p00
done · peak unseen_mismatch 75.1
α schedule
α 3.2472 ×25 α 3.1691 ×25 α 3.1198 ×25 α 3.0817 ×25 α 3.0498 ×25 α 3.0221 ×25 α 2.9973 ×25 α 2.9748 ×25 α 2.954 ×25 α 2.9347 ×25 α 2.9166 ×25 α 2.8996 ×25 α 2.8834 ×25 α 2.868 ×25 α 2.8533 ×25 α 2.8393 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.8393
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 | 50.2 ± 1.7 | 49.9 ± 1.6 | 49.6 ± 1.6 |
| 50 | 51.7 ± 1.6 | 50.4 ± 1.7 | 50.1 ± 1.3 | 49.8 ± 1.9 |
| 75 | 52.3 ± 1.7 | 50.1 ± 1.6 | 50.6 ± 1.6 | 49.5 ± 1.7 |
| 100 | 55.8 ± 7.1 | 53.3 ± 3.4 | 52.4 ± 4.6 | 50.0 ± 2.2 |
| 125 | 66.6 ± 12.8 | 62.0 ± 10.8 | 58.4 ± 11.1 | 55.7 ± 8.7 |
| 150 | 76.6 ± 12.0 | 73.7 ± 11.5 | 66.7 ± 12.4 | 63.4 ± 11.4 |
| 175 | 84.8 ± 7.0 | 82.2 ± 6.9 | 72.6 ± 11.2 | 69.1 ± 10.1 |
| 200 | 88.3 ± 4.2 | 85.6 ± 4.4 | 74.6 ± 9.2 | 72.1 ± 9.1 |
| 225 | 89.8 ± 3.5 | 87.4 ± 3.3 | 75.4 ± 7.2 | 72.6 ± 7.6 |
| 250 | 90.3 ± 2.1 | 87.7 ± 1.9 | 74.7 ± 5.5 | 72.0 ± 5.2 |
| 275 | 90.9 ± 2.4 | 88.6 ± 2.0 | 76.2 ± 5.0 | 73.2 ± 4.8 |
| 300 | 91.5 ± 1.8 | 89.8 ± 2.1 | 77.1 ± 4.4 | 74.1 ± 4.2 |
| 325 | 91.8 ± 1.7 | 89.8 ± 1.7 | 77.0 ± 4.6 | 73.7 ± 4.8 |
| 350 | 92.7 ± 1.4 | 90.9 ± 1.8 | 78.0 ± 4.2 | 75.1 ± 4.7 |
| 375 | 92.4 ± 2.0 | 90.6 ± 2.0 | 77.1 ± 4.9 | 73.7 ± 4.9 |
| 400 | 92.8 ± 1.5 | 90.9 ± 1.8 | 78.6 ± 3.9 | 74.4 ± 4.3 |