r5_c0
done · peak unseen_mismatch 90.0
α schedule
α 1.8258 ×25 α 1.9279 ×25 α 2.0009 ×25 α 2.0615 ×25 α 2.1146 ×25 α 2.1627 ×25 α 2.2069 ×25 α 2.2482 ×25 α 2.2871 ×25 α 2.3239 ×25 α 2.3591 ×25 α 2.3927 ×25 α 2.4251 ×25 α 2.4563 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.4563
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 | 50.6 ± 1.5 | 50.1 ± 1.5 | 50.2 ± 1.5 | 49.3 ± 2.0 |
| 50 | 50.9 ± 1.6 | 49.9 ± 1.5 | 50.1 ± 1.5 | 50.2 ± 1.3 |
| 75 | 51.0 ± 1.3 | 49.6 ± 1.6 | 50.0 ± 1.6 | 49.6 ± 1.4 |
| 100 | 51.1 ± 2.4 | 50.1 ± 1.5 | 50.1 ± 1.3 | 50.0 ± 1.7 |
| 125 | 53.5 ± 3.9 | 49.5 ± 2.1 | 52.0 ± 3.4 | 49.0 ± 2.0 |
| 150 | 61.2 ± 15.9 | 52.8 ± 7.3 | 59.1 ± 15.0 | 50.9 ± 6.7 |
| 175 | 71.5 ± 18.5 | 62.5 ± 13.4 | 68.9 ± 18.8 | 60.6 ± 12.3 |
| 200 | 80.0 ± 19.0 | 73.0 ± 16.8 | 77.9 ± 20.2 | 69.8 ± 16.9 |
| 225 | 87.7 ± 13.6 | 81.5 ± 15.4 | 84.4 ± 16.1 | 77.6 ± 16.3 |
| 250 | 95.2 ± 5.6 | 91.6 ± 7.1 | 92.5 ± 8.3 | 87.4 ± 9.8 |
| 275 | 97.2 ± 2.7 | 95.0 ± 3.3 | 94.0 ± 6.1 | 90.0 ± 6.3 |
| 300 | 97.5 ± 2.0 | 95.4 ± 2.6 | 93.3 ± 5.5 | 89.3 ± 6.1 |
| 325 | 96.8 ± 2.1 | 95.1 ± 2.3 | 91.8 ± 5.4 | 87.7 ± 6.5 |
| 350 | 96.7 ± 1.9 | 94.8 ± 2.4 | 90.5 ± 5.1 | 86.6 ± 6.2 |