s1_e5
done · peak unseen_mismatch 94.9
α schedule
α 1.0 ×50 α 1.1739 ×50 α 1.3478 ×50 α 1.5217 ×50 α 1.6957 ×50 α 1.8696 ×50 α 2.0435 ×50 α 2.2174 ×50 α 2.3913 ×50 α 2.5652 ×50 α 2.7391 ×50 α 2.913 ×50 α 3.087 ×50 α 3.2609 ×50 α 3.4348 ×50 α 3.6087 ×50 α 3.7826 ×50 α 3.9565 ×50 α 4.1304 ×50 α 4.3043 ×50 α 4.4783 ×50 α 4.6522 ×50 α 4.8261 ×50 α 5.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 5.0
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 ± 1.3 | 49.1 ± 0.7 | 50.0 ± 2.0 | 50.2 ± 1.2 |
| 100 | 50.4 ± 1.5 | 49.8 ± 1.8 | 50.4 ± 1.4 | 49.3 ± 0.6 |
| 150 | 52.4 ± 7.1 | 47.4 ± 5.1 | 53.1 ± 7.1 | 47.7 ± 4.6 |
| 200 | 65.4 ± 19.2 | 46.0 ± 7.6 | 64.2 ± 19.5 | 44.8 ± 8.6 |
| 250 | 84.3 ± 20.3 | 46.4 ± 14.1 | 83.5 ± 20.9 | 47.4 ± 14.1 |
| 300 | 95.0 ± 10.5 | 63.4 ± 18.8 | 95.0 ± 10.5 | 64.5 ± 18.5 |
| 350 | 99.9 ± 0.2 | 83.0 ± 20.9 | 99.8 ± 0.3 | 83.0 ± 19.7 |
| 400 | 99.6 ± 0.9 | 82.6 ± 27.3 | 99.0 ± 1.9 | 82.1 ± 27.4 |
| 450 | 99.7 ± 0.4 | 93.6 ± 7.8 | 98.7 ± 1.8 | 92.5 ± 7.4 |
| 500 | 98.9 ± 1.1 | 97.3 ± 1.8 | 97.2 ± 2.9 | 94.9 ± 3.5 |
| 550 | 97.6 ± 2.0 | 96.2 ± 2.1 | 94.0 ± 5.2 | 91.4 ± 5.4 |
| 600 | 96.1 ± 2.2 | 95.2 ± 2.1 | 89.3 ± 5.5 | 88.1 ± 5.6 |
| 650 | 95.2 ± 1.8 | 94.1 ± 1.5 | 86.6 ± 5.3 | 84.5 ± 3.8 |
| 700 | 93.4 ± 2.1 | 92.6 ± 1.6 | 82.6 ± 4.4 | 80.1 ± 4.1 |
| 750 | 91.5 ± 1.6 | 90.8 ± 1.3 | 78.6 ± 4.7 | 76.8 ± 4.2 |
| 800 | 90.0 ± 1.3 | 89.1 ± 1.2 | 76.2 ± 3.8 | 74.3 ± 2.5 |
| 850 | 89.2 ± 1.1 | 88.4 ± 1.3 | 73.3 ± 4.1 | 71.2 ± 2.9 |
| 900 | 88.4 ± 0.8 | 87.5 ± 1.6 | 71.3 ± 5.2 | 68.7 ± 2.7 |
| 950 | 87.4 ± 0.8 | 87.1 ± 1.9 | 70.4 ± 4.3 | 67.9 ± 2.7 |
| 1000 | 87.0 ± 0.9 | 86.4 ± 2.1 | 68.9 ± 4.6 | 66.9 ± 2.5 |
| 1050 | 86.9 ± 1.2 | 86.1 ± 1.7 | 69.5 ± 3.2 | 67.4 ± 2.9 |
| 1100 | 86.5 ± 0.9 | 86.0 ± 1.8 | 67.9 ± 3.5 | 66.0 ± 2.1 |
| 1150 | 86.6 ± 1.2 | 85.8 ± 1.3 | 67.5 ± 3.8 | 65.0 ± 3.2 |
| 1200 | 86.0 ± 1.5 | 85.5 ± 1.5 | 66.0 ± 4.3 | 64.3 ± 2.2 |