lin_up
done · peak unseen_mismatch 90.8
α schedule
α 0.1562 ×25 α 0.4688 ×25 α 0.7812 ×25 α 1.0938 ×25 α 1.4062 ×25 α 1.7188 ×25 α 2.0312 ×25 α 2.3438 ×25 α 2.6562 ×25 α 2.9688 ×25 α 3.2812 ×25 α 3.5938 ×25 α 3.9062 ×25 α 4.2188 ×25 α 4.5312 ×25 α 4.8438 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.8438
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 | 50.6 ± 1.9 | 49.5 ± 1.9 | 50.0 ± 2.0 | 50.1 ± 1.7 |
| 50 | 50.6 ± 1.7 | 50.0 ± 1.5 | 50.1 ± 1.5 | 50.2 ± 1.8 |
| 75 | 50.8 ± 1.7 | 49.7 ± 1.8 | 50.0 ± 1.9 | 49.2 ± 1.4 |
| 100 | 51.0 ± 2.0 | 49.9 ± 1.4 | 50.4 ± 1.7 | 50.0 ± 1.5 |
| 125 | 51.6 ± 3.3 | 48.8 ± 2.3 | 51.8 ± 3.5 | 49.4 ± 2.2 |
| 150 | 56.5 ± 10.6 | 47.0 ± 6.0 | 56.0 ± 10.6 | 47.2 ± 5.9 |
| 175 | 65.3 ± 18.3 | 44.0 ± 8.4 | 64.6 ± 18.2 | 43.9 ± 8.9 |
| 200 | 74.0 ± 21.5 | 44.3 ± 10.6 | 74.1 ± 21.2 | 44.5 ± 10.5 |
| 225 | 85.5 ± 19.3 | 49.2 ± 18.8 | 85.3 ± 19.1 | 48.8 ± 18.5 |
| 250 | 91.9 ± 16.3 | 59.8 ± 20.6 | 91.8 ± 16.2 | 59.1 ± 20.3 |
| 275 | 93.9 ± 14.6 | 68.7 ± 20.1 | 93.2 ± 15.6 | 67.5 ± 19.3 |
| 300 | 95.1 ± 11.6 | 84.3 ± 16.2 | 94.0 ± 13.2 | 82.5 ± 15.9 |
| 325 | 96.1 ± 9.4 | 91.6 ± 10.3 | 94.8 ± 10.6 | 89.2 ± 10.7 |
| 350 | 96.9 ± 4.7 | 93.8 ± 5.6 | 95.3 ± 6.1 | 90.8 ± 6.9 |
| 375 | 96.8 ± 2.9 | 94.4 ± 3.4 | 94.0 ± 5.0 | 90.7 ± 5.3 |
| 400 | 95.6 ± 2.9 | 93.5 ± 3.3 | 91.8 ± 5.1 | 88.2 ± 5.9 |