burst_48
done · peak unseen_mismatch 90.1
α schedule
α 5.0 ×9 α 5.0 ×9 α 5.0 ×9 α 5.0 ×9 α 5.0 ×9 α 5.0 ×9 α 5.0 ×9 α 5.0 ×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 3.1429 ×8 α 3.1429 ×8 α 3.1429 ×8 α 3.1429 ×8 α 3.1429 ×8 α 3.1429 ×8 α 3.1429 ×8 α 3.1429 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 1.2857 ×8 α 1.2857 ×8 α 1.2857 ×8 α 1.2857 ×8 α 1.2857 ×8 α 1.2857 ×8 α 1.2857 ×8 α 1.2857 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8 α 0.7 ×8
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 | 51.4 ± 1.7 | 49.9 ± 1.5 | 49.7 ± 1.5 | 49.7 ± 1.6 |
| 50 | 52.0 ± 1.8 | 50.5 ± 1.9 | 50.1 ± 1.7 | 49.7 ± 1.5 |
| 75 | 53.1 ± 3.3 | 51.5 ± 2.3 | 51.0 ± 1.6 | 49.6 ± 1.5 |
| 100 | 55.1 ± 5.4 | 50.9 ± 1.9 | 52.9 ± 4.7 | 49.6 ± 1.5 |
| 125 | 66.8 ± 16.7 | 52.8 ± 7.3 | 65.1 ± 17.3 | 51.8 ± 6.5 |
| 150 | 80.6 ± 20.6 | 50.9 ± 16.1 | 80.2 ± 20.5 | 49.9 ± 15.9 |
| 175 | 91.1 ± 15.6 | 68.9 ± 21.6 | 90.3 ± 16.5 | 67.7 ± 21.1 |
| 200 | 96.4 ± 8.7 | 86.0 ± 17.6 | 95.1 ± 9.8 | 84.2 ± 17.4 |
| 225 | 98.0 ± 9.3 | 90.1 ± 15.2 | 97.6 ± 10.3 | 89.2 ± 15.3 |
| 250 | 98.2 ± 8.6 | 87.3 ± 20.1 | 98.2 ± 8.7 | 86.7 ± 20.4 |
| 275 | 99.0 ± 4.9 | 82.7 ± 27.8 | 98.9 ± 5.6 | 82.8 ± 27.5 |
| 300 | 100.0 ± 0.1 | 86.4 ± 23.1 | 100.0 ± 0.0 | 86.0 ± 23.2 |
| 325 | 100.0 ± 0.1 | 91.0 ± 15.8 | 99.9 ± 0.2 | 90.1 ± 16.3 |
| 350 | 100.0 ± 0.0 | 87.1 ± 21.4 | 100.0 ± 0.0 | 86.7 ± 21.6 |
| 375 | 100.0 ± 0.0 | 88.6 ± 22.5 | 100.0 ± 0.0 | 87.8 ± 22.4 |
| 400 | 100.0 ± 0.0 | 86.9 ± 24.3 | 100.0 ± 0.0 | 86.4 ± 24.2 |