r8_c6
done · peak unseen_mismatch 62.3
α schedule
α 2.6106 ×25 α 2.6099 ×25 α 2.6066 ×25 α 0.2255 ×25 α 0.2238 ×25 α 2.5445 ×25 α 2.4912 ×25 α 2.4125 ×25 α 0.1998 ×25 α 0.1869 ×25 α 0.17 ×25 α 1.7112 ×25 α 1.4025 ×25 α 0.0889 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.0889
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 ± 1.3 | 50.6 ± 1.6 | 50.1 ± 1.5 | 49.3 ± 1.6 |
| 50 | 50.9 ± 1.5 | 50.3 ± 1.7 | 49.8 ± 1.1 | 50.0 ± 1.8 |
| 75 | 51.8 ± 1.6 | 49.9 ± 1.8 | 50.3 ± 1.3 | 49.3 ± 1.4 |
| 100 | 51.6 ± 3.7 | 49.8 ± 2.0 | 51.1 ± 3.2 | 49.9 ± 1.7 |
| 125 | 56.0 ± 10.9 | 47.7 ± 6.5 | 54.4 ± 11.2 | 47.2 ± 6.6 |
| 150 | 61.9 ± 16.9 | 47.1 ± 7.7 | 61.0 ± 17.4 | 46.6 ± 7.1 |
| 175 | 72.7 ± 20.2 | 52.1 ± 9.8 | 72.2 ± 20.4 | 51.2 ± 8.9 |
| 200 | 81.6 ± 20.1 | 63.6 ± 16.9 | 80.5 ± 20.5 | 62.3 ± 15.4 |
| 225 | 88.8 ± 17.7 | 58.0 ± 22.8 | 88.6 ± 17.7 | 58.2 ± 22.3 |
| 250 | 95.8 ± 11.5 | 45.1 ± 29.8 | 95.4 ± 12.5 | 45.2 ± 29.5 |
| 275 | 98.1 ± 8.9 | 41.0 ± 26.5 | 98.2 ± 8.4 | 40.8 ± 26.5 |
| 300 | 100.0 ± 0.1 | 44.2 ± 28.4 | 100.0 ± 0.1 | 44.5 ± 28.1 |
| 325 | 100.0 ± 0.0 | 47.7 ± 31.1 | 100.0 ± 0.0 | 48.2 ± 30.6 |
| 350 | 100.0 ± 0.0 | 46.4 ± 26.3 | 100.0 ± 0.0 | 46.6 ± 25.5 |