free-d41b355f
done · peak unseen_mismatch 82.7
α schedule
α 4.217 ×20 α 1.735 ×12 α 4.668 ×24 α 2.304 ×10 α 2.413 ×36 α 1.163 ×22 α 3.003 ×31 α 2.01 ×13 α 0.7634 ×16 α 1.572 ×18 α 2.216 ×28 α 2.395 ×14 α 4.871 ×10 α 3.467 ×46 α 3.467 ×100
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.467
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 |
|---|---|---|---|---|
| 100 | 55.9 ± 6.6 | 51.7 ± 2.8 | 53.2 ± 5.9 | 50.1 ± 2.1 |
| 110 | 58.3 ± 8.4 | 52.6 ± 4.3 | 55.4 ± 8.2 | 51.0 ± 2.9 |
| 120 | 61.7 ± 12.5 | 54.4 ± 7.3 | 59.0 ± 12.0 | 51.9 ± 6.1 |
| 130 | 68.0 ± 16.4 | 56.8 ± 9.5 | 65.4 ± 16.7 | 54.2 ± 8.2 |
| 140 | 73.7 ± 17.4 | 62.9 ± 13.6 | 70.2 ± 18.3 | 59.1 ± 12.4 |
| 150 | 78.9 ± 17.5 | 68.6 ± 14.3 | 75.1 ± 18.6 | 64.0 ± 13.6 |
| 160 | 82.5 ± 16.6 | 73.5 ± 15.0 | 78.4 ± 18.3 | 68.8 ± 14.6 |
| 170 | 85.8 ± 16.0 | 77.4 ± 14.8 | 82.1 ± 17.4 | 72.6 ± 14.9 |
| 180 | 88.9 ± 15.6 | 77.9 ± 15.2 | 86.4 ± 17.2 | 74.3 ± 15.2 |
| 190 | 91.2 ± 15.8 | 74.5 ± 17.1 | 89.9 ± 17.2 | 71.9 ± 16.4 |
| 200 | 92.4 ± 14.3 | 76.4 ± 18.6 | 90.6 ± 17.0 | 73.6 ± 18.1 |
| 210 | 93.1 ± 13.5 | 78.7 ± 16.6 | 91.0 ± 15.4 | 75.7 ± 16.5 |
| 220 | 93.0 ± 14.1 | 81.0 ± 18.3 | 91.1 ± 14.9 | 78.0 ± 18.0 |
| 230 | 91.0 ± 15.2 | 80.2 ± 17.0 | 88.7 ± 15.2 | 76.2 ± 16.7 |
| 240 | 91.1 ± 14.1 | 78.0 ± 16.4 | 87.7 ± 15.1 | 73.5 ± 14.7 |
| 250 | 92.5 ± 10.2 | 85.2 ± 11.5 | 87.7 ± 12.5 | 78.9 ± 10.7 |
| 260 | 92.3 ± 9.8 | 88.9 ± 9.4 | 87.0 ± 11.8 | 82.7 ± 9.9 |
| 270 | 91.5 ± 7.4 | 88.9 ± 7.1 | 84.3 ± 9.6 | 81.2 ± 6.8 |
| 280 | 93.4 ± 5.0 | 89.7 ± 4.6 | 86.4 ± 8.7 | 81.6 ± 6.4 |
| 290 | 93.2 ± 3.7 | 91.1 ± 2.7 | 85.0 ± 7.6 | 81.4 ± 5.9 |
| 300 | 93.1 ± 2.1 | 90.4 ± 2.5 | 85.1 ± 5.5 | 81.0 ± 4.9 |
| 310 | 92.8 ± 3.3 | 90.9 ± 2.5 | 83.3 ± 7.9 | 80.6 ± 6.4 |
| 320 | 92.5 ± 1.9 | 90.4 ± 2.0 | 81.6 ± 5.6 | 78.3 ± 4.5 |
| 330 | 92.3 ± 1.6 | 90.2 ± 2.1 | 81.5 ± 5.9 | 78.1 ± 5.2 |
| 340 | 92.2 ± 1.4 | 90.3 ± 1.5 | 79.8 ± 5.3 | 77.0 ± 5.2 |
| 350 | 91.5 ± 1.6 | 89.4 ± 2.4 | 79.0 ± 4.5 | 75.9 ± 5.0 |
| 360 | 91.8 ± 1.5 | 90.1 ± 2.2 | 79.2 ± 4.4 | 75.7 ± 4.0 |
| 370 | 91.0 ± 2.0 | 89.5 ± 1.4 | 77.8 ± 4.8 | 74.6 ± 4.0 |
| 380 | 91.2 ± 1.5 | 89.7 ± 1.6 | 78.1 ± 4.4 | 74.8 ± 3.9 |
| 390 | 91.1 ± 1.4 | 89.6 ± 2.0 | 78.3 ± 4.8 | 75.2 ± 4.6 |
| 400 | 90.9 ± 1.4 | 89.5 ± 1.6 | 77.8 ± 4.3 | 73.7 ± 3.2 |