r6_c1
done · peak unseen_mismatch 83.7
α schedule
α 1.3762 ×25 α 1.3767 ×25 α 1.3796 ×25 α 1.342 ×25 α 1.359 ×25 α 1.4372 ×25 α 1.4886 ×25 α 1.5659 ×25 α 1.6207 ×25 α 1.7662 ×25 α 1.9588 ×25 α 2.282 ×25 α 2.6051 ×25 α 2.9054 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.9054
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 | 50.6 ± 1.2 | 50.2 ± 1.5 | 49.7 ± 1.2 | 49.5 ± 1.9 |
| 50 | 50.4 ± 1.5 | 50.1 ± 1.6 | 49.8 ± 1.3 | 49.9 ± 1.7 |
| 75 | 50.7 ± 1.5 | 49.7 ± 1.9 | 50.5 ± 1.5 | 49.6 ± 1.6 |
| 100 | 50.4 ± 1.9 | 49.9 ± 1.5 | 50.5 ± 1.3 | 50.1 ± 1.5 |
| 125 | 52.0 ± 3.1 | 49.2 ± 2.0 | 51.0 ± 3.8 | 49.3 ± 2.4 |
| 150 | 56.2 ± 12.8 | 48.2 ± 5.1 | 55.3 ± 12.4 | 47.7 ± 4.6 |
| 175 | 60.4 ± 16.2 | 47.8 ± 8.6 | 60.2 ± 16.5 | 47.8 ± 7.6 |
| 200 | 69.7 ± 21.3 | 44.2 ± 17.1 | 69.6 ± 21.2 | 43.9 ± 16.8 |
| 225 | 77.8 ± 22.9 | 50.5 ± 17.8 | 77.9 ± 22.6 | 50.9 ± 17.9 |
| 250 | 84.3 ± 21.0 | 55.3 ± 22.5 | 83.8 ± 21.2 | 55.5 ± 22.7 |
| 275 | 89.1 ± 19.2 | 59.7 ± 24.1 | 88.6 ± 19.8 | 59.9 ± 23.7 |
| 300 | 92.7 ± 16.3 | 69.0 ± 26.7 | 92.5 ± 16.4 | 68.2 ± 26.2 |
| 325 | 94.9 ± 13.8 | 77.9 ± 27.0 | 94.4 ± 14.2 | 77.2 ± 26.6 |
| 350 | 96.3 ± 10.8 | 84.9 ± 22.8 | 95.4 ± 11.5 | 83.7 ± 22.7 |