p17
done · peak unseen_mismatch 50.0
α schedule
α 0.4024 ×25 α 0.4025 ×25 α 0.4027 ×25 α 0.4031 ×25 α 0.4037 ×25 α 0.4045 ×25 α 0.4056 ×25 α 0.407 ×25 α 0.4086 ×25 α 0.4107 ×25 α 0.413 ×25 α 0.4158 ×25 α 0.4189 ×25 α 0.4224 ×25 α 0.4264 ×25 α 0.4308 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.4308
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 | 50.3 ± 1.6 | 49.7 ± 1.4 | 49.8 ± 1.5 | 49.5 ± 1.5 |
| 50 | 50.6 ± 1.6 | 50.0 ± 1.3 | 49.9 ± 1.2 | 50.0 ± 1.4 |
| 75 | 50.4 ± 1.6 | 49.4 ± 1.9 | 50.4 ± 1.5 | 49.5 ± 1.6 |
| 100 | 50.2 ± 2.1 | 49.9 ± 1.6 | 50.0 ± 1.5 | 49.9 ± 1.4 |
| 125 | 51.3 ± 2.2 | 49.3 ± 1.8 | 50.7 ± 2.7 | 49.2 ± 1.9 |
| 150 | 55.7 ± 11.4 | 47.2 ± 6.7 | 55.0 ± 11.5 | 46.7 ± 6.6 |
| 175 | 59.7 ± 18.1 | 43.0 ± 14.0 | 59.8 ± 17.9 | 43.3 ± 14.0 |
| 200 | 61.9 ± 19.3 | 41.4 ± 17.1 | 62.1 ± 19.2 | 42.2 ± 17.5 |
| 225 | 69.2 ± 21.2 | 37.9 ± 16.5 | 69.8 ± 20.9 | 39.3 ± 17.2 |
| 250 | 79.9 ± 22.9 | 31.2 ± 21.1 | 80.0 ± 22.8 | 31.4 ± 21.0 |
| 275 | 84.1 ± 22.0 | 26.9 ± 22.4 | 83.5 ± 22.6 | 26.8 ± 22.7 |
| 300 | 85.9 ± 21.5 | 23.4 ± 20.2 | 86.1 ± 21.2 | 23.3 ± 20.3 |
| 325 | 90.3 ± 18.8 | 22.9 ± 21.2 | 90.2 ± 19.0 | 23.1 ± 21.4 |
| 350 | 92.1 ± 17.9 | 21.3 ± 22.0 | 91.9 ± 18.1 | 21.5 ± 22.6 |
| 375 | 95.7 ± 12.6 | 22.0 ± 27.1 | 95.6 ± 13.1 | 22.2 ± 27.0 |
| 400 | 97.4 ± 10.0 | 20.7 ± 23.7 | 97.4 ± 10.0 | 21.0 ± 23.8 |