fixed_30
done · peak unseen_mismatch 74.7
α schedule
α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25 α 3.0 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.0
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 | 51.2 ± 1.7 | 50.1 ± 1.7 | 50.0 ± 2.0 | 50.3 ± 1.3 |
| 50 | 51.6 ± 1.8 | 50.7 ± 1.5 | 50.3 ± 1.8 | 49.8 ± 1.8 |
| 75 | 52.5 ± 2.3 | 50.4 ± 2.1 | 50.4 ± 2.0 | 49.0 ± 1.6 |
| 100 | 57.3 ± 6.0 | 53.2 ± 4.4 | 53.5 ± 4.6 | 50.9 ± 2.3 |
| 125 | 69.6 ± 12.5 | 63.4 ± 10.4 | 62.5 ± 12.7 | 56.8 ± 8.5 |
| 150 | 82.3 ± 9.2 | 76.8 ± 9.7 | 72.3 ± 12.5 | 66.5 ± 10.4 |
| 175 | 88.5 ± 5.4 | 84.2 ± 6.7 | 77.7 ± 10.3 | 72.3 ± 9.2 |
| 200 | 90.1 ± 3.9 | 87.5 ± 4.0 | 78.5 ± 8.4 | 74.7 ± 7.3 |
| 225 | 90.4 ± 3.1 | 88.3 ± 2.4 | 78.2 ± 7.1 | 74.2 ± 6.1 |
| 250 | 90.8 ± 2.6 | 88.5 ± 1.8 | 77.9 ± 7.0 | 74.0 ± 5.6 |
| 275 | 91.3 ± 2.0 | 89.1 ± 1.6 | 78.5 ± 5.9 | 74.3 ± 5.4 |
| 300 | 91.3 ± 2.1 | 89.8 ± 1.7 | 78.0 ± 5.2 | 74.4 ± 5.0 |
| 325 | 91.7 ± 1.6 | 89.9 ± 1.7 | 78.0 ± 4.5 | 74.2 ± 5.0 |
| 350 | 91.9 ± 1.5 | 90.0 ± 1.5 | 77.5 ± 4.6 | 73.4 ± 4.7 |
| 375 | 91.8 ± 1.5 | 90.0 ± 1.5 | 77.6 ± 3.9 | 73.2 ± 4.9 |
| 400 | 92.3 ± 1.7 | 90.2 ± 1.6 | 78.0 ± 4.1 | 73.2 ± 4.4 |