p05
done · peak unseen_mismatch 50.1
α schedule
α 1.1174 ×25 α 0.9279 ×25 α 0.8151 ×25 α 0.7304 ×25 α 0.661 ×25 α 0.6016 ×25 α 0.5493 ×25 α 0.5022 ×25 α 0.4592 ×25 α 0.4196 ×25 α 0.3828 ×25 α 0.3483 ×25 α 0.3158 ×25 α 0.2851 ×25 α 0.2559 ×25 α 0.2281 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.2281
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 | 50.0 ± 1.6 | 49.7 ± 1.7 | 49.6 ± 1.5 |
| 50 | 50.7 ± 1.6 | 49.9 ± 1.7 | 50.0 ± 1.5 | 49.9 ± 1.5 |
| 75 | 50.6 ± 1.4 | 49.5 ± 1.8 | 50.2 ± 1.3 | 49.6 ± 1.5 |
| 100 | 50.1 ± 2.1 | 49.7 ± 1.3 | 50.0 ± 1.3 | 50.1 ± 1.2 |
| 125 | 51.2 ± 1.8 | 49.4 ± 1.8 | 50.2 ± 2.4 | 49.1 ± 2.2 |
| 150 | 53.0 ± 8.8 | 48.4 ± 8.0 | 52.3 ± 9.0 | 47.6 ± 7.9 |
| 175 | 56.3 ± 12.8 | 45.3 ± 9.8 | 56.4 ± 12.8 | 46.1 ± 10.0 |
| 200 | 62.6 ± 18.9 | 41.4 ± 14.2 | 62.6 ± 18.8 | 42.0 ± 14.4 |
| 225 | 68.4 ± 20.9 | 38.6 ± 15.9 | 68.8 ± 20.8 | 39.5 ± 16.3 |
| 250 | 78.8 ± 23.5 | 32.8 ± 22.1 | 78.8 ± 23.4 | 33.0 ± 21.6 |
| 275 | 82.1 ± 22.3 | 30.9 ± 19.1 | 81.6 ± 22.8 | 30.8 ± 19.3 |
| 300 | 86.9 ± 20.3 | 25.9 ± 18.4 | 86.9 ± 20.3 | 26.2 ± 18.8 |
| 325 | 90.7 ± 18.3 | 23.4 ± 21.1 | 90.6 ± 18.7 | 23.6 ± 21.2 |
| 350 | 93.6 ± 14.6 | 25.6 ± 24.9 | 93.9 ± 14.2 | 26.0 ± 25.0 |
| 375 | 96.0 ± 13.7 | 22.9 ± 21.5 | 96.0 ± 13.4 | 23.3 ± 21.9 |
| 400 | 96.3 ± 12.6 | 29.1 ± 24.2 | 96.5 ± 12.0 | 29.2 ± 24.4 |