s5_e1
done · peak unseen_mismatch 93.0
α schedule
α 5.0 ×50 α 4.8261 ×50 α 4.6522 ×50 α 4.4783 ×50 α 4.3043 ×50 α 4.1304 ×50 α 3.9565 ×50 α 3.7826 ×50 α 3.6087 ×50 α 3.4348 ×50 α 3.2609 ×50 α 3.087 ×50 α 2.913 ×50 α 2.7391 ×50 α 2.5652 ×50 α 2.3913 ×50 α 2.2174 ×50 α 2.0435 ×50 α 1.8696 ×50 α 1.6957 ×50 α 1.5217 ×50 α 1.3478 ×50 α 1.1739 ×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 | 51.3 ± 1.5 | 50.3 ± 1.3 | 50.3 ± 1.5 | 50.3 ± 1.8 |
| 100 | 61.6 ± 6.9 | 57.9 ± 5.1 | 53.8 ± 3.7 | 51.3 ± 2.2 |
| 150 | 79.6 ± 3.3 | 77.8 ± 2.2 | 65.6 ± 5.2 | 62.8 ± 4.6 |
| 200 | 81.0 ± 2.4 | 80.2 ± 2.2 | 64.7 ± 3.8 | 62.7 ± 3.5 |
| 250 | 81.9 ± 2.8 | 81.5 ± 2.9 | 65.6 ± 4.1 | 63.1 ± 4.2 |
| 300 | 84.4 ± 3.0 | 83.3 ± 2.6 | 67.0 ± 6.0 | 65.0 ± 6.1 |
| 350 | 85.7 ± 2.3 | 84.3 ± 2.1 | 67.7 ± 3.9 | 65.4 ± 3.8 |
| 400 | 88.4 ± 1.1 | 87.4 ± 2.0 | 73.0 ± 2.3 | 69.8 ± 4.4 |
| 450 | 88.0 ± 1.4 | 86.9 ± 1.6 | 70.7 ± 4.0 | 66.7 ± 4.1 |
| 500 | 89.0 ± 2.2 | 87.1 ± 2.3 | 71.5 ± 3.9 | 66.7 ± 3.6 |
| 550 | 90.2 ± 1.7 | 87.7 ± 2.3 | 73.3 ± 3.5 | 67.4 ± 3.6 |
| 600 | 91.7 ± 1.9 | 89.2 ± 2.4 | 74.4 ± 5.8 | 68.4 ± 5.9 |
| 650 | 92.4 ± 1.5 | 90.1 ± 2.2 | 77.1 ± 4.1 | 70.5 ± 4.6 |
| 700 | 93.2 ± 1.5 | 90.9 ± 1.5 | 78.1 ± 3.5 | 71.5 ± 3.3 |
| 750 | 94.3 ± 1.2 | 92.4 ± 2.0 | 81.6 ± 3.9 | 74.3 ± 5.1 |
| 800 | 95.9 ± 1.4 | 93.6 ± 1.8 | 84.9 ± 4.0 | 77.9 ± 5.2 |
| 850 | 96.5 ± 0.8 | 95.0 ± 1.3 | 87.2 ± 2.9 | 79.4 ± 4.1 |
| 900 | 97.1 ± 0.9 | 95.9 ± 1.3 | 88.2 ± 3.2 | 79.9 ± 4.8 |
| 950 | 98.5 ± 0.7 | 96.8 ± 1.3 | 91.5 ± 3.5 | 81.6 ± 5.5 |
| 1000 | 99.1 ± 0.5 | 97.7 ± 1.3 | 93.7 ± 1.6 | 85.2 ± 5.1 |
| 1050 | 99.6 ± 0.2 | 98.1 ± 1.3 | 95.7 ± 2.0 | 86.3 ± 7.1 |
| 1100 | 99.8 ± 0.1 | 98.9 ± 0.8 | 97.0 ± 1.1 | 89.1 ± 5.8 |
| 1150 | 99.9 ± 0.1 | 99.2 ± 0.7 | 98.5 ± 1.5 | 89.2 ± 6.5 |
| 1200 | 100.0 ± 0.0 | 99.6 ± 0.7 | 99.6 ± 0.5 | 93.0 ± 8.9 |