s3_e1
done · peak unseen_mismatch 93.2
α schedule
α 3.0 ×50 α 2.913 ×50 α 2.8261 ×50 α 2.7391 ×50 α 2.6522 ×50 α 2.5652 ×50 α 2.4783 ×50 α 2.3913 ×50 α 2.3043 ×50 α 2.2174 ×50 α 2.1304 ×50 α 2.0435 ×50 α 1.9565 ×50 α 1.8696 ×50 α 1.7826 ×50 α 1.6957 ×50 α 1.6087 ×50 α 1.5217 ×50 α 1.4348 ×50 α 1.3478 ×50 α 1.2609 ×50 α 1.1739 ×50 α 1.087 ×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 ± 1.2 | 49.8 ± 1.4 | 49.8 ± 1.6 | 51.1 ± 1.2 |
| 100 | 52.6 ± 2.4 | 50.9 ± 2.2 | 51.7 ± 1.4 | 50.0 ± 1.3 |
| 150 | 78.3 ± 7.8 | 69.8 ± 10.7 | 71.7 ± 9.6 | 63.3 ± 9.4 |
| 200 | 91.9 ± 5.1 | 89.0 ± 3.8 | 84.3 ± 9.0 | 78.7 ± 7.2 |
| 250 | 93.2 ± 3.1 | 91.9 ± 3.2 | 84.0 ± 6.9 | 80.2 ± 8.1 |
| 300 | 93.7 ± 1.0 | 92.6 ± 1.5 | 83.6 ± 3.1 | 79.0 ± 3.4 |
| 350 | 94.4 ± 1.3 | 92.5 ± 2.2 | 82.1 ± 2.9 | 77.2 ± 3.4 |
| 400 | 95.3 ± 1.0 | 93.5 ± 1.5 | 84.7 ± 3.3 | 79.0 ± 5.3 |
| 450 | 95.5 ± 1.0 | 93.9 ± 1.4 | 83.9 ± 3.2 | 78.4 ± 3.6 |
| 500 | 96.3 ± 1.1 | 94.3 ± 0.8 | 87.4 ± 3.1 | 80.3 ± 3.2 |
| 550 | 96.8 ± 1.0 | 95.2 ± 1.1 | 87.7 ± 3.6 | 81.3 ± 4.9 |
| 600 | 97.3 ± 0.9 | 95.2 ± 1.7 | 88.9 ± 2.6 | 81.3 ± 5.8 |
| 650 | 97.9 ± 0.7 | 96.1 ± 1.2 | 89.9 ± 3.0 | 82.5 ± 4.6 |
| 700 | 98.5 ± 0.4 | 96.5 ± 1.2 | 91.9 ± 1.7 | 83.5 ± 3.7 |
| 750 | 98.7 ± 0.4 | 97.1 ± 0.9 | 92.1 ± 2.9 | 84.4 ± 5.5 |
| 800 | 99.0 ± 0.4 | 97.7 ± 1.0 | 94.1 ± 2.0 | 87.5 ± 5.2 |
| 850 | 99.5 ± 0.2 | 98.5 ± 0.7 | 95.6 ± 2.1 | 89.1 ± 4.7 |
| 900 | 99.5 ± 0.2 | 98.4 ± 0.8 | 95.3 ± 2.0 | 88.8 ± 4.4 |
| 950 | 99.8 ± 0.2 | 98.8 ± 0.7 | 97.3 ± 1.5 | 89.9 ± 5.0 |
| 1000 | 99.9 ± 0.1 | 98.8 ± 1.1 | 97.8 ± 1.4 | 89.8 ± 7.0 |
| 1050 | 99.9 ± 0.1 | 98.8 ± 1.2 | 98.2 ± 1.3 | 89.6 ± 8.1 |
| 1100 | 100.0 ± 0.1 | 99.2 ± 0.8 | 98.7 ± 0.7 | 92.0 ± 4.2 |
| 1150 | 100.0 ± 0.0 | 99.2 ± 0.8 | 98.9 ± 1.0 | 91.4 ± 7.8 |
| 1200 | 100.0 ± 0.0 | 99.5 ± 1.0 | 99.5 ± 0.3 | 93.2 ± 8.1 |