r6_c5
done · peak unseen_mismatch 98.9
α schedule
α 4.3443 ×25 α 4.0663 ×25 α 3.9039 ×25 α 0.9978 ×25 α 0.9719 ×25 α 3.6011 ×25 α 3.5277 ×25 α 3.4619 ×25 α 0.8974 ×25 α 0.8828 ×25 α 0.8694 ×25 α 3.2483 ×25 α 3.2035 ×25 α 0.8338 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.8338
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 | 50.4 ± 1.6 | 49.8 ± 1.5 | 49.5 ± 1.9 |
| 50 | 51.4 ± 1.4 | 50.4 ± 1.6 | 50.2 ± 1.2 | 49.8 ± 1.7 |
| 75 | 52.7 ± 2.4 | 50.3 ± 1.9 | 50.6 ± 1.7 | 49.4 ± 1.5 |
| 100 | 54.3 ± 5.8 | 51.5 ± 2.1 | 52.2 ± 4.2 | 49.9 ± 1.8 |
| 125 | 63.6 ± 15.6 | 52.2 ± 6.7 | 60.6 ± 16.4 | 50.4 ± 6.8 |
| 150 | 76.3 ± 18.9 | 66.7 ± 15.4 | 73.3 ± 19.2 | 63.4 ± 14.1 |
| 175 | 86.2 ± 13.4 | 79.9 ± 13.5 | 81.2 ± 14.5 | 75.1 ± 13.5 |
| 200 | 91.5 ± 7.1 | 87.8 ± 7.6 | 85.3 ± 9.6 | 81.6 ± 9.5 |
| 225 | 99.5 ± 0.8 | 94.4 ± 13.8 | 99.1 ± 1.5 | 93.3 ± 13.5 |
| 250 | 99.9 ± 0.1 | 92.9 ± 13.6 | 99.8 ± 0.4 | 91.8 ± 14.5 |
| 275 | 100.0 ± 0.0 | 95.9 ± 8.2 | 100.0 ± 0.0 | 95.3 ± 7.9 |
| 300 | 97.9 ± 2.1 | 95.5 ± 3.2 | 93.8 ± 5.7 | 89.1 ± 7.8 |
| 325 | 96.6 ± 2.4 | 93.1 ± 4.6 | 91.6 ± 5.5 | 85.2 ± 8.6 |
| 350 | 100.0 ± 0.0 | 99.5 ± 1.5 | 100.0 ± 0.0 | 98.9 ± 3.9 |