s1_e3
done · peak unseen_mismatch 96.5
α schedule
α 1.0 ×50 α 1.087 ×50 α 1.1739 ×50 α 1.2609 ×50 α 1.3478 ×50 α 1.4348 ×50 α 1.5217 ×50 α 1.6087 ×50 α 1.6957 ×50 α 1.7826 ×50 α 1.8696 ×50 α 1.9565 ×50 α 2.0435 ×50 α 2.1304 ×50 α 2.2174 ×50 α 2.3043 ×50 α 2.3913 ×50 α 2.4783 ×50 α 2.5652 ×50 α 2.6522 ×50 α 2.7391 ×50 α 2.8261 ×50 α 2.913 ×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 | 50.6 ± 1.3 | 49.1 ± 0.7 | 50.0 ± 2.0 | 50.2 ± 1.2 |
| 100 | 50.7 ± 1.2 | 49.7 ± 2.0 | 50.1 ± 1.0 | 49.5 ± 0.7 |
| 150 | 50.3 ± 2.7 | 49.1 ± 2.4 | 51.2 ± 2.4 | 48.9 ± 1.9 |
| 200 | 62.5 ± 18.5 | 44.9 ± 9.5 | 61.0 ± 19.3 | 44.3 ± 9.2 |
| 250 | 80.2 ± 21.8 | 38.6 ± 12.1 | 79.8 ± 22.1 | 38.8 ± 12.7 |
| 300 | 90.8 ± 18.5 | 42.4 ± 10.3 | 91.0 ± 17.9 | 43.7 ± 10.9 |
| 350 | 96.4 ± 10.6 | 52.9 ± 22.8 | 96.1 ± 11.5 | 53.1 ± 23.1 |
| 400 | 100.0 ± 0.0 | 58.5 ± 27.4 | 100.0 ± 0.0 | 58.8 ± 27.6 |
| 450 | 100.0 ± 0.0 | 77.3 ± 29.5 | 100.0 ± 0.0 | 77.2 ± 29.3 |
| 500 | 99.9 ± 0.2 | 89.0 ± 21.8 | 99.9 ± 0.2 | 89.0 ± 20.9 |
| 550 | 99.9 ± 0.3 | 92.1 ± 22.0 | 99.6 ± 0.7 | 91.5 ± 22.0 |
| 600 | 99.6 ± 0.8 | 98.0 ± 3.6 | 98.8 ± 1.9 | 96.5 ± 4.8 |
| 650 | 99.0 ± 1.0 | 98.6 ± 1.2 | 96.9 ± 3.2 | 96.0 ± 3.4 |
| 700 | 98.4 ± 1.4 | 97.9 ± 1.2 | 96.2 ± 3.5 | 94.9 ± 3.2 |
| 750 | 97.8 ± 1.2 | 97.4 ± 1.0 | 93.9 ± 4.0 | 92.0 ± 3.9 |
| 800 | 96.5 ± 1.3 | 96.2 ± 1.0 | 90.9 ± 3.8 | 89.0 ± 3.3 |
| 850 | 96.2 ± 0.7 | 95.5 ± 1.0 | 89.2 ± 3.3 | 87.5 ± 3.5 |
| 900 | 95.3 ± 1.0 | 94.5 ± 1.0 | 85.4 ± 4.0 | 84.2 ± 3.2 |
| 950 | 94.2 ± 0.7 | 93.8 ± 1.4 | 84.0 ± 3.4 | 82.0 ± 3.7 |
| 1000 | 93.7 ± 1.2 | 93.1 ± 1.4 | 81.4 ± 4.2 | 80.1 ± 3.5 |
| 1050 | 93.0 ± 1.4 | 92.0 ± 1.5 | 79.9 ± 3.9 | 78.0 ± 3.1 |
| 1100 | 92.3 ± 1.5 | 91.2 ± 1.5 | 77.8 ± 4.9 | 76.2 ± 4.3 |
| 1150 | 92.3 ± 1.3 | 91.3 ± 0.9 | 78.0 ± 4.1 | 75.8 ± 3.6 |
| 1200 | 91.8 ± 1.1 | 90.8 ± 1.0 | 76.7 ± 4.0 | 75.1 ± 3.5 |