late_fast
done · peak unseen_mismatch 84.7
α schedule
α 4.9999 ×25 α 4.997 ×25 α 4.9859 ×25 α 4.9613 ×25 α 4.9177 ×25 α 4.8497 ×25 α 4.7519 ×25 α 4.6189 ×25 α 4.4452 ×25 α 4.2255 ×25 α 3.9543 ×25 α 3.6262 ×25 α 3.2357 ×25 α 2.7775 ×25 α 2.2461 ×25 α 1.6361 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.6361
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 | 51.3 ± 2.0 | 50.0 ± 1.6 | 50.1 ± 2.1 | 50.3 ± 1.5 |
| 50 | 51.9 ± 1.9 | 50.7 ± 1.2 | 50.5 ± 2.0 | 49.8 ± 1.9 |
| 75 | 53.4 ± 2.6 | 51.1 ± 2.1 | 50.5 ± 2.5 | 49.1 ± 1.6 |
| 100 | 64.5 ± 8.0 | 60.3 ± 7.1 | 55.6 ± 5.7 | 52.7 ± 3.6 |
| 125 | 76.8 ± 8.2 | 73.0 ± 7.5 | 62.3 ± 9.1 | 59.2 ± 6.8 |
| 150 | 79.5 ± 3.9 | 76.2 ± 4.1 | 62.6 ± 5.6 | 60.4 ± 4.5 |
| 175 | 79.5 ± 3.3 | 77.1 ± 3.6 | 62.6 ± 5.8 | 60.8 ± 4.7 |
| 200 | 80.5 ± 2.6 | 78.6 ± 2.5 | 62.9 ± 4.6 | 61.1 ± 3.7 |
| 225 | 81.4 ± 2.6 | 79.4 ± 2.2 | 63.4 ± 4.3 | 61.8 ± 3.6 |
| 250 | 82.5 ± 2.6 | 80.9 ± 2.6 | 64.7 ± 4.2 | 63.0 ± 3.9 |
| 275 | 84.0 ± 2.9 | 82.6 ± 2.7 | 66.6 ± 5.1 | 64.4 ± 4.8 |
| 300 | 85.4 ± 2.0 | 83.9 ± 2.1 | 67.9 ± 3.6 | 65.8 ± 4.2 |
| 325 | 88.0 ± 2.3 | 86.3 ± 2.3 | 71.3 ± 5.4 | 68.3 ± 5.4 |
| 350 | 90.4 ± 2.3 | 88.8 ± 2.1 | 73.4 ± 5.8 | 70.3 ± 5.4 |
| 375 | 93.4 ± 1.9 | 91.7 ± 1.9 | 78.8 ± 5.6 | 74.8 ± 5.6 |
| 400 | 97.7 ± 1.3 | 96.3 ± 1.3 | 89.4 ± 4.2 | 84.7 ± 5.7 |