s3_e2
done · peak unseen_mismatch 86.6
α schedule
α 3.0 ×50 α 2.9565 ×50 α 2.913 ×50 α 2.8696 ×50 α 2.8261 ×50 α 2.7826 ×50 α 2.7391 ×50 α 2.6957 ×50 α 2.6522 ×50 α 2.6087 ×50 α 2.5652 ×50 α 2.5217 ×50 α 2.4783 ×50 α 2.4348 ×50 α 2.3913 ×50 α 2.3478 ×50 α 2.3043 ×50 α 2.2609 ×50 α 2.2174 ×50 α 2.1739 ×50 α 2.1304 ×50 α 2.087 ×50 α 2.0435 ×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 | 50.6 ± 1.2 | 49.8 ± 1.4 | 49.8 ± 1.6 | 51.1 ± 1.2 |
| 100 | 52.8 ± 2.5 | 50.6 ± 2.4 | 51.7 ± 1.3 | 49.8 ± 1.4 |
| 150 | 76.6 ± 8.2 | 69.2 ± 9.8 | 70.9 ± 9.8 | 62.8 ± 9.5 |
| 200 | 91.0 ± 4.6 | 88.8 ± 3.9 | 82.9 ± 8.6 | 78.2 ± 7.8 |
| 250 | 92.7 ± 3.1 | 91.1 ± 3.1 | 82.4 ± 7.1 | 78.4 ± 7.2 |
| 300 | 92.8 ± 1.6 | 91.0 ± 1.6 | 81.6 ± 4.1 | 77.2 ± 5.0 |
| 350 | 93.2 ± 1.6 | 91.7 ± 2.0 | 80.0 ± 4.6 | 75.3 ± 4.9 |
| 400 | 94.1 ± 0.6 | 92.7 ± 1.3 | 81.3 ± 3.6 | 77.2 ± 5.1 |
| 450 | 93.5 ± 0.9 | 92.3 ± 1.0 | 79.6 ± 4.1 | 75.3 ± 4.7 |
| 500 | 94.1 ± 1.1 | 92.7 ± 1.8 | 81.8 ± 4.4 | 76.8 ± 4.2 |
| 550 | 95.1 ± 1.0 | 93.3 ± 1.1 | 83.2 ± 4.6 | 77.6 ± 5.1 |
| 600 | 95.2 ± 0.8 | 93.3 ± 1.2 | 82.7 ± 3.9 | 77.4 ± 4.8 |
| 650 | 95.5 ± 0.5 | 94.1 ± 1.1 | 84.2 ± 3.5 | 78.4 ± 3.7 |
| 700 | 96.3 ± 0.7 | 94.3 ± 1.2 | 85.5 ± 3.1 | 79.0 ± 5.0 |
| 750 | 96.5 ± 0.9 | 94.8 ± 1.4 | 87.8 ± 3.2 | 80.9 ± 4.8 |
| 800 | 96.5 ± 0.6 | 94.8 ± 1.2 | 87.6 ± 3.4 | 81.5 ± 5.8 |
| 850 | 96.9 ± 0.5 | 95.4 ± 1.0 | 88.2 ± 3.4 | 81.5 ± 4.6 |
| 900 | 96.9 ± 0.6 | 95.2 ± 0.8 | 87.8 ± 2.8 | 81.9 ± 4.8 |
| 950 | 97.5 ± 0.7 | 95.9 ± 1.1 | 88.9 ± 2.6 | 83.6 ± 3.4 |
| 1000 | 97.8 ± 0.7 | 96.2 ± 1.0 | 90.4 ± 2.4 | 84.2 ± 3.3 |
| 1050 | 97.6 ± 0.9 | 95.9 ± 1.3 | 89.9 ± 3.0 | 82.9 ± 5.7 |
| 1100 | 97.9 ± 0.5 | 96.3 ± 0.9 | 90.0 ± 3.3 | 83.6 ± 4.3 |
| 1150 | 98.0 ± 0.6 | 96.6 ± 0.8 | 91.2 ± 2.5 | 84.9 ± 4.3 |
| 1200 | 98.5 ± 0.5 | 97.3 ± 0.8 | 92.7 ± 3.1 | 86.6 ± 4.0 |