control
done · peak unseen_mismatch 84.5
α schedule
α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25 α 1.6 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.6
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 | 50.4 ± 1.5 | 50.0 ± 1.6 | 49.9 ± 1.6 | 49.4 ± 1.8 |
| 50 | 50.6 ± 1.3 | 50.2 ± 1.6 | 50.0 ± 1.4 | 49.8 ± 1.4 |
| 75 | 50.7 ± 1.7 | 49.6 ± 1.7 | 50.4 ± 1.4 | 49.5 ± 1.8 |
| 100 | 50.8 ± 2.7 | 50.1 ± 1.5 | 50.4 ± 1.7 | 49.9 ± 1.3 |
| 125 | 52.2 ± 2.5 | 49.2 ± 2.1 | 50.9 ± 2.8 | 49.0 ± 2.3 |
| 150 | 56.9 ± 11.6 | 49.4 ± 6.0 | 55.5 ± 11.2 | 48.6 ± 5.7 |
| 175 | 65.4 ± 18.5 | 46.9 ± 11.2 | 64.5 ± 18.3 | 47.1 ± 11.2 |
| 200 | 75.9 ± 21.5 | 50.6 ± 14.4 | 75.5 ± 21.0 | 50.2 ± 14.0 |
| 225 | 83.6 ± 22.9 | 52.4 ± 19.5 | 83.6 ± 22.5 | 52.9 ± 18.7 |
| 250 | 85.2 ± 21.5 | 60.1 ± 25.4 | 84.9 ± 22.0 | 59.7 ± 24.9 |
| 275 | 89.9 ± 17.9 | 64.1 ± 24.3 | 89.3 ± 18.5 | 63.4 ± 23.1 |
| 300 | 95.6 ± 12.9 | 68.8 ± 24.7 | 95.4 ± 12.6 | 68.0 ± 23.9 |
| 325 | 97.6 ± 9.5 | 71.3 ± 25.3 | 97.7 ± 9.0 | 70.9 ± 24.2 |
| 350 | 99.3 ± 3.1 | 78.4 ± 28.6 | 99.0 ± 4.1 | 77.0 ± 27.9 |
| 375 | 100.0 ± 0.1 | 82.3 ± 27.4 | 99.7 ± 0.8 | 81.3 ± 26.6 |
| 400 | 99.9 ± 0.4 | 85.1 ± 25.8 | 99.6 ± 1.0 | 84.5 ± 24.9 |