s2_e1
done · peak unseen_mismatch 95.8
α schedule
α 2.0 ×50 α 1.9565 ×50 α 1.913 ×50 α 1.8696 ×50 α 1.8261 ×50 α 1.7826 ×50 α 1.7391 ×50 α 1.6957 ×50 α 1.6522 ×50 α 1.6087 ×50 α 1.5652 ×50 α 1.5217 ×50 α 1.4783 ×50 α 1.4348 ×50 α 1.3913 ×50 α 1.3478 ×50 α 1.3043 ×50 α 1.2609 ×50 α 1.2174 ×50 α 1.1739 ×50 α 1.1304 ×50 α 1.087 ×50 α 1.0435 ×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 | 51.2 ± 0.9 | 49.3 ± 0.9 | 50.0 ± 1.5 | 50.5 ± 1.4 |
| 100 | 51.3 ± 2.0 | 49.8 ± 1.5 | 51.1 ± 1.4 | 49.3 ± 1.4 |
| 150 | 63.7 ± 10.5 | 51.2 ± 6.9 | 60.8 ± 10.6 | 51.8 ± 5.3 |
| 200 | 95.8 ± 5.2 | 62.6 ± 24.4 | 94.8 ± 6.5 | 60.8 ± 22.8 |
| 250 | 99.8 ± 0.6 | 81.9 ± 19.2 | 99.5 ± 1.6 | 80.6 ± 18.5 |
| 300 | 99.8 ± 0.2 | 85.4 ± 15.6 | 99.4 ± 1.0 | 84.5 ± 14.6 |
| 350 | 99.7 ± 0.4 | 89.6 ± 14.4 | 98.9 ± 1.7 | 87.3 ± 13.9 |
| 400 | 99.2 ± 0.8 | 96.1 ± 4.2 | 98.4 ± 2.1 | 94.1 ± 4.3 |
| 450 | 99.2 ± 0.6 | 97.4 ± 2.7 | 97.2 ± 2.2 | 93.8 ± 3.7 |
| 500 | 99.2 ± 0.7 | 98.3 ± 0.9 | 96.9 ± 2.3 | 93.4 ± 3.7 |
| 550 | 99.5 ± 0.5 | 98.5 ± 1.0 | 97.9 ± 1.8 | 95.8 ± 2.3 |
| 600 | 99.2 ± 0.6 | 98.3 ± 0.9 | 96.7 ± 2.8 | 93.8 ± 3.7 |
| 650 | 99.2 ± 0.4 | 98.3 ± 0.8 | 97.2 ± 1.6 | 93.2 ± 3.3 |
| 700 | 99.5 ± 0.2 | 98.5 ± 0.9 | 97.0 ± 1.8 | 93.3 ± 4.1 |
| 750 | 99.7 ± 0.2 | 98.8 ± 0.8 | 97.8 ± 1.3 | 93.1 ± 4.6 |
| 800 | 99.7 ± 0.2 | 98.9 ± 0.6 | 97.8 ± 1.4 | 93.9 ± 3.5 |
| 850 | 99.8 ± 0.2 | 99.0 ± 0.9 | 98.6 ± 1.0 | 94.6 ± 3.4 |
| 900 | 99.8 ± 0.1 | 98.9 ± 0.9 | 97.9 ± 1.5 | 93.3 ± 4.4 |
| 950 | 99.9 ± 0.1 | 99.4 ± 0.6 | 98.8 ± 1.0 | 95.3 ± 4.3 |
| 1000 | 99.9 ± 0.1 | 99.2 ± 0.8 | 99.0 ± 1.0 | 94.9 ± 4.7 |
| 1050 | 100.0 ± 0.1 | 98.7 ± 1.9 | 98.9 ± 1.5 | 93.8 ± 7.1 |
| 1100 | 100.0 ± 0.1 | 99.3 ± 0.7 | 99.4 ± 0.6 | 95.6 ± 4.6 |
| 1150 | 100.0 ± 0.1 | 98.6 ± 2.5 | 99.1 ± 0.7 | 93.2 ± 6.8 |
| 1200 | 100.0 ± 0.0 | 97.1 ± 7.4 | 99.5 ± 0.5 | 92.4 ± 9.8 |