s5_e3
done · peak unseen_mismatch 73.0
α schedule
α 5.0 ×50 α 4.913 ×50 α 4.8261 ×50 α 4.7391 ×50 α 4.6522 ×50 α 4.5652 ×50 α 4.4783 ×50 α 4.3913 ×50 α 4.3043 ×50 α 4.2174 ×50 α 4.1304 ×50 α 4.0435 ×50 α 3.9565 ×50 α 3.8696 ×50 α 3.7826 ×50 α 3.6957 ×50 α 3.6087 ×50 α 3.5217 ×50 α 3.4348 ×50 α 3.3478 ×50 α 3.2609 ×50 α 3.1739 ×50 α 3.087 ×50 α 3.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.0
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 | 51.8 ± 2.1 | 49.9 ± 1.2 | 50.2 ± 1.8 | 50.0 ± 1.4 |
| 100 | 60.2 ± 7.2 | 58.5 ± 7.0 | 54.0 ± 4.9 | 52.4 ± 4.2 |
| 150 | 75.7 ± 5.6 | 74.2 ± 5.7 | 61.7 ± 4.2 | 58.7 ± 3.9 |
| 200 | 79.8 ± 1.8 | 78.3 ± 1.4 | 63.0 ± 3.2 | 61.1 ± 3.3 |
| 250 | 82.1 ± 2.0 | 80.7 ± 1.7 | 63.1 ± 2.8 | 61.5 ± 2.4 |
| 300 | 84.0 ± 2.2 | 82.6 ± 2.0 | 65.6 ± 4.4 | 63.2 ± 3.5 |
| 350 | 85.4 ± 2.5 | 83.9 ± 2.4 | 66.5 ± 5.8 | 64.8 ± 5.1 |
| 400 | 85.3 ± 2.4 | 83.6 ± 2.9 | 67.0 ± 4.9 | 64.0 ± 5.6 |
| 450 | 86.4 ± 1.1 | 84.3 ± 2.1 | 67.1 ± 3.4 | 63.5 ± 3.9 |
| 500 | 87.0 ± 1.6 | 85.1 ± 1.6 | 68.7 ± 3.5 | 64.1 ± 4.5 |
| 550 | 87.9 ± 2.5 | 85.7 ± 3.1 | 70.2 ± 6.3 | 67.0 ± 7.7 |
| 600 | 88.2 ± 1.5 | 85.9 ± 1.9 | 70.5 ± 4.1 | 65.7 ± 5.3 |
| 650 | 88.9 ± 1.6 | 86.7 ± 1.8 | 70.7 ± 5.5 | 66.0 ± 5.7 |
| 700 | 88.8 ± 1.7 | 86.7 ± 2.1 | 70.3 ± 4.1 | 67.0 ± 4.4 |
| 750 | 89.9 ± 1.1 | 87.7 ± 2.3 | 73.2 ± 3.4 | 66.6 ± 5.5 |
| 800 | 89.9 ± 1.6 | 87.2 ± 2.4 | 72.4 ± 4.2 | 67.2 ± 6.2 |
| 850 | 90.5 ± 1.1 | 88.1 ± 2.2 | 72.8 ± 4.8 | 67.2 ± 5.8 |
| 900 | 91.1 ± 0.9 | 89.1 ± 1.3 | 74.7 ± 4.5 | 69.6 ± 4.9 |
| 950 | 91.2 ± 1.1 | 88.8 ± 1.6 | 73.8 ± 4.0 | 68.5 ± 4.5 |
| 1000 | 92.0 ± 1.1 | 90.0 ± 1.4 | 76.5 ± 3.2 | 71.2 ± 5.5 |
| 1050 | 92.5 ± 1.2 | 89.8 ± 2.4 | 77.3 ± 3.5 | 70.3 ± 7.0 |
| 1100 | 92.4 ± 0.9 | 90.1 ± 1.5 | 77.2 ± 4.3 | 70.2 ± 5.7 |
| 1150 | 93.3 ± 0.7 | 91.0 ± 1.6 | 79.4 ± 3.2 | 73.0 ± 5.8 |
| 1200 | 93.8 ± 1.2 | 91.0 ± 1.4 | 79.1 ± 3.8 | 72.2 ± 4.7 |