s3_e0
done · peak unseen_mismatch 92.0
α schedule
α 3.0 ×50 α 2.8696 ×50 α 2.7391 ×50 α 2.6087 ×50 α 2.4783 ×50 α 2.3478 ×50 α 2.2174 ×50 α 2.087 ×50 α 1.9565 ×50 α 1.8261 ×50 α 1.6957 ×50 α 1.5652 ×50 α 1.4348 ×50 α 1.3043 ×50 α 1.1739 ×50 α 1.0435 ×50 α 0.913 ×50 α 0.7826 ×50 α 0.6522 ×50 α 0.5217 ×50 α 0.3913 ×50 α 0.2609 ×50 α 0.1304 ×50 α 0.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.0
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 ± 3.1 | 50.8 ± 2.1 | 51.8 ± 1.9 | 49.8 ± 1.4 |
| 150 | 77.0 ± 8.8 | 69.0 ± 12.2 | 70.1 ± 9.9 | 63.0 ± 9.9 |
| 200 | 93.8 ± 4.7 | 88.6 ± 6.0 | 87.7 ± 8.7 | 80.2 ± 8.1 |
| 250 | 94.8 ± 2.5 | 93.0 ± 2.9 | 86.9 ± 6.7 | 83.1 ± 6.6 |
| 300 | 94.7 ± 1.8 | 93.6 ± 1.9 | 86.3 ± 4.7 | 82.6 ± 4.3 |
| 350 | 95.5 ± 1.0 | 93.8 ± 1.5 | 85.8 ± 3.3 | 79.9 ± 4.5 |
| 400 | 97.1 ± 0.8 | 95.4 ± 1.1 | 87.3 ± 2.0 | 82.3 ± 3.4 |
| 450 | 97.2 ± 0.8 | 95.2 ± 1.7 | 88.3 ± 3.1 | 81.2 ± 5.0 |
| 500 | 98.0 ± 0.7 | 96.1 ± 1.1 | 90.6 ± 2.4 | 83.6 ± 2.8 |
| 550 | 98.6 ± 0.5 | 97.2 ± 1.1 | 92.2 ± 2.4 | 85.2 ± 3.7 |
| 600 | 99.0 ± 0.3 | 97.7 ± 1.4 | 94.3 ± 1.6 | 86.0 ± 6.2 |
| 650 | 99.6 ± 0.3 | 98.4 ± 1.0 | 96.1 ± 2.2 | 89.6 ± 6.0 |
| 700 | 99.8 ± 0.1 | 98.8 ± 0.5 | 97.5 ± 1.4 | 90.4 ± 4.1 |
| 750 | 99.9 ± 0.1 | 99.3 ± 0.5 | 98.9 ± 0.8 | 92.0 ± 5.4 |
| 800 | 100.0 ± 0.0 | 99.1 ± 1.1 | 99.1 ± 0.7 | 91.0 ± 7.7 |
| 850 | 100.0 ± 0.0 | 99.1 ± 0.8 | 99.5 ± 0.7 | 91.7 ± 3.5 |
| 900 | 100.0 ± 0.0 | 98.0 ± 3.0 | 99.5 ± 0.3 | 89.5 ± 8.0 |
| 950 | 100.0 ± 0.0 | 88.9 ± 16.5 | 99.5 ± 1.1 | 79.0 ± 20.5 |
| 1000 | 100.0 ± 0.0 | 81.0 ± 16.9 | 99.9 ± 0.2 | 67.1 ± 17.9 |
| 1050 | 100.0 ± 0.0 | 54.1 ± 24.1 | 99.9 ± 0.1 | 47.4 ± 21.7 |
| 1100 | 100.0 ± 0.0 | 30.3 ± 14.3 | 100.0 ± 0.1 | 29.2 ± 14.9 |
| 1150 | 100.0 ± 0.0 | 19.7 ± 12.4 | 99.6 ± 0.7 | 22.7 ± 12.8 |
| 1200 | 100.0 ± 0.0 | 15.2 ± 12.0 | 99.4 ± 0.7 | 19.0 ± 11.3 |