r1_c7
done · peak unseen_mismatch 70.8
α schedule
α 2.6295 ×25 α 2.9113 ×25 α 3.0733 ×25 α 3.1927 ×25 α 3.2892 ×25 α 3.371 ×25 α 3.4425 ×25 α 3.5063 ×25 α 3.5642 ×25 α 3.6173 ×25 α 3.6664 ×25 α 3.7122 ×25 α 3.7552 ×25 α 3.7957 ×25 α 3.8341 ×25 α 3.8706 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.8706
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 | 50.7 ± 1.6 | 50.1 ± 1.5 | 49.5 ± 1.6 |
| 50 | 51.3 ± 1.5 | 50.2 ± 1.6 | 49.8 ± 1.3 | 50.0 ± 1.8 |
| 75 | 51.7 ± 1.6 | 49.8 ± 1.8 | 50.0 ± 1.6 | 49.4 ± 1.6 |
| 100 | 53.8 ± 5.3 | 51.6 ± 2.6 | 51.8 ± 4.3 | 49.7 ± 1.8 |
| 125 | 64.9 ± 12.4 | 59.2 ± 9.2 | 58.4 ± 11.0 | 53.7 ± 7.2 |
| 150 | 73.4 ± 14.1 | 70.0 ± 12.9 | 66.0 ± 13.0 | 62.2 ± 11.3 |
| 175 | 80.8 ± 9.7 | 77.9 ± 9.3 | 69.6 ± 11.9 | 66.3 ± 11.5 |
| 200 | 86.1 ± 5.6 | 83.5 ± 5.4 | 73.5 ± 10.1 | 70.4 ± 9.1 |
| 225 | 87.5 ± 4.2 | 85.0 ± 4.2 | 73.7 ± 8.4 | 70.8 ± 8.5 |
| 250 | 87.3 ± 3.4 | 85.2 ± 3.0 | 72.9 ± 7.0 | 70.3 ± 6.3 |
| 275 | 87.8 ± 3.1 | 85.5 ± 2.4 | 71.8 ± 6.2 | 69.1 ± 6.0 |
| 300 | 87.8 ± 2.9 | 86.1 ± 2.2 | 72.2 ± 5.6 | 69.7 ± 4.8 |
| 325 | 88.0 ± 2.3 | 85.9 ± 2.0 | 71.9 ± 4.7 | 69.2 ± 4.5 |
| 350 | 88.4 ± 2.3 | 86.9 ± 2.2 | 72.5 ± 4.7 | 69.7 ± 4.8 |
| 375 | 88.8 ± 2.4 | 86.7 ± 2.0 | 71.9 ± 4.2 | 69.4 ± 4.3 |
| 400 | 88.9 ± 2.3 | 87.3 ± 1.8 | 72.5 ± 4.0 | 70.3 ± 4.3 |