.5 | 81.3 ± 11.0 | 77.4 ± 10.4 |
| 250 | 91.3 ± 4.6 | 89.2 ± 4.4 | 81.2 ± 8.9 | 77.7 ± 8.0 |
| 275 | 92.0 ± 3.7 | 89.9 ± 3.6 | 81.6 ± 8.1 | 78.2 ± 7.6 |
| 300 | 91.8 ± 3.3 | 89.5 ± 3.0 | 79.9 ± 7.8 | 76.4 ± 7.3 |
| 325 | 90.8 ± 2.7 | 88.7 ± 2.6 | 78.3 ± 6.9 | 75.0 ± 6.3 |
| 350 | 90.7 ± 2.5 | 88.5 ± 2.3 | 77.8 ± 5.5 | 74.1 ± 5.3 |
| 375 | 90.5 ± 2.1 | 88.2 ± 2.0 | 76.5 ± 4.8 | 72.8 ± 4.2 |
| 400 | 90.1 ± 1.8 | 87.8 ± 1.6 | 75.4 ± 4.2 | 72.0 ± 3.0 |