s1_e4
done · peak unseen_mismatch 98.7
α schedule
α 1.0 ×50 α 1.1304 ×50 α 1.2609 ×50 α 1.3913 ×50 α 1.5217 ×50 α 1.6522 ×50 α 1.7826 ×50 α 1.913 ×50 α 2.0435 ×50 α 2.1739 ×50 α 2.3043 ×50 α 2.4348 ×50 α 2.5652 ×50 α 2.6957 ×50 α 2.8261 ×50 α 2.9565 ×50 α 3.087 ×50 α 3.2174 ×50 α 3.3478 ×50 α 3.4783 ×50 α 3.6087 ×50 α 3.7391 ×50 α 3.8696 ×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 | 50.6 ± 1.3 | 49.1 ± 0.7 | 50.0 ± 2.0 | 50.2 ± 1.2 |
| 100 | 50.5 ± 1.1 | 49.6 ± 1.8 | 50.3 ± 1.7 | 49.5 ± 0.7 |
| 150 | 52.1 ± 4.9 | 48.1 ± 3.6 | 52.6 ± 4.8 | 48.6 ± 3.1 |
| 200 | 64.1 ± 18.4 | 43.0 ± 13.4 | 61.8 ± 19.3 | 42.1 ± 12.9 |
| 250 | 82.9 ± 20.7 | 51.4 ± 12.5 | 82.3 ± 21.5 | 51.6 ± 12.4 |
| 300 | 99.6 ± 0.9 | 58.2 ± 16.6 | 99.7 ± 0.7 | 58.6 ± 16.6 |
| 350 | 100.0 ± 0.0 | 79.4 ± 16.2 | 100.0 ± 0.0 | 78.7 ± 16.0 |
| 400 | 100.0 ± 0.0 | 83.6 ± 19.1 | 100.0 ± 0.0 | 84.2 ± 18.6 |
| 450 | 99.8 ± 0.5 | 93.5 ± 11.2 | 99.6 ± 1.1 | 92.5 ± 11.2 |
| 500 | 99.8 ± 0.4 | 99.4 ± 0.9 | 99.5 ± 0.8 | 98.7 ± 1.7 |
| 550 | 99.3 ± 0.9 | 99.0 ± 1.4 | 98.6 ± 1.6 | 97.9 ± 2.3 |
| 600 | 98.8 ± 1.0 | 98.2 ± 1.3 | 95.9 ± 3.4 | 95.0 ± 4.3 |
| 650 | 98.1 ± 1.1 | 97.5 ± 1.2 | 93.5 ± 4.2 | 92.0 ± 4.7 |
| 700 | 97.0 ± 1.5 | 96.0 ± 1.8 | 90.6 ± 3.7 | 89.1 ± 5.0 |
| 750 | 95.4 ± 1.5 | 94.5 ± 1.6 | 86.4 ± 3.5 | 84.7 ± 4.0 |
| 800 | 94.1 ± 0.9 | 93.6 ± 1.1 | 84.3 ± 3.5 | 82.5 ± 4.2 |
| 850 | 92.6 ± 0.9 | 91.9 ± 1.2 | 81.0 ± 3.6 | 79.2 ± 4.5 |
| 900 | 91.5 ± 1.3 | 91.1 ± 1.4 | 78.1 ± 4.6 | 76.6 ± 4.8 |
| 950 | 91.2 ± 0.7 | 90.4 ± 0.9 | 76.3 ± 3.3 | 74.8 ± 3.5 |
| 1000 | 90.6 ± 1.1 | 89.8 ± 1.3 | 76.5 ± 3.9 | 74.5 ± 4.4 |
| 1050 | 90.1 ± 0.9 | 89.2 ± 1.5 | 75.0 ± 4.0 | 72.6 ± 4.3 |
| 1100 | 89.4 ± 1.1 | 88.9 ± 1.4 | 73.4 ± 4.5 | 71.7 ± 4.4 |
| 1150 | 88.9 ± 1.2 | 88.3 ± 1.7 | 72.4 ± 4.7 | 70.1 ± 4.6 |
| 1200 | 88.6 ± 1.3 | 88.2 ± 1.8 | 72.0 ± 4.0 | 70.1 ± 4.3 |