p07
done · peak unseen_mismatch 91.1
α schedule
α 2.7685 ×25 α 2.607 ×25 α 2.4896 ×25 α 2.3913 ×25 α 2.3045 ×25 α 2.2257 ×25 α 2.1528 ×25 α 2.0846 ×25 α 2.0202 ×25 α 1.959 ×25 α 1.9005 ×25 α 1.8443 ×25 α 1.7902 ×25 α 1.7379 ×25 α 1.6872 ×25 α 1.6381 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.6381
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 | 51.0 ± 1.3 | 50.3 ± 1.7 | 50.1 ± 1.4 | 49.3 ± 1.9 |
| 50 | 51.1 ± 1.6 | 50.4 ± 1.8 | 49.9 ± 1.2 | 49.7 ± 1.7 |
| 75 | 51.4 ± 1.6 | 49.7 ± 1.6 | 50.3 ± 1.6 | 49.4 ± 1.7 |
| 100 | 53.7 ± 5.9 | 50.7 ± 1.8 | 52.1 ± 4.6 | 49.8 ± 1.6 |
| 125 | 63.2 ± 13.6 | 56.2 ± 8.3 | 58.9 ± 13.0 | 51.9 ± 5.8 |
| 150 | 72.9 ± 17.9 | 67.0 ± 14.8 | 68.9 ± 17.8 | 62.7 ± 13.0 |
| 175 | 83.1 ± 15.8 | 76.8 ± 14.7 | 78.0 ± 17.4 | 71.3 ± 14.5 |
| 200 | 90.7 ± 12.1 | 85.3 ± 12.1 | 85.9 ± 13.7 | 79.5 ± 13.2 |
| 225 | 96.2 ± 4.1 | 92.1 ± 5.4 | 91.8 ± 7.1 | 86.7 ± 7.2 |
| 250 | 97.6 ± 1.6 | 95.5 ± 2.6 | 93.6 ± 4.6 | 89.8 ± 6.0 |
| 275 | 98.3 ± 1.2 | 96.7 ± 1.9 | 94.1 ± 3.6 | 90.3 ± 5.2 |
| 300 | 98.1 ± 1.0 | 96.7 ± 1.6 | 94.1 ± 3.8 | 90.4 ± 5.2 |
| 325 | 98.2 ± 1.0 | 96.9 ± 1.7 | 93.6 ± 4.0 | 90.0 ± 5.9 |
| 350 | 98.3 ± 0.8 | 96.9 ± 1.2 | 93.5 ± 3.3 | 89.5 ± 4.5 |
| 375 | 98.5 ± 0.7 | 97.5 ± 0.9 | 94.1 ± 2.7 | 90.1 ± 3.5 |
| 400 | 98.7 ± 0.7 | 97.6 ± 1.1 | 95.1 ± 2.5 | 91.1 ± 3.2 |