s4_e2
done · peak unseen_mismatch 83.8
α schedule
α 4.0 ×50 α 3.913 ×50 α 3.8261 ×50 α 3.7391 ×50 α 3.6522 ×50 α 3.5652 ×50 α 3.4783 ×50 α 3.3913 ×50 α 3.3043 ×50 α 3.2174 ×50 α 3.1304 ×50 α 3.0435 ×50 α 2.9565 ×50 α 2.8696 ×50 α 2.7826 ×50 α 2.6957 ×50 α 2.6087 ×50 α 2.5217 ×50 α 2.4348 ×50 α 2.3478 ×50 α 2.2609 ×50 α 2.1739 ×50 α 2.087 ×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 | 51.7 ± 1.9 | 49.6 ± 1.0 | 49.9 ± 2.1 | 50.5 ± 1.2 |
| 100 | 58.6 ± 8.3 | 56.6 ± 7.0 | 54.1 ± 5.6 | 53.0 ± 4.4 |
| 150 | 77.5 ± 6.6 | 76.6 ± 6.3 | 64.2 ± 6.8 | 61.8 ± 6.4 |
| 200 | 82.8 ± 1.5 | 81.3 ± 2.0 | 66.4 ± 3.5 | 64.2 ± 4.4 |
| 250 | 85.5 ± 2.3 | 83.8 ± 1.7 | 67.8 ± 5.1 | 66.1 ± 3.5 |
| 300 | 88.5 ± 2.1 | 87.0 ± 1.8 | 71.7 ± 3.7 | 69.2 ± 4.3 |
| 350 | 89.5 ± 2.4 | 87.1 ± 2.1 | 71.8 ± 5.0 | 68.5 ± 4.9 |
| 400 | 89.2 ± 2.2 | 87.4 ± 2.3 | 71.9 ± 4.9 | 69.0 ± 4.9 |
| 450 | 91.0 ± 1.3 | 88.6 ± 1.9 | 73.8 ± 5.3 | 69.7 ± 4.9 |
| 500 | 92.3 ± 1.3 | 90.5 ± 1.9 | 75.7 ± 4.4 | 70.1 ± 5.9 |
| 550 | 91.9 ± 1.3 | 89.6 ± 1.6 | 76.9 ± 3.2 | 72.3 ± 4.0 |
| 600 | 92.3 ± 1.4 | 90.4 ± 1.5 | 76.3 ± 4.4 | 71.6 ± 4.1 |
| 650 | 93.8 ± 1.2 | 91.6 ± 1.4 | 78.2 ± 3.8 | 73.2 ± 4.6 |
| 700 | 93.4 ± 1.2 | 91.7 ± 1.3 | 78.9 ± 2.8 | 73.0 ± 4.1 |
| 750 | 94.8 ± 1.1 | 92.5 ± 1.4 | 82.1 ± 3.9 | 76.4 ± 5.0 |
| 800 | 94.7 ± 1.1 | 92.5 ± 1.5 | 81.4 ± 2.5 | 75.0 ± 4.9 |
| 850 | 95.1 ± 0.8 | 93.0 ± 1.5 | 83.0 ± 3.7 | 77.0 ± 4.6 |
| 900 | 96.1 ± 0.8 | 94.0 ± 1.8 | 84.6 ± 3.5 | 77.2 ± 5.9 |
| 950 | 96.2 ± 0.8 | 94.3 ± 1.3 | 85.7 ± 2.6 | 78.0 ± 4.3 |
| 1000 | 96.5 ± 1.0 | 94.2 ± 1.1 | 86.3 ± 3.9 | 77.8 ± 4.6 |
| 1050 | 96.9 ± 0.7 | 95.1 ± 1.2 | 86.7 ± 2.3 | 79.2 ± 5.0 |
| 1100 | 97.4 ± 0.3 | 95.5 ± 0.9 | 88.5 ± 2.3 | 81.8 ± 4.7 |
| 1150 | 98.4 ± 0.4 | 96.9 ± 0.6 | 90.4 ± 2.6 | 83.8 ± 4.3 |
| 1200 | 98.2 ± 0.9 | 96.6 ± 0.9 | 91.2 ± 2.6 | 83.5 ± 5.3 |