r2_c4
done · peak unseen_mismatch 50.6
α schedule
α 1.5241 ×25 α 1.5177 ×25 α 1.5048 ×25 α 1.4854 ×25 α 1.4594 ×25 α 1.4269 ×25 α 1.3878 ×25 α 1.3421 ×25 α 1.2898 ×25 α 1.231 ×25 α 1.1655 ×25 α 1.0935 ×25 α 1.0148 ×25 α 0.9295 ×25 α 0.8376 ×25 α 0.739 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.739
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 | 50.5 ± 1.4 | 50.0 ± 1.5 | 49.9 ± 1.4 | 49.4 ± 1.9 |
| 50 | 50.6 ± 2.0 | 50.1 ± 1.7 | 50.0 ± 1.6 | 49.8 ± 1.6 |
| 75 | 50.7 ± 1.5 | 49.5 ± 1.7 | 50.3 ± 1.3 | 49.5 ± 1.9 |
| 100 | 50.3 ± 2.1 | 50.4 ± 1.4 | 50.1 ± 1.5 | 50.0 ± 1.2 |
| 125 | 51.8 ± 2.7 | 49.1 ± 2.0 | 50.9 ± 3.2 | 49.0 ± 2.7 |
| 150 | 55.7 ± 11.5 | 49.4 ± 4.8 | 54.7 ± 11.5 | 48.8 ± 5.5 |
| 175 | 61.9 ± 16.7 | 44.7 ± 10.8 | 61.7 ± 16.7 | 44.9 ± 11.0 |
| 200 | 71.7 ± 21.9 | 45.8 ± 12.5 | 71.3 ± 22.0 | 46.2 ± 12.5 |
| 225 | 77.3 ± 23.6 | 49.2 ± 15.5 | 77.4 ± 23.6 | 49.7 ± 15.9 |
| 250 | 82.5 ± 21.9 | 46.4 ± 21.8 | 82.2 ± 22.1 | 46.2 ± 21.5 |
| 275 | 86.8 ± 21.1 | 43.3 ± 22.2 | 86.7 ± 21.2 | 42.9 ± 21.8 |
| 300 | 91.3 ± 17.2 | 42.9 ± 24.3 | 91.2 ± 17.3 | 43.7 ± 24.2 |
| 325 | 96.5 ± 10.3 | 45.2 ± 22.9 | 96.3 ± 10.8 | 45.6 ± 22.3 |
| 350 | 98.0 ± 9.7 | 46.7 ± 21.2 | 97.9 ± 9.9 | 46.8 ± 21.6 |
| 375 | 98.1 ± 9.5 | 49.8 ± 21.1 | 98.1 ± 9.4 | 50.6 ± 21.4 |
| 400 | 99.9 ± 0.6 | 45.2 ± 19.7 | 99.9 ± 0.4 | 45.0 ± 19.7 |