r6_c4
done · peak unseen_mismatch 94.2
α schedule
α 2.6129 ×25 α 3.3291 ×25 α 3.881 ×25 α 1.0587 ×25 α 1.1634 ×25 α 5.0 ×25 α 5.0 ×25 α 5.0 ×25 α 1.5187 ×25 α 1.5973 ×25 α 1.6731 ×25 α 5.0 ×25 α 5.0 ×25 α 1.8865 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.8865
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 ± 1.4 | 50.6 ± 1.7 | 50.1 ± 1.5 | 49.3 ± 1.6 |
| 50 | 51.1 ± 1.3 | 50.2 ± 1.5 | 49.6 ± 1.2 | 50.0 ± 1.7 |
| 75 | 51.9 ± 1.6 | 50.0 ± 1.7 | 50.3 ± 1.6 | 49.3 ± 1.6 |
| 100 | 52.5 ± 4.4 | 50.4 ± 1.8 | 51.4 ± 3.9 | 49.7 ± 1.4 |
| 125 | 58.5 ± 12.2 | 49.7 ± 3.7 | 56.2 ± 12.3 | 48.8 ± 2.9 |
| 150 | 69.4 ± 18.1 | 59.2 ± 10.1 | 67.1 ± 18.5 | 56.3 ± 9.1 |
| 175 | 78.1 ± 18.1 | 72.2 ± 16.8 | 74.5 ± 18.9 | 67.9 ± 15.7 |
| 200 | 84.5 ± 12.1 | 80.1 ± 12.2 | 78.3 ± 14.4 | 73.5 ± 13.4 |
| 225 | 95.2 ± 5.4 | 88.2 ± 12.2 | 92.3 ± 7.7 | 85.0 ± 12.7 |
| 250 | 98.6 ± 1.5 | 94.3 ± 5.0 | 96.8 ± 3.2 | 91.0 ± 6.0 |
| 275 | 99.3 ± 0.8 | 97.2 ± 2.5 | 98.0 ± 2.2 | 94.2 ± 4.7 |
| 300 | 96.1 ± 2.4 | 93.8 ± 3.1 | 90.9 ± 5.6 | 87.4 ± 6.9 |
| 325 | 94.1 ± 3.0 | 91.6 ± 3.3 | 87.6 ± 5.5 | 83.6 ± 6.9 |
| 350 | 97.9 ± 1.0 | 96.5 ± 1.2 | 93.3 ± 3.0 | 90.2 ± 3.2 |