r1_c4
done · peak unseen_mismatch 91.3
α schedule
α 1.5633 ×25 α 1.5646 ×25 α 1.5699 ×25 α 1.5821 ×25 α 1.6046 ×25 α 1.6405 ×25 α 1.6932 ×25 α 1.7662 ×25 α 1.8631 ×25 α 1.9873 ×25 α 2.1426 ×25 α 2.3326 ×25 α 2.5609 ×25 α 2.8314 ×25 α 3.1478 ×25 α 3.5139 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.5139
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 | 50.4 ± 1.6 | 50.1 ± 1.5 | 49.9 ± 1.6 | 49.5 ± 2.1 |
| 50 | 50.7 ± 1.8 | 50.1 ± 1.6 | 50.1 ± 1.5 | 49.7 ± 1.4 |
| 75 | 50.5 ± 1.6 | 49.6 ± 1.6 | 50.3 ± 1.3 | 49.5 ± 1.5 |
| 100 | 50.5 ± 2.1 | 50.2 ± 1.6 | 50.2 ± 1.4 | 49.9 ± 1.4 |
| 125 | 52.1 ± 2.8 | 49.0 ± 2.1 | 51.2 ± 3.2 | 49.1 ± 2.3 |
| 150 | 56.4 ± 12.4 | 49.4 ± 6.7 | 55.3 ± 12.3 | 48.6 ± 6.7 |
| 175 | 64.0 ± 18.3 | 50.5 ± 10.0 | 63.6 ± 18.1 | 49.6 ± 9.8 |
| 200 | 72.4 ± 21.4 | 52.2 ± 14.6 | 71.7 ± 21.6 | 51.6 ± 14.6 |
| 225 | 80.2 ± 22.4 | 58.7 ± 20.9 | 79.4 ± 22.7 | 58.2 ± 20.2 |
| 250 | 86.2 ± 20.3 | 68.4 ± 24.2 | 85.5 ± 20.7 | 67.3 ± 23.2 |
| 275 | 90.5 ± 16.8 | 75.7 ± 23.0 | 89.4 ± 17.5 | 74.2 ± 22.7 |
| 300 | 94.9 ± 11.0 | 78.1 ± 24.3 | 93.8 ± 11.6 | 76.2 ± 23.8 |
| 325 | 97.9 ± 5.6 | 89.6 ± 15.0 | 96.6 ± 6.6 | 87.7 ± 14.5 |
| 350 | 97.8 ± 4.6 | 90.4 ± 17.1 | 96.0 ± 5.9 | 88.0 ± 16.7 |
| 375 | 97.2 ± 3.1 | 95.1 ± 3.9 | 94.2 ± 5.6 | 91.3 ± 6.2 |
| 400 | 96.6 ± 2.7 | 95.1 ± 3.1 | 92.5 ± 5.8 | 89.9 ± 6.7 |