r3_c7
done · peak unseen_mismatch 95.0
α schedule
α 4.1789 ×25 α 4.1717 ×25 α 4.1509 ×25 α 4.1121 ×25 α 4.0516 ×25 α 3.966 ×25 α 3.8525 ×25 α 3.7082 ×25 α 3.5307 ×25 α 3.3176 ×25 α 3.0666 ×25 α 2.7755 ×25 α 2.4423 ×25 α 2.0649 ×25 α 1.6415 ×25 α 1.1702 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.1702
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 | 50.8 ± 1.2 | 50.5 ± 1.6 | 49.9 ± 1.5 | 49.5 ± 1.8 |
| 50 | 51.5 ± 1.5 | 50.5 ± 1.7 | 49.9 ± 1.0 | 49.9 ± 1.6 |
| 75 | 52.8 ± 2.2 | 50.2 ± 1.8 | 50.7 ± 1.7 | 49.3 ± 1.5 |
| 100 | 59.5 ± 7.8 | 57.0 ± 5.5 | 53.5 ± 4.7 | 51.1 ± 3.2 |
| 125 | 71.1 ± 10.9 | 68.1 ± 10.2 | 59.0 ± 9.9 | 56.9 ± 8.2 |
| 150 | 78.0 ± 6.4 | 75.9 ± 6.3 | 64.4 ± 7.9 | 61.6 ± 7.0 |
| 175 | 82.1 ± 4.4 | 80.3 ± 3.7 | 66.6 ± 8.1 | 63.9 ± 7.7 |
| 200 | 83.4 ± 3.3 | 81.3 ± 3.1 | 65.9 ± 6.1 | 63.8 ± 5.9 |
| 225 | 84.6 ± 2.7 | 82.7 ± 2.6 | 66.7 ± 5.4 | 64.5 ± 5.5 |
| 250 | 85.7 ± 2.5 | 84.2 ± 2.4 | 67.9 ± 4.6 | 66.2 ± 4.7 |
| 275 | 87.8 ± 2.5 | 86.3 ± 2.2 | 70.2 ± 5.4 | 68.2 ± 5.2 |
| 300 | 90.5 ± 2.4 | 88.9 ± 2.0 | 74.1 ± 5.7 | 71.6 ± 5.7 |
| 325 | 92.4 ± 1.8 | 90.5 ± 1.5 | 76.5 ± 5.0 | 74.0 ± 4.3 |
| 350 | 95.3 ± 1.4 | 93.7 ± 1.2 | 82.3 ± 4.6 | 79.3 ± 4.1 |
| 375 | 97.5 ± 0.9 | 96.4 ± 1.1 | 88.4 ± 3.4 | 84.9 ± 3.8 |
| 400 | 99.7 ± 0.3 | 99.2 ± 0.7 | 97.5 ± 1.7 | 95.0 ± 2.9 |