r2_c0
done · peak unseen_mismatch 97.3
α schedule
α 3.3144 ×25 α 3.3084 ×25 α 3.2908 ×25 α 3.2577 ×25 α 3.2058 ×25 α 3.1322 ×25 α 3.0342 ×25 α 2.9092 ×25 α 2.7552 ×25 α 2.5698 ×25 α 2.351 ×25 α 2.0969 ×25 α 1.8056 ×25 α 1.4751 ×25 α 1.1039 ×25 α 0.69 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.69
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 | 50.2 ± 1.7 | 50.0 ± 1.6 | 49.5 ± 1.7 |
| 50 | 51.6 ± 1.4 | 50.4 ± 1.5 | 50.2 ± 1.1 | 49.9 ± 1.7 |
| 75 | 52.3 ± 1.7 | 50.1 ± 1.6 | 50.5 ± 1.5 | 49.4 ± 1.5 |
| 100 | 56.2 ± 6.6 | 54.0 ± 3.9 | 52.4 ± 4.1 | 50.2 ± 2.2 |
| 125 | 66.8 ± 12.4 | 63.5 ± 10.5 | 58.2 ± 11.1 | 55.6 ± 8.8 |
| 150 | 77.1 ± 10.5 | 74.3 ± 10.3 | 65.5 ± 11.8 | 63.1 ± 11.0 |
| 175 | 84.6 ± 5.5 | 81.9 ± 5.4 | 71.5 ± 10.0 | 68.4 ± 8.9 |
| 200 | 88.0 ± 3.7 | 85.6 ± 3.5 | 74.2 ± 8.5 | 71.6 ± 7.7 |
| 225 | 89.7 ± 3.3 | 87.4 ± 3.2 | 75.4 ± 8.0 | 72.8 ± 7.9 |
| 250 | 91.0 ± 2.1 | 88.8 ± 1.9 | 76.3 ± 5.9 | 73.6 ± 5.5 |
| 275 | 93.2 ± 1.9 | 91.3 ± 1.8 | 79.5 ± 5.1 | 77.1 ± 4.7 |
| 300 | 94.7 ± 1.5 | 92.9 ± 1.6 | 82.4 ± 4.4 | 78.7 ± 4.5 |
| 325 | 96.5 ± 1.1 | 95.1 ± 1.4 | 86.2 ± 4.2 | 82.5 ± 5.1 |
| 350 | 98.7 ± 0.8 | 97.6 ± 1.3 | 92.8 ± 3.7 | 89.4 ± 4.3 |
| 375 | 99.7 ± 0.3 | 99.2 ± 0.6 | 98.3 ± 1.1 | 95.1 ± 3.7 |
| 400 | 100.0 ± 0.0 | 99.4 ± 1.1 | 99.8 ± 0.3 | 97.3 ± 4.1 |