start_4
done · peak unseen_mismatch 96.9
α schedule
α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 4.0 ×19 α 4.0 ×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 2.6429 ×19 α 2.6429 ×19 α 2.6429 ×19 α 2.6429 ×19 α 2.6429 ×19 α 2.6429 ×19 α 2.6429 ×19 α 2.6429 ×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 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×18 α 1.2857 ×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 | 51.5 ± 1.6 | 49.7 ± 1.6 | 49.9 ± 2.0 | 49.7 ± 2.1 |
| 50 | 51.1 ± 1.7 | 49.9 ± 1.3 | 50.1 ± 1.5 | 50.4 ± 1.5 |
| 75 | 52.9 ± 2.1 | 50.9 ± 1.9 | 50.6 ± 1.7 | 50.0 ± 2.1 |
| 100 | 58.7 ± 5.5 | 55.5 ± 4.2 | 52.5 ± 4.0 | 51.1 ± 2.2 |
| 125 | 72.0 ± 10.1 | 67.7 ± 9.2 | 61.9 ± 9.3 | 57.9 ± 7.3 |
| 150 | 80.1 ± 7.8 | 77.2 ± 8.1 | 68.6 ± 9.4 | 64.3 ± 7.9 |
| 175 | 94.0 ± 6.5 | 89.8 ± 8.7 | 90.0 ± 9.2 | 84.2 ± 10.7 |
| 200 | 99.4 ± 0.7 | 93.6 ± 12.9 | 98.7 ± 1.7 | 91.4 ± 13.5 |
| 225 | 99.9 ± 0.2 | 93.1 ± 16.7 | 99.5 ± 0.7 | 91.4 ± 16.6 |
| 250 | 99.9 ± 0.1 | 93.7 ± 19.9 | 99.7 ± 0.3 | 92.2 ± 19.1 |
| 275 | 99.9 ± 0.2 | 93.3 ± 18.4 | 99.6 ± 0.6 | 91.3 ± 18.5 |
| 300 | 100.0 ± 0.1 | 95.7 ± 14.8 | 99.7 ± 0.5 | 93.7 ± 15.0 |
| 325 | 97.2 ± 1.5 | 93.5 ± 4.7 | 91.6 ± 4.7 | 84.8 ± 6.4 |
| 350 | 96.5 ± 1.4 | 94.2 ± 2.3 | 90.1 ± 4.1 | 84.8 ± 5.5 |
| 375 | 96.0 ± 1.2 | 94.1 ± 2.0 | 89.7 ± 3.2 | 84.4 ± 4.9 |
| 400 | 95.4 ± 1.5 | 93.6 ± 2.0 | 88.0 ± 3.8 | 81.8 ± 5.3 |
| 450 | 95.1 ± 1.3 | 93.2 ± 1.5 | 85.2 ± 3.6 | 80.1 ± 4.9 |
| 500 | 99.9 ± 0.1 | 99.5 ± 0.6 | 99.2 ± 0.9 | 96.9 ± 3.1 |
| 550 | 99.9 ± 0.1 | 99.2 ± 1.5 | 99.1 ± 1.1 | 94.7 ± 6.6 |
| 600 | 100.0 ± 0.1 | 99.2 ± 1.5 | 99.2 ± 0.9 | 93.3 ± 8.5 |
| 650 | 99.8 ± 0.2 | 98.7 ± 1.0 | 97.2 ± 1.6 | 89.6 ± 6.7 |
| 700 | 99.8 ± 0.2 | 98.5 ± 1.2 | 97.0 ± 1.6 | 88.8 ± 6.6 |
| 750 | 99.9 ± 0.1 | 98.5 ± 1.0 | 97.0 ± 1.8 | 88.2 ± 6.9 |
| 800 | 100.0 ± 0.1 | 99.0 ± 1.8 | 99.0 ± 1.1 | 91.2 ± 8.2 |
| 850 | 100.0 ± 0.1 | 99.3 ± 0.7 | 99.0 ± 1.2 | 91.9 ± 5.9 |
| 900 | 100.0 ± 0.1 | 99.0 ± 0.9 | 98.8 ± 0.9 | 89.8 ± 6.5 |