T450
done · peak unseen_mismatch 95.3
α schedule
α 5.0 ×10 α 5.0 ×10 α 5.0 ×10 α 5.0 ×10 α 5.0 ×10 α 5.0 ×10 α 5.0 ×10 α 5.0 ×10 α 0.7 ×10 α 0.7 ×10 α 0.7 ×10 α 0.7 ×10 α 0.7 ×10 α 0.7 ×10 α 0.7 ×10 α 0.7 ×10 α 3.1429 ×10 α 3.1429 ×10 α 3.1429 ×9 α 3.1429 ×9 α 3.1429 ×9 α 3.1429 ×9 α 3.1429 ×9 α 3.1429 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 1.2857 ×9 α 1.2857 ×9 α 1.2857 ×9 α 1.2857 ×9 α 1.2857 ×9 α 1.2857 ×9 α 1.2857 ×9 α 1.2857 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9 α 0.7 ×9
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.7
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 | 51.5 ± 1.5 | 49.9 ± 1.3 | 49.8 ± 1.9 | 49.6 ± 2.2 |
| 50 | 50.9 ± 1.8 | 50.0 ± 1.2 | 50.2 ± 1.6 | 50.4 ± 1.5 |
| 75 | 54.0 ± 1.9 | 51.6 ± 1.9 | 51.1 ± 1.7 | 50.1 ± 2.2 |
| 100 | 56.4 ± 4.9 | 52.4 ± 2.9 | 53.1 ± 3.9 | 49.7 ± 1.6 |
| 125 | 67.7 ± 16.9 | 51.2 ± 6.3 | 65.8 ± 17.3 | 50.7 ± 5.9 |
| 150 | 82.2 ± 18.6 | 45.9 ± 14.6 | 81.2 ± 19.3 | 46.4 ± 14.3 |
| 175 | 95.9 ± 8.3 | 50.3 ± 22.1 | 95.4 ± 9.2 | 49.4 ± 22.0 |
| 200 | 99.3 ± 1.7 | 71.8 ± 25.0 | 99.2 ± 2.0 | 71.0 ± 24.7 |
| 225 | 99.7 ± 0.5 | 89.8 ± 16.7 | 99.6 ± 0.7 | 88.5 ± 16.9 |
| 250 | 100.0 ± 0.0 | 94.1 ± 13.7 | 99.9 ± 0.2 | 93.7 ± 13.8 |
| 275 | 100.0 ± 0.0 | 91.5 ± 15.2 | 100.0 ± 0.0 | 91.3 ± 14.6 |
| 300 | 100.0 ± 0.0 | 89.1 ± 19.8 | 100.0 ± 0.0 | 88.7 ± 19.5 |
| 325 | 100.0 ± 0.0 | 83.2 ± 25.3 | 100.0 ± 0.0 | 83.1 ± 25.2 |
| 350 | 100.0 ± 0.0 | 93.6 ± 13.0 | 100.0 ± 0.0 | 92.7 ± 14.1 |
| 375 | 100.0 ± 0.1 | 95.9 ± 8.0 | 99.9 ± 0.3 | 95.0 ± 9.2 |
| 400 | 100.0 ± 0.0 | 96.1 ± 8.0 | 100.0 ± 0.0 | 95.3 ± 9.2 |
| 450 | 100.0 ± 0.0 | 89.8 ± 18.4 | 100.0 ± 0.0 | 89.1 ± 18.1 |