r3_c6
done · peak unseen_mismatch 91.6
α schedule
α 1.6368 ×25 α 1.6549 ×25 α 1.6848 ×25 α 1.7249 ×25 α 1.7742 ×25 α 1.8322 ×25 α 1.8982 ×25 α 1.9719 ×25 α 2.0531 ×25 α 2.1415 ×25 α 2.2368 ×25 α 2.3388 ×25 α 2.4474 ×25 α 2.5625 ×25 α 2.6839 ×25 α 2.8114 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.8114
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 | 50.6 ± 1.3 | 50.0 ± 1.6 | 49.8 ± 1.5 | 49.6 ± 1.9 |
| 50 | 50.8 ± 1.4 | 50.1 ± 1.6 | 49.8 ± 1.7 | 49.9 ± 1.3 |
| 75 | 50.7 ± 1.7 | 50.0 ± 1.7 | 50.4 ± 1.6 | 49.6 ± 1.7 |
| 100 | 50.8 ± 2.1 | 50.1 ± 1.5 | 50.2 ± 1.4 | 49.8 ± 1.8 |
| 125 | 52.7 ± 2.8 | 49.6 ± 2.0 | 51.6 ± 3.5 | 49.3 ± 2.3 |
| 150 | 57.7 ± 11.5 | 51.3 ± 3.7 | 55.6 ± 10.7 | 50.3 ± 3.2 |
| 175 | 67.5 ± 18.4 | 53.2 ± 9.5 | 66.3 ± 18.6 | 52.2 ± 10.0 |
| 200 | 78.8 ± 21.0 | 60.4 ± 15.0 | 78.0 ± 21.3 | 59.6 ± 14.4 |
| 225 | 84.6 ± 20.7 | 67.8 ± 17.7 | 83.5 ± 21.5 | 66.8 ± 17.1 |
| 250 | 89.2 ± 16.6 | 75.9 ± 22.5 | 87.6 ± 17.6 | 74.0 ± 22.7 |
| 275 | 94.4 ± 12.6 | 87.2 ± 15.5 | 93.2 ± 13.2 | 85.1 ± 15.5 |
| 300 | 96.2 ± 8.6 | 90.5 ± 12.1 | 94.7 ± 9.4 | 87.5 ± 12.2 |
| 325 | 97.9 ± 3.2 | 94.8 ± 7.1 | 95.8 ± 4.6 | 91.6 ± 7.8 |
| 350 | 97.7 ± 2.0 | 95.7 ± 3.2 | 94.5 ± 4.2 | 90.9 ± 5.7 |
| 375 | 97.1 ± 1.8 | 95.4 ± 2.4 | 92.5 ± 5.0 | 89.2 ± 6.2 |
| 400 | 96.6 ± 1.9 | 95.0 ± 2.4 | 91.4 ± 4.6 | 88.4 ± 5.9 |