down_floor
done · peak unseen_mismatch 89.1
α schedule
α 4.8844 ×25 α 4.6531 ×25 α 4.4219 ×25 α 4.1906 ×25 α 3.9594 ×25 α 3.7281 ×25 α 3.4969 ×25 α 3.2656 ×25 α 3.0344 ×25 α 2.8031 ×25 α 2.5719 ×25 α 2.3406 ×25 α 2.1094 ×25 α 1.8781 ×25 α 1.6469 ×25 α 1.4156 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.4156
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 | 51.4 ± 1.9 | 50.1 ± 1.6 | 50.1 ± 2.1 | 50.1 ± 1.6 |
| 50 | 51.9 ± 2.0 | 50.7 ± 1.2 | 50.5 ± 2.0 | 50.0 ± 2.0 |
| 75 | 53.6 ± 2.7 | 51.1 ± 2.2 | 50.5 ± 2.5 | 49.2 ± 1.4 |
| 100 | 63.8 ± 7.7 | 59.4 ± 6.5 | 55.5 ± 6.0 | 52.8 ± 3.5 |
| 125 | 76.1 ± 9.2 | 72.7 ± 8.5 | 63.5 ± 9.6 | 60.0 ± 7.7 |
| 150 | 81.5 ± 4.1 | 78.3 ± 4.4 | 65.5 ± 6.2 | 62.8 ± 5.2 |
| 175 | 82.9 ± 3.7 | 80.2 ± 3.4 | 65.8 ± 6.1 | 63.6 ± 4.7 |
| 200 | 85.1 ± 3.0 | 82.7 ± 2.8 | 67.5 ± 5.2 | 64.9 ± 4.2 |
| 225 | 87.0 ± 2.7 | 85.4 ± 2.4 | 69.9 ± 5.2 | 67.4 ± 4.6 |
| 250 | 88.7 ± 2.5 | 86.8 ± 2.2 | 71.8 ± 5.1 | 69.5 ± 4.3 |
| 275 | 91.2 ± 2.0 | 89.6 ± 1.6 | 76.6 ± 5.0 | 73.5 ± 4.4 |
| 300 | 93.0 ± 1.8 | 91.3 ± 2.0 | 78.4 ± 5.3 | 75.2 ± 5.8 |
| 325 | 95.2 ± 1.4 | 93.7 ± 1.6 | 83.2 ± 4.2 | 79.1 ± 4.4 |
| 350 | 96.6 ± 1.1 | 95.1 ± 1.4 | 85.8 ± 3.8 | 81.7 ± 5.0 |
| 375 | 97.9 ± 0.9 | 96.3 ± 1.4 | 88.6 ± 2.7 | 84.1 ± 4.1 |
| 400 | 99.0 ± 0.7 | 98.1 ± 1.1 | 93.4 ± 3.6 | 89.1 ± 5.1 |