r5_c7
done · peak unseen_mismatch 69.3
α schedule
α 2.6421 ×25 α 2.9758 ×25 α 3.1662 ×25 α 3.3062 ×25 α 3.4189 ×25 α 3.5143 ×25 α 3.5976 ×25 α 3.6719 ×25 α 3.7391 ×25 α 3.8007 ×25 α 3.8577 ×25 α 3.9107 ×25 α 3.9605 ×25 α 4.0074 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.0074
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 ± 1.3 | 50.5 ± 1.5 | 50.2 ± 1.4 | 49.3 ± 1.5 |
| 50 | 51.2 ± 1.4 | 50.3 ± 1.6 | 49.7 ± 1.1 | 50.1 ± 1.6 |
| 75 | 51.5 ± 1.5 | 49.9 ± 1.8 | 50.1 ± 1.5 | 49.4 ± 1.7 |
| 100 | 54.5 ± 5.8 | 51.9 ± 2.6 | 52.1 ± 4.1 | 50.0 ± 1.9 |
| 125 | 64.8 ± 11.4 | 59.7 ± 9.4 | 57.5 ± 10.3 | 53.6 ± 6.3 |
| 150 | 74.4 ± 13.3 | 71.2 ± 12.3 | 65.6 ± 12.9 | 62.2 ± 11.2 |
| 175 | 81.3 ± 10.0 | 78.6 ± 9.5 | 69.4 ± 12.1 | 66.2 ± 11.5 |
| 200 | 85.4 ± 5.6 | 82.6 ± 5.2 | 72.3 ± 9.8 | 68.9 ± 9.3 |
| 225 | 86.3 ± 4.4 | 83.9 ± 4.1 | 71.7 ± 8.0 | 68.9 ± 7.9 |
| 250 | 86.3 ± 3.2 | 84.1 ± 2.7 | 71.1 ± 6.9 | 68.3 ± 6.2 |
| 275 | 86.7 ± 3.2 | 84.4 ± 2.8 | 70.6 ± 5.5 | 67.4 ± 5.2 |
| 300 | 87.0 ± 2.7 | 84.9 ± 2.1 | 70.4 ± 4.6 | 68.0 ± 4.3 |
| 325 | 87.3 ± 2.8 | 85.4 ± 2.0 | 70.6 ± 5.3 | 68.3 ± 5.0 |
| 350 | 88.2 ± 2.6 | 86.2 ± 1.8 | 71.6 ± 5.4 | 69.3 ± 4.9 |