r2_c6
done · peak unseen_mismatch 89.8
α schedule
α 1.6149 ×25 α 1.6357 ×25 α 1.6729 ×25 α 1.7249 ×25 α 1.7909 ×25 α 1.8703 ×25 α 1.9626 ×25 α 2.0676 ×25 α 2.1848 ×25 α 2.314 ×25 α 2.455 ×25 α 2.6076 ×25 α 2.7716 ×25 α 2.9468 ×25 α 3.1332 ×25 α 3.3304 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.3304
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 | 50.4 ± 1.4 | 50.3 ± 1.6 | 49.8 ± 1.6 | 49.5 ± 2.0 |
| 50 | 50.8 ± 1.4 | 50.0 ± 1.6 | 50.1 ± 1.4 | 49.8 ± 1.3 |
| 75 | 50.6 ± 1.5 | 49.7 ± 1.7 | 50.2 ± 1.3 | 49.8 ± 1.7 |
| 100 | 50.5 ± 2.1 | 50.1 ± 1.5 | 50.0 ± 1.3 | 49.8 ± 1.5 |
| 125 | 52.2 ± 2.8 | 49.4 ± 1.8 | 51.3 ± 3.0 | 49.2 ± 2.1 |
| 150 | 55.3 ± 9.5 | 50.4 ± 2.4 | 53.8 ± 9.3 | 49.3 ± 1.7 |
| 175 | 63.5 ± 16.0 | 51.9 ± 6.1 | 62.3 ± 15.9 | 50.9 ± 4.8 |
| 200 | 77.3 ± 20.0 | 60.5 ± 17.8 | 76.0 ± 20.2 | 59.1 ± 17.3 |
| 225 | 84.0 ± 21.0 | 69.1 ± 19.4 | 83.4 ± 21.2 | 67.8 ± 18.2 |
| 250 | 88.2 ± 18.4 | 76.7 ± 18.5 | 86.8 ± 19.8 | 73.7 ± 19.1 |
| 275 | 92.9 ± 13.9 | 87.2 ± 14.6 | 90.8 ± 14.2 | 83.8 ± 14.6 |
| 300 | 94.2 ± 12.5 | 90.3 ± 13.9 | 92.1 ± 12.7 | 87.4 ± 14.0 |
| 325 | 95.7 ± 8.8 | 93.5 ± 9.4 | 92.9 ± 9.4 | 89.8 ± 10.0 |
| 350 | 96.0 ± 5.9 | 94.2 ± 6.6 | 91.9 ± 7.5 | 89.0 ± 7.9 |
| 375 | 95.7 ± 2.3 | 93.9 ± 2.5 | 90.8 ± 5.0 | 87.0 ± 5.5 |
| 400 | 95.3 ± 2.3 | 93.4 ± 2.8 | 88.6 ± 5.6 | 85.5 ± 5.6 |