r8_c4
done · peak unseen_mismatch 55.2
α schedule
α 2.8022 ×25 α 2.4463 ×25 α 2.2242 ×25 α 0.6678 ×25 α 0.6215 ×25 α 1.7868 ×25 α 1.6764 ×25 α 1.5762 ×25 α 0.4828 ×25 α 0.455 ×25 α 0.4289 ×25 α 1.2432 ×25 α 1.1719 ×25 α 0.3591 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.3591
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 | 51.1 ± 1.7 | 50.3 ± 1.8 | 49.9 ± 1.2 | 49.9 ± 1.6 |
| 75 | 51.4 ± 1.5 | 49.7 ± 1.7 | 50.4 ± 1.5 | 49.3 ± 1.6 |
| 100 | 52.1 ± 3.9 | 49.9 ± 2.0 | 51.2 ± 2.8 | 49.6 ± 1.8 |
| 125 | 57.5 ± 12.0 | 47.4 ± 5.1 | 55.8 ± 12.4 | 47.2 ± 4.7 |
| 150 | 63.3 ± 18.3 | 45.5 ± 10.9 | 62.2 ± 18.8 | 44.6 ± 10.7 |
| 175 | 73.5 ± 20.0 | 46.9 ± 13.1 | 72.8 ± 20.6 | 46.4 ± 12.5 |
| 200 | 84.5 ± 20.5 | 55.7 ± 23.5 | 84.1 ± 20.7 | 55.2 ± 23.1 |
| 225 | 91.1 ± 16.7 | 53.1 ± 21.3 | 90.6 ± 17.1 | 53.5 ± 21.2 |
| 250 | 97.1 ± 9.6 | 44.6 ± 25.6 | 96.9 ± 10.2 | 44.9 ± 25.3 |
| 275 | 100.0 ± 0.2 | 37.3 ± 29.0 | 100.0 ± 0.2 | 38.1 ± 28.8 |
| 300 | 100.0 ± 0.1 | 41.4 ± 31.0 | 100.0 ± 0.1 | 41.5 ± 30.8 |
| 325 | 100.0 ± 0.0 | 42.2 ± 31.9 | 100.0 ± 0.0 | 41.8 ± 31.4 |
| 350 | 100.0 ± 0.0 | 44.3 ± 27.3 | 100.0 ± 0.0 | 44.6 ± 26.6 |