r5_c1
done · peak unseen_mismatch 93.4
α schedule
α 1.62 ×25 α 1.6391 ×25 α 1.6652 ×25 α 1.6963 ×25 α 1.7316 ×25 α 1.7706 ×25 α 1.8127 ×25 α 1.8578 ×25 α 1.9056 ×25 α 1.956 ×25 α 2.0087 ×25 α 2.0636 ×25 α 2.1207 ×25 α 2.1797 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.1797
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 | 50.5 ± 1.5 | 50.2 ± 1.5 | 50.0 ± 1.6 | 49.3 ± 1.9 |
| 50 | 50.7 ± 1.6 | 49.9 ± 1.4 | 50.2 ± 1.8 | 49.8 ± 1.5 |
| 75 | 50.8 ± 1.6 | 49.7 ± 1.7 | 50.2 ± 1.4 | 49.8 ± 1.8 |
| 100 | 50.4 ± 2.0 | 50.1 ± 1.2 | 50.0 ± 1.2 | 49.9 ± 1.5 |
| 125 | 52.0 ± 2.7 | 49.4 ± 1.7 | 51.0 ± 3.0 | 49.3 ± 1.9 |
| 150 | 54.7 ± 9.0 | 50.2 ± 2.3 | 53.6 ± 8.8 | 49.2 ± 2.1 |
| 175 | 64.2 ± 17.7 | 51.5 ± 7.4 | 63.3 ± 17.4 | 50.5 ± 6.6 |
| 200 | 76.7 ± 20.5 | 58.2 ± 16.6 | 76.3 ± 20.2 | 56.8 ± 15.5 |
| 225 | 84.8 ± 21.0 | 67.7 ± 20.3 | 83.9 ± 21.7 | 66.9 ± 20.2 |
| 250 | 89.1 ± 18.8 | 73.5 ± 25.0 | 88.0 ± 19.9 | 73.0 ± 24.5 |
| 275 | 91.2 ± 17.0 | 82.7 ± 18.8 | 90.4 ± 17.4 | 81.0 ± 18.8 |
| 300 | 94.6 ± 13.0 | 84.8 ± 20.5 | 93.6 ± 13.4 | 82.5 ± 20.2 |
| 325 | 98.0 ± 5.0 | 92.1 ± 10.7 | 96.7 ± 5.2 | 89.6 ± 10.7 |
| 350 | 99.1 ± 1.1 | 95.5 ± 5.5 | 97.5 ± 3.1 | 93.4 ± 6.1 |