incumbent
done · peak unseen_mismatch 99.6
α schedule
α 5.0 ×19 α 5.0 ×19 α 5.0 ×19 α 5.0 ×19 α 5.0 ×19 α 5.0 ×19 α 5.0 ×19 α 5.0 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 3.1429 ×19 α 3.1429 ×19 α 3.1429 ×19 α 3.1429 ×19 α 3.1429 ×19 α 3.1429 ×19 α 3.1429 ×19 α 3.1429 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 0.7 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.2857 ×19 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 0.7 ×18 α 0.7 ×18 α 0.7 ×18 α 0.7 ×18 α 0.7 ×18 α 0.7 ×18 α 0.7 ×18 α 0.7 ×18
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 ± 1.6 | 50.5 ± 2.0 | 50.2 ± 2.0 | 49.7 ± 1.3 |
| 50 | 51.6 ± 1.7 | 50.9 ± 1.7 | 49.7 ± 1.9 | 49.5 ± 1.7 |
| 75 | 53.3 ± 3.1 | 51.6 ± 2.0 | 50.8 ± 2.1 | 50.6 ± 1.6 |
| 100 | 63.7 ± 8.4 | 60.1 ± 7.2 | 55.4 ± 6.4 | 52.4 ± 3.7 |
| 125 | 76.1 ± 7.4 | 72.7 ± 7.5 | 62.0 ± 8.3 | 58.5 ± 6.3 |
| 150 | 79.0 ± 3.9 | 75.8 ± 5.0 | 63.0 ± 5.8 | 59.9 ± 4.2 |
| 175 | 92.8 ± 7.1 | 90.2 ± 7.2 | 86.8 ± 11.0 | 83.7 ± 10.3 |
| 200 | 99.4 ± 2.0 | 97.1 ± 5.4 | 98.7 ± 3.2 | 95.9 ± 6.9 |
| 225 | 100.0 ± 0.1 | 96.7 ± 11.2 | 100.0 ± 0.1 | 96.4 ± 10.5 |
| 250 | 100.0 ± 0.0 | 96.5 ± 11.6 | 100.0 ± 0.0 | 96.2 ± 11.2 |
| 275 | 100.0 ± 0.0 | 98.0 ± 8.0 | 100.0 ± 0.1 | 97.8 ± 8.0 |
| 300 | 100.0 ± 0.0 | 98.7 ± 5.2 | 100.0 ± 0.0 | 98.4 ± 5.7 |
| 325 | 97.0 ± 1.8 | 93.4 ± 4.8 | 90.3 ± 4.9 | 82.5 ± 7.3 |
| 350 | 95.7 ± 1.7 | 92.7 ± 2.4 | 88.8 ± 3.8 | 82.0 ± 5.9 |
| 375 | 95.3 ± 1.6 | 92.5 ± 2.1 | 87.0 ± 4.5 | 80.8 ± 5.4 |
| 400 | 94.7 ± 1.4 | 91.9 ± 1.8 | 84.9 ± 4.2 | 78.8 ± 4.5 |
| 450 | 93.7 ± 1.3 | 91.4 ± 1.5 | 82.5 ± 3.6 | 76.7 ± 3.5 |
| 500 | 100.0 ± 0.0 | 99.9 ± 0.1 | 100.0 ± 0.1 | 99.6 ± 0.7 |
| 550 | 100.0 ± 0.0 | 99.8 ± 0.5 | 99.9 ± 0.1 | 98.9 ± 3.0 |
| 600 | 100.0 ± 0.0 | 99.8 ± 0.4 | 99.8 ± 0.4 | 98.3 ± 2.1 |
| 650 | 99.7 ± 0.2 | 98.8 ± 1.1 | 97.3 ± 1.7 | 92.0 ± 4.6 |
| 700 | 99.7 ± 0.2 | 98.6 ± 0.9 | 96.8 ± 2.2 | 90.5 ± 4.8 |
| 750 | 99.8 ± 0.2 | 98.6 ± 1.0 | 97.2 ± 1.7 | 90.3 ± 5.6 |
| 800 | 100.0 ± 0.0 | 99.2 ± 1.8 | 99.9 ± 0.3 | 95.9 ± 5.6 |
| 850 | 100.0 ± 0.0 | 96.0 ± 8.8 | 99.8 ± 0.4 | 89.1 ± 13.3 |
| 900 | 100.0 ± 0.0 | 94.0 ± 8.7 | 99.8 ± 0.4 | 85.2 ± 14.0 |