duty_0.25
done · peak unseen_mismatch 96.5
α schedule
α 3.0 ×19 α 3.0 ×19 α 3.0 ×19 α 3.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 2.1429 ×19 α 2.1429 ×19 α 2.1429 ×19 α 2.1429 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.0 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18 α 1.0 ×18
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.0
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 ± 1.5 | 49.5 ± 1.6 | 49.8 ± 1.9 | 49.7 ± 2.1 |
| 50 | 50.7 ± 1.9 | 49.6 ± 1.4 | 50.1 ± 1.5 | 50.8 ± 1.1 |
| 75 | 51.9 ± 1.8 | 50.1 ± 1.9 | 50.5 ± 1.7 | 50.0 ± 1.9 |
| 100 | 51.5 ± 1.8 | 50.5 ± 1.9 | 50.8 ± 1.9 | 49.5 ± 1.3 |
| 125 | 56.3 ± 7.1 | 50.3 ± 3.5 | 54.8 ± 7.1 | 49.7 ± 2.5 |
| 150 | 71.9 ± 19.1 | 52.3 ± 10.9 | 70.2 ± 19.8 | 52.0 ± 9.9 |
| 175 | 86.8 ± 18.1 | 49.1 ± 18.0 | 86.0 ± 18.9 | 49.5 ± 17.4 |
| 200 | 93.4 ± 14.7 | 45.1 ± 26.2 | 93.1 ± 15.2 | 44.8 ± 25.8 |
| 225 | 97.4 ± 11.4 | 41.5 ± 26.6 | 97.4 ± 10.9 | 42.0 ± 26.4 |
| 250 | 97.4 ± 11.2 | 42.1 ± 29.0 | 97.6 ± 10.5 | 42.0 ± 28.5 |
| 275 | 97.6 ± 10.6 | 41.4 ± 34.3 | 97.5 ± 10.9 | 42.1 ± 34.1 |
| 300 | 97.5 ± 10.7 | 41.7 ± 31.7 | 97.6 ± 10.5 | 41.8 ± 31.6 |
| 325 | 98.8 ± 5.1 | 46.3 ± 33.0 | 98.8 ± 5.3 | 46.2 ± 33.4 |
| 350 | 100.0 ± 0.0 | 62.0 ± 33.3 | 100.0 ± 0.0 | 61.7 ± 33.3 |
| 375 | 100.0 ± 0.0 | 71.8 ± 33.8 | 99.9 ± 0.3 | 71.7 ± 33.6 |
| 400 | 100.0 ± 0.0 | 73.4 ± 30.8 | 100.0 ± 0.0 | 73.2 ± 30.9 |
| 450 | 100.0 ± 0.0 | 73.2 ± 37.1 | 100.0 ± 0.0 | 73.3 ± 36.6 |
| 500 | 100.0 ± 0.0 | 78.8 ± 29.4 | 100.0 ± 0.0 | 78.1 ± 29.1 |
| 550 | 100.0 ± 0.0 | 85.2 ± 20.1 | 100.0 ± 0.0 | 84.2 ± 20.8 |
| 600 | 100.0 ± 0.0 | 91.1 ± 17.8 | 99.9 ± 0.2 | 89.8 ± 18.9 |
| 650 | 100.0 ± 0.0 | 90.2 ± 15.6 | 99.8 ± 0.5 | 88.2 ± 16.2 |
| 700 | 100.0 ± 0.0 | 96.2 ± 7.2 | 99.7 ± 1.2 | 94.3 ± 8.1 |
| 750 | 100.0 ± 0.0 | 97.2 ± 5.7 | 99.9 ± 0.2 | 95.8 ± 6.2 |
| 800 | 100.0 ± 0.0 | 98.6 ± 2.2 | 99.7 ± 0.7 | 96.5 ± 4.3 |
| 850 | 100.0 ± 0.1 | 97.3 ± 5.0 | 99.7 ± 0.6 | 95.4 ± 5.1 |
| 900 | 100.0 ± 0.1 | 98.6 ± 2.7 | 99.7 ± 0.7 | 96.3 ± 4.9 |