s2_e2
done · peak unseen_mismatch 93.8
α schedule
α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50 α 2.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.0
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 | 51.2 ± 0.9 | 49.3 ± 0.9 | 50.0 ± 1.5 | 50.5 ± 1.4 |
| 100 | 51.3 ± 2.0 | 49.8 ± 1.4 | 50.9 ± 1.8 | 49.1 ± 1.3 |
| 150 | 63.3 ± 8.3 | 51.4 ± 6.1 | 60.9 ± 8.1 | 50.5 ± 4.6 |
| 200 | 96.0 ± 5.4 | 63.0 ± 23.7 | 94.4 ± 7.8 | 61.3 ± 23.1 |
| 250 | 99.6 ± 0.9 | 81.4 ± 24.4 | 98.5 ± 2.8 | 79.1 ± 23.9 |
| 300 | 99.4 ± 0.8 | 93.7 ± 15.0 | 98.6 ± 2.4 | 92.0 ± 14.3 |
| 350 | 99.1 ± 0.9 | 96.7 ± 2.5 | 97.4 ± 3.1 | 93.8 ± 4.5 |
| 400 | 98.5 ± 1.3 | 97.4 ± 1.5 | 95.5 ± 3.8 | 92.8 ± 5.6 |
| 450 | 98.2 ± 1.1 | 96.8 ± 1.4 | 93.1 ± 4.4 | 89.4 ± 5.9 |
| 500 | 97.4 ± 1.3 | 96.2 ± 1.6 | 92.1 ± 4.1 | 88.3 ± 5.5 |
| 550 | 97.4 ± 0.8 | 96.4 ± 1.2 | 92.0 ± 2.7 | 87.8 ± 4.6 |
| 600 | 97.1 ± 0.6 | 96.0 ± 1.2 | 90.6 ± 2.2 | 86.8 ± 3.3 |
| 650 | 97.3 ± 0.9 | 96.0 ± 0.9 | 90.8 ± 2.7 | 86.1 ± 2.6 |
| 700 | 97.3 ± 0.5 | 96.3 ± 1.0 | 90.6 ± 2.1 | 86.9 ± 3.6 |
| 750 | 97.7 ± 0.8 | 96.3 ± 1.1 | 91.3 ± 2.9 | 87.3 ± 4.1 |
| 800 | 97.8 ± 0.8 | 96.6 ± 0.8 | 89.6 ± 2.7 | 85.3 ± 3.1 |
| 850 | 97.6 ± 0.6 | 96.4 ± 0.9 | 91.0 ± 2.8 | 87.0 ± 3.3 |
| 900 | 97.7 ± 0.7 | 96.7 ± 1.0 | 90.6 ± 1.5 | 86.4 ± 3.4 |
| 950 | 97.9 ± 0.7 | 96.7 ± 1.1 | 90.7 ± 2.7 | 87.2 ± 3.9 |
| 1000 | 98.0 ± 0.7 | 96.9 ± 1.0 | 90.8 ± 3.2 | 86.7 ± 3.5 |
| 1050 | 98.0 ± 0.5 | 96.9 ± 1.2 | 90.9 ± 2.6 | 86.3 ± 4.5 |
| 1100 | 98.0 ± 0.6 | 96.9 ± 0.8 | 90.9 ± 3.2 | 87.3 ± 3.7 |
| 1150 | 97.8 ± 0.5 | 96.5 ± 1.0 | 90.8 ± 2.7 | 85.6 ± 3.2 |
| 1200 | 98.2 ± 0.6 | 97.3 ± 0.8 | 91.8 ± 1.4 | 86.9 ± 4.0 |