r2_c3
done · peak unseen_mismatch 99.0
α schedule
α 2.8037 ×25 α 2.7978 ×25 α 2.7808 ×25 α 2.7489 ×25 α 2.699 ×25 α 2.6285 ×25 α 2.5348 ×25 α 2.4157 ×25 α 2.269 ×25 α 2.0927 ×25 α 1.885 ×25 α 1.644 ×25 α 1.3679 ×25 α 1.0552 ×25 α 0.7042 ×25 α 0.3133 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.3133
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 | 50.8 ± 1.2 | 50.2 ± 1.8 | 50.0 ± 1.4 | 49.2 ± 1.8 |
| 50 | 51.3 ± 1.4 | 50.4 ± 1.8 | 49.9 ± 1.2 | 49.8 ± 1.7 |
| 75 | 51.5 ± 1.8 | 50.0 ± 1.8 | 50.4 ± 1.5 | 49.5 ± 1.7 |
| 100 | 54.0 ± 5.9 | 51.7 ± 2.5 | 51.7 ± 4.4 | 50.0 ± 1.7 |
| 125 | 63.6 ± 13.3 | 58.7 ± 9.9 | 58.2 ± 12.4 | 54.0 ± 7.8 |
| 150 | 73.3 ± 15.9 | 70.0 ± 14.5 | 67.1 ± 15.3 | 63.1 ± 13.2 |
| 175 | 83.0 ± 12.6 | 79.3 ± 12.9 | 74.9 ± 13.6 | 70.4 ± 13.7 |
| 200 | 90.6 ± 8.3 | 87.0 ± 8.6 | 82.5 ± 11.4 | 77.8 ± 11.6 |
| 225 | 93.8 ± 5.5 | 91.3 ± 6.0 | 86.1 ± 9.1 | 82.3 ± 9.3 |
| 250 | 95.8 ± 2.4 | 93.7 ± 2.9 | 88.7 ± 6.1 | 85.3 ± 6.6 |
| 275 | 97.0 ± 1.5 | 95.6 ± 2.0 | 91.0 ± 5.1 | 87.1 ± 5.7 |
| 300 | 98.0 ± 1.1 | 96.7 ± 1.5 | 93.0 ± 3.7 | 89.8 ± 4.5 |
| 325 | 99.2 ± 0.5 | 98.2 ± 0.9 | 96.0 ± 2.7 | 92.9 ± 3.7 |
| 350 | 99.9 ± 0.1 | 99.6 ± 0.4 | 99.0 ± 0.8 | 97.1 ± 2.2 |
| 375 | 100.0 ± 0.0 | 99.9 ± 0.2 | 99.9 ± 0.3 | 99.0 ± 1.0 |
| 400 | 100.0 ± 0.0 | 97.2 ± 8.4 | 100.0 ± 0.1 | 95.8 ± 8.1 |