r7_c2
done · peak unseen_mismatch 89.7
α schedule
α 2.9291 ×25 α 2.9411 ×25 α 2.9512 ×25 α 0.7194 ×25 α 0.7215 ×25 α 2.9772 ×25 α 2.985 ×25 α 2.9926 ×25 α 0.729 ×25 α 0.7307 ×25 α 0.7324 ×25 α 3.0209 ×25 α 3.0276 ×25 α 0.7373 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.7373
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 ± 1.2 | 50.5 ± 1.6 | 50.1 ± 1.5 | 49.4 ± 1.8 |
| 50 | 51.4 ± 1.5 | 50.3 ± 1.5 | 49.7 ± 1.1 | 50.1 ± 1.6 |
| 75 | 52.0 ± 1.8 | 49.8 ± 1.6 | 50.5 ± 1.5 | 49.6 ± 1.7 |
| 100 | 52.4 ± 4.1 | 50.3 ± 1.9 | 51.2 ± 3.1 | 49.8 ± 1.5 |
| 125 | 57.3 ± 10.9 | 48.9 ± 2.8 | 55.2 ± 11.3 | 48.2 ± 2.5 |
| 150 | 66.6 ± 18.5 | 52.3 ± 9.2 | 64.7 ± 18.8 | 50.7 ± 8.4 |
| 175 | 78.7 ± 20.0 | 65.7 ± 16.7 | 76.9 ± 20.3 | 63.1 ± 15.7 |
| 200 | 87.8 ± 15.8 | 81.0 ± 16.3 | 85.4 ± 16.9 | 78.2 ± 16.7 |
| 225 | 95.0 ± 11.4 | 82.5 ± 18.6 | 94.4 ± 12.3 | 81.4 ± 18.1 |
| 250 | 99.6 ± 1.6 | 79.4 ± 24.6 | 99.3 ± 2.5 | 78.6 ± 24.5 |
| 275 | 100.0 ± 0.0 | 78.2 ± 25.0 | 100.0 ± 0.1 | 78.0 ± 24.8 |
| 300 | 99.7 ± 0.9 | 82.7 ± 25.0 | 99.2 ± 3.2 | 82.0 ± 24.2 |
| 325 | 99.1 ± 1.4 | 90.0 ± 17.7 | 97.9 ± 3.0 | 87.7 ± 16.8 |
| 350 | 100.0 ± 0.0 | 90.0 ± 19.5 | 100.0 ± 0.0 | 89.7 ± 19.2 |