r2_c5
done · peak unseen_mismatch 91.2
α schedule
α 1.5358 ×25 α 1.5359 ×25 α 1.5365 ×25 α 1.5383 ×25 α 1.5425 ×25 α 1.5505 ×25 α 1.5639 ×25 α 1.5849 ×25 α 1.6157 ×25 α 1.659 ×25 α 1.7178 ×25 α 1.7952 ×25 α 1.8947 ×25 α 2.0202 ×25 α 2.1758 ×25 α 2.3656 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.3656
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 ± 1.6 | 50.2 ± 1.4 | 49.8 ± 1.5 | 49.4 ± 2.1 |
| 50 | 50.6 ± 1.8 | 50.0 ± 1.6 | 50.2 ± 1.5 | 49.8 ± 1.8 |
| 75 | 50.4 ± 1.4 | 49.5 ± 1.7 | 50.3 ± 1.2 | 49.4 ± 1.7 |
| 100 | 50.2 ± 2.2 | 50.0 ± 1.5 | 50.2 ± 1.6 | 50.0 ± 1.5 |
| 125 | 51.9 ± 2.5 | 49.0 ± 2.0 | 51.0 ± 2.8 | 48.9 ± 2.4 |
| 150 | 55.2 ± 10.4 | 49.8 ± 3.7 | 54.3 ± 10.3 | 48.7 ± 4.4 |
| 175 | 62.8 ± 17.2 | 48.3 ± 8.1 | 62.0 ± 17.4 | 48.4 ± 7.6 |
| 200 | 72.6 ± 22.8 | 51.5 ± 14.7 | 72.4 ± 22.8 | 51.1 ± 13.6 |
| 225 | 76.9 ± 23.9 | 52.5 ± 17.0 | 76.9 ± 23.7 | 52.5 ± 16.2 |
| 250 | 78.9 ± 22.7 | 53.5 ± 19.2 | 78.4 ± 23.0 | 53.1 ± 18.8 |
| 275 | 85.5 ± 20.5 | 66.7 ± 21.5 | 84.7 ± 21.2 | 65.4 ± 20.7 |
| 300 | 91.4 ± 15.5 | 71.5 ± 24.5 | 91.0 ± 15.8 | 70.5 ± 24.0 |
| 325 | 96.7 ± 10.1 | 80.5 ± 23.0 | 96.4 ± 10.9 | 79.5 ± 23.2 |
| 350 | 98.8 ± 5.1 | 85.5 ± 18.7 | 98.5 ± 6.0 | 84.8 ± 18.2 |
| 375 | 99.7 ± 0.5 | 89.1 ± 16.4 | 99.4 ± 1.0 | 88.3 ± 15.1 |
| 400 | 99.4 ± 0.7 | 92.8 ± 11.3 | 98.7 ± 1.7 | 91.2 ± 10.5 |