s1_e1
done · peak unseen_mismatch 89.6
α schedule
α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50 α 1.0 ×50
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 |
|---|---|---|---|---|
| 50 | 50.6 ± 1.3 | 49.1 ± 0.7 | 50.0 ± 2.0 | 50.2 ± 1.2 |
| 100 | 50.4 ± 0.9 | 49.9 ± 1.8 | 50.1 ± 1.0 | 49.7 ± 1.0 |
| 150 | 51.5 ± 5.2 | 48.6 ± 3.3 | 52.0 ± 5.5 | 49.2 ± 2.8 |
| 200 | 63.0 ± 18.5 | 42.5 ± 12.9 | 61.4 ± 19.1 | 42.3 ± 12.7 |
| 250 | 85.6 ± 18.7 | 36.8 ± 12.0 | 84.6 ± 19.3 | 37.6 ± 12.1 |
| 300 | 90.8 ± 18.4 | 30.5 ± 17.3 | 90.9 ± 18.1 | 30.7 ± 17.6 |
| 350 | 100.0 ± 0.0 | 33.5 ± 28.1 | 100.0 ± 0.1 | 33.2 ± 28.3 |
| 400 | 100.0 ± 0.0 | 40.0 ± 20.4 | 100.0 ± 0.0 | 40.4 ± 20.5 |
| 450 | 100.0 ± 0.0 | 37.7 ± 23.5 | 100.0 ± 0.0 | 37.9 ± 24.0 |
| 500 | 100.0 ± 0.0 | 45.3 ± 33.6 | 100.0 ± 0.0 | 44.5 ± 33.2 |
| 550 | 100.0 ± 0.0 | 45.9 ± 26.4 | 100.0 ± 0.0 | 46.2 ± 25.7 |
| 600 | 100.0 ± 0.0 | 50.4 ± 30.9 | 100.0 ± 0.0 | 49.7 ± 31.2 |
| 650 | 100.0 ± 0.0 | 60.5 ± 31.0 | 100.0 ± 0.0 | 60.9 ± 30.1 |
| 700 | 100.0 ± 0.0 | 47.1 ± 31.8 | 100.0 ± 0.0 | 45.9 ± 31.2 |
| 750 | 100.0 ± 0.0 | 58.5 ± 27.6 | 100.0 ± 0.0 | 58.3 ± 27.2 |
| 800 | 100.0 ± 0.0 | 70.2 ± 24.6 | 100.0 ± 0.0 | 69.9 ± 24.5 |
| 850 | 100.0 ± 0.0 | 69.8 ± 34.9 | 100.0 ± 0.0 | 69.5 ± 34.8 |
| 900 | 100.0 ± 0.0 | 69.1 ± 29.7 | 100.0 ± 0.0 | 69.0 ± 29.7 |
| 950 | 100.0 ± 0.0 | 74.9 ± 33.0 | 100.0 ± 0.0 | 74.9 ± 32.9 |
| 1000 | 100.0 ± 0.0 | 69.8 ± 32.8 | 100.0 ± 0.0 | 68.7 ± 33.2 |
| 1050 | 100.0 ± 0.0 | 76.9 ± 30.1 | 100.0 ± 0.0 | 77.0 ± 30.0 |
| 1100 | 100.0 ± 0.0 | 84.8 ± 26.1 | 99.9 ± 0.2 | 84.0 ± 26.2 |
| 1150 | 100.0 ± 0.0 | 91.0 ± 11.5 | 100.0 ± 0.0 | 89.6 ± 12.3 |
| 1200 | 100.0 ± 0.0 | 86.3 ± 22.2 | 100.0 ± 0.1 | 85.6 ± 22.1 |