p13
done · peak unseen_mismatch 92.7
α schedule
α 2.4604 ×25 α 2.4599 ×25 α 2.4579 ×25 α 2.4528 ×25 α 2.4429 ×25 α 2.4263 ×25 α 2.4011 ×25 α 2.3652 ×25 α 2.3163 ×25 α 2.2521 ×25 α 2.1702 ×25 α 2.0681 ×25 α 1.9433 ×25 α 1.7931 ×25 α 1.6148 ×25 α 1.4058 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.4058
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 ± 1.4 | 50.3 ± 1.6 | 50.0 ± 1.6 | 49.4 ± 1.8 |
| 50 | 50.9 ± 1.5 | 50.2 ± 1.6 | 49.8 ± 1.1 | 49.9 ± 1.8 |
| 75 | 51.4 ± 1.7 | 49.8 ± 1.7 | 50.5 ± 1.5 | 49.2 ± 1.5 |
| 100 | 52.5 ± 3.8 | 50.8 ± 1.6 | 50.7 ± 2.5 | 49.7 ± 1.4 |
| 125 | 59.7 ± 10.7 | 53.1 ± 4.4 | 55.4 ± 10.0 | 50.3 ± 2.9 |
| 150 | 68.5 ± 15.5 | 63.3 ± 12.1 | 64.3 ± 15.0 | 58.0 ± 9.8 |
| 175 | 79.9 ± 16.2 | 74.5 ± 14.6 | 73.7 ± 17.7 | 68.3 ± 14.3 |
| 200 | 86.5 ± 12.5 | 81.4 ± 13.3 | 79.9 ± 14.5 | 74.0 ± 14.0 |
| 225 | 93.4 ± 6.5 | 89.8 ± 7.8 | 87.6 ± 9.1 | 82.3 ± 10.1 |
| 250 | 96.0 ± 2.5 | 93.5 ± 2.7 | 89.7 ± 6.6 | 85.6 ± 6.5 |
| 275 | 97.0 ± 1.7 | 95.1 ± 1.9 | 90.7 ± 5.4 | 86.8 ± 5.7 |
| 300 | 97.3 ± 1.3 | 95.7 ± 1.6 | 91.1 ± 4.6 | 87.4 ± 5.5 |
| 325 | 96.9 ± 1.3 | 95.5 ± 1.7 | 89.8 ± 4.9 | 86.1 ± 5.7 |
| 350 | 97.8 ± 0.9 | 96.3 ± 1.1 | 91.5 ± 3.7 | 87.3 ± 4.8 |
| 375 | 98.5 ± 0.9 | 97.4 ± 1.2 | 93.5 ± 3.1 | 89.9 ± 4.1 |
| 400 | 99.4 ± 0.4 | 98.4 ± 1.1 | 96.4 ± 1.8 | 92.7 ± 3.6 |