s5_e4
done · peak unseen_mismatch 69.1
α schedule
α 5.0 ×50 α 4.9565 ×50 α 4.913 ×50 α 4.8696 ×50 α 4.8261 ×50 α 4.7826 ×50 α 4.7391 ×50 α 4.6957 ×50 α 4.6522 ×50 α 4.6087 ×50 α 4.5652 ×50 α 4.5217 ×50 α 4.4783 ×50 α 4.4348 ×50 α 4.3913 ×50 α 4.3478 ×50 α 4.3043 ×50 α 4.2609 ×50 α 4.2174 ×50 α 4.1739 ×50 α 4.1304 ×50 α 4.087 ×50 α 4.0435 ×50 α 4.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.0
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 | 61.8 ± 6.8 | 58.2 ± 5.2 | 53.9 ± 3.8 | 51.2 ± 2.2 |
| 150 | 78.6 ± 3.9 | 76.9 ± 3.3 | 65.2 ± 5.3 | 62.3 ± 4.9 |
| 200 | 80.2 ± 2.9 | 79.0 ± 3.0 | 63.5 ± 3.9 | 61.7 ± 3.3 |
| 250 | 81.1 ± 3.4 | 80.3 ± 2.7 | 64.0 ± 3.6 | 62.3 ± 4.3 |
| 300 | 82.6 ± 3.7 | 81.8 ± 3.3 | 66.1 ± 6.7 | 64.0 ± 6.5 |
| 350 | 82.4 ± 2.5 | 82.1 ± 2.1 | 64.5 ± 3.5 | 62.2 ± 3.8 |
| 400 | 84.8 ± 2.1 | 84.2 ± 2.6 | 68.7 ± 3.5 | 65.5 ± 5.0 |
| 450 | 84.6 ± 2.6 | 83.8 ± 3.0 | 67.1 ± 4.2 | 63.6 ± 4.7 |
| 500 | 85.9 ± 3.5 | 84.6 ± 3.3 | 68.9 ± 5.5 | 64.5 ± 4.8 |
| 550 | 85.5 ± 2.4 | 84.4 ± 3.2 | 68.1 ± 3.4 | 64.6 ± 3.3 |
| 600 | 86.0 ± 2.7 | 84.5 ± 2.8 | 68.7 ± 4.7 | 65.3 ± 4.1 |
| 650 | 86.8 ± 2.0 | 84.9 ± 2.6 | 68.3 ± 3.7 | 64.3 ± 3.6 |
| 700 | 87.5 ± 2.4 | 85.3 ± 2.1 | 70.0 ± 3.7 | 65.5 ± 3.3 |
| 750 | 87.2 ± 1.8 | 85.7 ± 2.1 | 70.9 ± 3.1 | 66.0 ± 3.3 |
| 800 | 87.0 ± 1.8 | 86.1 ± 2.4 | 70.4 ± 3.1 | 65.4 ± 4.1 |
| 850 | 88.1 ± 1.5 | 86.6 ± 2.0 | 70.9 ± 3.0 | 65.7 ± 3.6 |
| 900 | 88.4 ± 2.0 | 86.5 ± 2.6 | 71.7 ± 4.1 | 65.2 ± 5.4 |
| 950 | 88.4 ± 1.7 | 86.8 ± 2.0 | 71.0 ± 3.3 | 64.9 ± 4.5 |
| 1000 | 88.7 ± 1.5 | 86.5 ± 1.8 | 71.2 ± 3.4 | 66.5 ± 4.7 |
| 1050 | 89.1 ± 1.2 | 87.2 ± 1.5 | 72.2 ± 4.0 | 66.7 ± 5.4 |
| 1100 | 89.8 ± 2.0 | 88.2 ± 2.0 | 73.2 ± 3.6 | 67.2 ± 4.2 |
| 1150 | 90.2 ± 1.7 | 87.7 ± 2.3 | 73.8 ± 3.9 | 67.2 ± 4.4 |
| 1200 | 90.3 ± 1.3 | 88.2 ± 1.5 | 73.8 ± 3.1 | 69.1 ± 3.7 |