r5_c3
done · peak unseen_mismatch 89.2
α schedule
α 1.5502 ×25 α 1.557 ×25 α 1.5743 ×25 α 1.6043 ×25 α 1.6487 ×25 α 1.7091 ×25 α 1.7865 ×25 α 1.8821 ×25 α 1.997 ×25 α 2.1319 ×25 α 2.2877 ×25 α 2.4653 ×25 α 2.6654 ×25 α 2.8886 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.8886
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 ± 1.7 | 50.1 ± 1.5 | 49.9 ± 1.7 | 49.5 ± 2.1 |
| 50 | 50.6 ± 1.7 | 50.2 ± 1.7 | 50.1 ± 1.6 | 49.6 ± 1.4 |
| 75 | 50.4 ± 1.5 | 49.6 ± 1.6 | 50.3 ± 1.4 | 49.5 ± 1.6 |
| 100 | 50.6 ± 2.1 | 50.0 ± 1.9 | 50.5 ± 1.6 | 49.9 ± 1.4 |
| 125 | 53.1 ± 4.1 | 48.8 ± 2.3 | 51.9 ± 4.7 | 48.9 ± 2.3 |
| 150 | 57.3 ± 15.5 | 48.5 ± 8.2 | 56.3 ± 15.5 | 47.8 ± 8.4 |
| 175 | 62.3 ± 17.3 | 50.2 ± 7.3 | 61.5 ± 17.3 | 50.5 ± 7.4 |
| 200 | 74.1 ± 22.1 | 51.8 ± 13.0 | 73.5 ± 22.0 | 51.7 ± 13.8 |
| 225 | 79.5 ± 22.4 | 59.2 ± 20.7 | 79.0 ± 22.6 | 58.3 ± 19.8 |
| 250 | 84.8 ± 21.0 | 65.3 ± 26.9 | 83.9 ± 21.5 | 64.2 ± 26.3 |
| 275 | 87.5 ± 19.9 | 76.0 ± 23.4 | 86.8 ± 19.8 | 74.7 ± 23.0 |
| 300 | 90.2 ± 16.3 | 78.4 ± 22.4 | 89.0 ± 16.8 | 76.6 ± 21.8 |
| 325 | 95.2 ± 9.0 | 87.8 ± 14.6 | 93.8 ± 9.4 | 85.3 ± 15.0 |
| 350 | 96.5 ± 6.2 | 92.6 ± 9.5 | 93.9 ± 7.6 | 89.2 ± 10.6 |