r7_c4
done · peak unseen_mismatch 81.7
α schedule
α 3.5497 ×25 α 3.5549 ×25 α 3.5693 ×25 α 1.8493 ×25 α 1.8701 ×25 α 3.6928 ×25 α 3.7671 ×25 α 3.8607 ×25 α 2.0444 ×25 α 2.1145 ×25 α 2.1967 ×25 α 4.4555 ×25 α 4.6653 ×25 α 2.5211 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.5211
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 | 50.9 ± 1.3 | 50.3 ± 1.6 | 50.0 ± 1.5 | 49.5 ± 1.7 |
| 50 | 51.2 ± 1.5 | 50.4 ± 1.7 | 50.0 ± 1.0 | 50.0 ± 1.7 |
| 75 | 52.1 ± 1.9 | 50.1 ± 1.7 | 50.6 ± 1.5 | 49.3 ± 1.5 |
| 100 | 55.1 ± 5.3 | 52.4 ± 2.1 | 52.0 ± 3.1 | 50.1 ± 1.6 |
| 125 | 66.5 ± 13.5 | 59.0 ± 7.9 | 60.5 ± 13.4 | 54.5 ± 6.6 |
| 150 | 78.5 ± 13.7 | 74.2 ± 13.1 | 71.1 ± 15.3 | 66.5 ± 13.7 |
| 175 | 85.9 ± 7.7 | 82.8 ± 8.4 | 75.8 ± 11.4 | 72.0 ± 11.2 |
| 200 | 87.8 ± 4.6 | 85.2 ± 4.7 | 76.3 ± 9.0 | 73.4 ± 8.8 |
| 225 | 93.2 ± 3.3 | 91.1 ± 3.5 | 83.7 ± 6.9 | 80.3 ± 7.5 |
| 250 | 94.1 ± 2.2 | 92.3 ± 2.2 | 84.9 ± 5.9 | 81.2 ± 5.8 |
| 275 | 95.0 ± 1.8 | 93.2 ± 1.6 | 85.3 ± 4.8 | 81.7 ± 4.5 |
| 300 | 90.8 ± 1.9 | 88.5 ± 2.1 | 78.6 ± 4.0 | 74.8 ± 4.2 |
| 325 | 90.1 ± 2.0 | 88.0 ± 1.9 | 77.6 ± 3.3 | 73.7 ± 3.7 |
| 350 | 94.2 ± 1.6 | 92.6 ± 1.6 | 83.1 ± 3.4 | 79.7 ± 3.6 |