p20
done · peak unseen_mismatch 70.3
α schedule
α 4.5614 ×25 α 4.5601 ×25 α 4.5551 ×25 α 4.5439 ×25 α 4.5237 ×25 α 4.4919 ×25 α 4.4458 ×25 α 4.3827 ×25 α 4.2999 ×25 α 4.1946 ×25 α 4.064 ×25 α 3.9054 ×25 α 3.7161 ×25 α 3.4931 ×25 α 3.2337 ×25 α 2.935 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.935
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 | 50.7 ± 1.3 | 50.5 ± 1.6 | 49.8 ± 1.5 | 49.6 ± 1.8 |
| 50 | 51.5 ± 1.7 | 50.5 ± 1.6 | 50.1 ± 1.3 | 49.9 ± 1.7 |
| 75 | 52.7 ± 2.2 | 50.4 ± 1.8 | 50.4 ± 1.2 | 49.4 ± 1.6 |
| 100 | 60.5 ± 8.9 | 57.9 ± 6.7 | 53.3 ± 5.5 | 51.6 ± 3.5 |
| 125 | 72.9 ± 9.4 | 69.0 ± 8.8 | 59.9 ± 8.5 | 57.1 ± 7.5 |
| 150 | 78.6 ± 5.2 | 76.1 ± 5.2 | 64.0 ± 7.8 | 61.1 ± 7.4 |
| 175 | 80.4 ± 3.4 | 78.6 ± 3.0 | 64.2 ± 6.5 | 62.2 ± 5.9 |
| 200 | 81.3 ± 2.7 | 79.4 ± 2.4 | 63.5 ± 4.9 | 61.5 ± 5.0 |
| 225 | 82.0 ± 3.0 | 80.2 ± 2.4 | 64.5 ± 5.1 | 62.6 ± 5.2 |
| 250 | 82.6 ± 2.4 | 81.0 ± 2.2 | 63.9 ± 4.2 | 62.7 ± 4.5 |
| 275 | 83.4 ± 2.6 | 82.3 ± 2.2 | 65.2 ± 4.7 | 63.8 ± 4.7 |
| 300 | 85.0 ± 2.8 | 83.5 ± 2.8 | 67.1 ± 5.7 | 65.6 ± 5.6 |
| 325 | 86.3 ± 2.6 | 85.2 ± 2.2 | 68.3 ± 5.7 | 66.7 ± 5.2 |
| 350 | 87.6 ± 2.1 | 86.1 ± 2.2 | 70.1 ± 3.7 | 68.3 ± 4.1 |
| 375 | 88.4 ± 1.8 | 87.0 ± 1.6 | 70.2 ± 4.4 | 67.9 ± 3.7 |
| 400 | 90.1 ± 1.8 | 88.8 ± 1.6 | 72.7 ± 4.3 | 70.3 ± 4.2 |