r5_c6
done · peak unseen_mismatch 73.9
α schedule
α 2.9863 ×25 α 3.013 ×25 α 3.0297 ×25 α 3.0425 ×25 α 3.0531 ×25 α 3.0624 ×25 α 3.0706 ×25 α 3.0781 ×25 α 3.085 ×25 α 3.0913 ×25 α 3.0973 ×25 α 3.1029 ×25 α 3.1082 ×25 α 3.1132 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.1132
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 | 50.7 ± 1.2 | 50.4 ± 1.6 | 50.0 ± 1.3 | 49.4 ± 1.8 |
| 50 | 51.6 ± 1.5 | 50.4 ± 1.5 | 50.0 ± 1.0 | 49.9 ± 1.8 |
| 75 | 52.1 ± 1.7 | 49.8 ± 1.8 | 50.6 ± 1.6 | 49.5 ± 1.5 |
| 100 | 55.7 ± 6.4 | 53.3 ± 4.1 | 52.1 ± 3.9 | 50.3 ± 1.5 |
| 125 | 65.4 ± 12.9 | 60.8 ± 11.0 | 57.9 ± 11.2 | 54.0 ± 7.8 |
| 150 | 74.7 ± 13.7 | 71.8 ± 13.0 | 66.0 ± 13.1 | 62.9 ± 10.9 |
| 175 | 83.4 ± 9.3 | 80.5 ± 9.0 | 71.9 ± 11.8 | 68.5 ± 10.6 |
| 200 | 87.5 ± 5.7 | 85.0 ± 5.4 | 75.2 ± 9.5 | 72.2 ± 8.6 |
| 225 | 88.9 ± 3.6 | 86.3 ± 3.3 | 75.2 ± 7.0 | 72.0 ± 6.5 |
| 250 | 89.6 ± 2.8 | 87.3 ± 2.5 | 75.5 ± 6.9 | 72.7 ± 6.3 |
| 275 | 90.2 ± 2.5 | 88.1 ± 1.9 | 75.1 ± 4.6 | 72.3 ± 5.0 |
| 300 | 90.9 ± 2.1 | 89.0 ± 1.8 | 76.7 ± 4.8 | 73.9 ± 4.8 |
| 325 | 91.2 ± 2.1 | 89.3 ± 1.9 | 76.7 ± 4.4 | 73.8 ± 4.8 |
| 350 | 91.3 ± 1.8 | 89.5 ± 1.7 | 76.1 ± 4.2 | 73.3 ± 4.5 |