s2_e3
done · peak unseen_mismatch 91.7
α schedule
α 2.0 ×50 α 2.0435 ×50 α 2.087 ×50 α 2.1304 ×50 α 2.1739 ×50 α 2.2174 ×50 α 2.2609 ×50 α 2.3043 ×50 α 2.3478 ×50 α 2.3913 ×50 α 2.4348 ×50 α 2.4783 ×50 α 2.5217 ×50 α 2.5652 ×50 α 2.6087 ×50 α 2.6522 ×50 α 2.6957 ×50 α 2.7391 ×50 α 2.7826 ×50 α 2.8261 ×50 α 2.8696 ×50 α 2.913 ×50 α 2.9565 ×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 | 50.5 ± 1.9 | 49.9 ± 0.5 | 50.1 ± 1.9 | 50.1 ± 1.5 |
| 100 | 51.4 ± 2.3 | 50.3 ± 2.0 | 50.5 ± 1.9 | 49.1 ± 1.3 |
| 150 | 62.8 ± 16.9 | 57.0 ± 12.0 | 62.6 ± 15.9 | 55.2 ± 10.7 |
| 200 | 82.4 ± 18.7 | 76.4 ± 17.5 | 79.3 ± 20.6 | 73.0 ± 17.6 |
| 250 | 95.5 ± 11.1 | 92.4 ± 12.0 | 94.1 ± 12.1 | 89.3 ± 12.8 |
| 300 | 98.7 ± 1.1 | 96.4 ± 2.0 | 96.0 ± 3.4 | 91.7 ± 4.0 |
| 350 | 97.8 ± 1.6 | 96.5 ± 2.1 | 92.8 ± 5.6 | 88.9 ± 6.3 |
| 400 | 96.4 ± 1.2 | 95.0 ± 1.7 | 88.8 ± 4.7 | 84.8 ± 5.8 |
| 450 | 95.7 ± 0.9 | 94.2 ± 1.7 | 86.0 ± 4.2 | 81.6 ± 4.5 |
| 500 | 95.1 ± 1.0 | 93.8 ± 1.8 | 83.6 ± 3.4 | 79.6 ± 3.6 |
| 550 | 94.3 ± 1.2 | 93.6 ± 1.2 | 83.0 ± 4.9 | 79.2 ± 5.2 |
| 600 | 94.2 ± 1.9 | 93.1 ± 2.1 | 82.7 ± 5.4 | 78.5 ± 4.8 |
| 650 | 94.6 ± 1.4 | 93.0 ± 1.5 | 82.1 ± 5.2 | 78.2 ± 5.3 |
| 700 | 94.5 ± 1.2 | 93.4 ± 2.5 | 82.3 ± 4.0 | 77.8 ± 3.8 |
| 750 | 94.8 ± 1.5 | 92.9 ± 2.1 | 82.9 ± 6.8 | 78.2 ± 6.1 |
| 800 | 94.7 ± 1.5 | 93.0 ± 2.1 | 82.2 ± 5.0 | 78.8 ± 5.4 |
| 850 | 94.4 ± 1.0 | 93.2 ± 1.7 | 82.4 ± 4.3 | 78.5 ± 3.2 |
| 900 | 94.5 ± 1.4 | 93.1 ± 2.0 | 82.4 ± 5.4 | 77.7 ± 5.1 |
| 950 | 94.5 ± 1.3 | 93.0 ± 2.0 | 82.6 ± 4.3 | 77.6 ± 4.2 |
| 1000 | 94.0 ± 1.3 | 92.9 ± 1.9 | 82.0 ± 4.2 | 78.2 ± 3.5 |
| 1050 | 94.3 ± 1.4 | 93.0 ± 2.5 | 82.9 ± 4.8 | 78.3 ± 5.2 |
| 1100 | 94.1 ± 1.4 | 92.6 ± 2.0 | 81.7 ± 4.1 | 77.3 ± 4.4 |
| 1150 | 94.2 ± 1.3 | 92.7 ± 1.9 | 81.4 ± 4.2 | 78.4 ± 4.6 |
| 1200 | 94.3 ± 1.1 | 92.7 ± 1.9 | 80.8 ± 5.1 | 76.2 ± 4.3 |