r1_c5
done · peak unseen_mismatch 88.8
α schedule
α 1.7439 ×25 α 1.8991 ×25 α 1.998 ×25 α 2.0751 ×25 α 2.1399 ×25 α 2.1965 ×25 α 2.2472 ×25 α 2.2934 ×25 α 2.336 ×25 α 2.3757 ×25 α 2.413 ×25 α 2.4482 ×25 α 2.4817 ×25 α 2.5135 ×25 α 2.544 ×25 α 2.5732 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.5732
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 ± 1.4 | 50.0 ± 1.5 | 50.0 ± 1.8 | 49.4 ± 1.9 |
| 50 | 50.7 ± 1.7 | 50.1 ± 1.5 | 49.8 ± 1.5 | 50.1 ± 1.2 |
| 75 | 50.7 ± 1.6 | 49.6 ± 1.9 | 50.2 ± 1.4 | 49.4 ± 1.5 |
| 100 | 51.0 ± 2.5 | 50.3 ± 1.7 | 50.5 ± 1.6 | 49.9 ± 1.5 |
| 125 | 53.5 ± 4.6 | 49.8 ± 1.6 | 51.8 ± 4.3 | 49.3 ± 1.6 |
| 150 | 60.4 ± 15.0 | 53.6 ± 7.1 | 58.4 ± 14.7 | 51.9 ± 5.8 |
| 175 | 71.9 ± 18.6 | 61.4 ± 11.9 | 69.7 ± 18.6 | 59.8 ± 11.0 |
| 200 | 79.2 ± 18.9 | 71.9 ± 16.8 | 77.1 ± 19.1 | 69.2 ± 16.5 |
| 225 | 86.5 ± 16.4 | 80.3 ± 15.6 | 83.8 ± 17.6 | 77.3 ± 15.4 |
| 250 | 92.8 ± 9.5 | 87.0 ± 10.9 | 89.3 ± 11.6 | 82.7 ± 12.2 |
| 275 | 96.3 ± 5.0 | 92.8 ± 5.6 | 92.2 ± 8.1 | 87.2 ± 7.9 |
| 300 | 97.0 ± 3.0 | 95.2 ± 3.2 | 92.6 ± 5.7 | 88.8 ± 5.9 |
| 325 | 97.2 ± 1.7 | 95.6 ± 2.3 | 91.7 ± 5.2 | 88.6 ± 6.1 |
| 350 | 96.8 ± 1.6 | 95.0 ± 1.8 | 90.9 ± 5.0 | 87.4 ± 5.1 |
| 375 | 96.1 ± 1.4 | 94.3 ± 2.0 | 88.7 ± 4.4 | 85.6 ± 4.7 |
| 400 | 95.8 ± 1.5 | 93.9 ± 1.8 | 87.7 ± 4.6 | 84.7 ± 5.4 |