lin_down
done · peak unseen_mismatch 97.2
α schedule
α 4.8438 ×25 α 4.5312 ×25 α 4.2188 ×25 α 3.9062 ×25 α 3.5938 ×25 α 3.2812 ×25 α 2.9688 ×25 α 2.6562 ×25 α 2.3438 ×25 α 2.0312 ×25 α 1.7188 ×25 α 1.4062 ×25 α 1.0938 ×25 α 0.7812 ×25 α 0.4688 ×25 α 0.1562 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.1562
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 | 51.4 ± 1.9 | 50.0 ± 1.6 | 50.1 ± 2.1 | 50.1 ± 1.6 |
| 50 | 51.9 ± 2.0 | 50.7 ± 1.3 | 50.4 ± 2.0 | 50.0 ± 1.9 |
| 75 | 53.5 ± 2.7 | 51.1 ± 2.3 | 50.6 ± 2.4 | 49.0 ± 1.4 |
| 100 | 63.4 ± 7.9 | 59.0 ± 6.6 | 55.4 ± 6.2 | 52.6 ± 3.5 |
| 125 | 76.8 ± 9.6 | 72.7 ± 8.8 | 64.7 ± 10.4 | 60.3 ± 8.6 |
| 150 | 83.0 ± 4.5 | 79.6 ± 4.9 | 68.2 ± 7.0 | 64.5 ± 6.0 |
| 175 | 84.9 ± 3.9 | 82.3 ± 4.0 | 69.2 ± 7.0 | 66.0 ± 5.7 |
| 200 | 88.2 ± 2.8 | 85.9 ± 2.6 | 72.3 ± 5.8 | 68.8 ± 4.5 |
| 225 | 91.4 ± 2.4 | 89.1 ± 2.1 | 77.4 ± 5.0 | 74.0 ± 4.8 |
| 250 | 94.4 ± 1.6 | 92.6 ± 1.6 | 81.5 ± 5.5 | 78.6 ± 4.8 |
| 275 | 96.7 ± 1.1 | 95.4 ± 1.3 | 87.3 ± 4.4 | 83.7 ± 4.6 |
| 300 | 98.7 ± 0.9 | 97.6 ± 1.1 | 92.5 ± 3.2 | 89.1 ± 3.7 |
| 325 | 99.8 ± 0.3 | 99.2 ± 0.6 | 97.8 ± 1.5 | 95.1 ± 3.7 |
| 350 | 100.0 ± 0.0 | 99.5 ± 2.1 | 99.5 ± 0.6 | 97.2 ± 5.5 |
| 375 | 100.0 ± 0.0 | 99.0 ± 2.9 | 99.9 ± 0.1 | 96.9 ± 6.5 |
| 400 | 100.0 ± 0.0 | 87.8 ± 16.0 | 100.0 ± 0.0 | 84.0 ± 17.4 |