r7_c3
done · peak unseen_mismatch 96.8
α schedule
α 2.8591 ×25 α 2.8586 ×25 α 2.8558 ×25 α 1.2615 ×25 α 1.2544 ×25 α 2.8038 ×25 α 2.7585 ×25 α 2.6912 ×25 α 1.1499 ×25 α 1.093 ×25 α 1.0184 ×25 α 2.0839 ×25 α 1.8145 ×25 α 0.657 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.657
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 | 50.9 ± 1.2 | 50.4 ± 1.6 | 50.0 ± 1.5 | 49.4 ± 1.7 |
| 50 | 51.3 ± 1.6 | 50.3 ± 1.6 | 49.9 ± 1.1 | 49.7 ± 1.7 |
| 75 | 51.8 ± 1.5 | 49.8 ± 1.7 | 50.4 ± 1.3 | 49.6 ± 1.5 |
| 100 | 52.5 ± 4.5 | 50.8 ± 1.6 | 51.2 ± 2.8 | 49.8 ± 1.5 |
| 125 | 59.1 ± 13.0 | 50.4 ± 2.9 | 57.2 ± 13.2 | 49.1 ± 2.9 |
| 150 | 68.9 ± 18.4 | 59.7 ± 10.8 | 66.7 ± 18.9 | 56.5 ± 9.3 |
| 175 | 79.3 ± 18.7 | 71.6 ± 16.9 | 76.6 ± 19.0 | 68.7 ± 16.0 |
| 200 | 88.6 ± 13.8 | 83.0 ± 13.6 | 84.8 ± 15.0 | 78.8 ± 14.3 |
| 225 | 96.5 ± 9.8 | 89.9 ± 13.5 | 95.6 ± 10.4 | 87.8 ± 14.5 |
| 250 | 98.2 ± 7.7 | 89.0 ± 14.7 | 97.8 ± 8.8 | 87.3 ± 15.5 |
| 275 | 99.7 ± 1.3 | 90.2 ± 14.1 | 99.7 ± 1.2 | 88.9 ± 14.3 |
| 300 | 99.4 ± 0.8 | 93.4 ± 10.6 | 97.9 ± 2.7 | 90.0 ± 10.7 |
| 325 | 99.4 ± 0.6 | 97.9 ± 2.8 | 97.8 ± 2.6 | 94.9 ± 5.2 |
| 350 | 100.0 ± 0.0 | 96.8 ± 9.7 | 100.0 ± 0.0 | 96.8 ± 9.4 |