s1_e2
done · peak unseen_mismatch 99.6
α schedule
α 1.0 ×50 α 1.0435 ×50 α 1.087 ×50 α 1.1304 ×50 α 1.1739 ×50 α 1.2174 ×50 α 1.2609 ×50 α 1.3043 ×50 α 1.3478 ×50 α 1.3913 ×50 α 1.4348 ×50 α 1.4783 ×50 α 1.5217 ×50 α 1.5652 ×50 α 1.6087 ×50 α 1.6522 ×50 α 1.6957 ×50 α 1.7391 ×50 α 1.7826 ×50 α 1.8261 ×50 α 1.8696 ×50 α 1.913 ×50 α 1.9565 ×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 | 50.6 ± 1.3 | 49.1 ± 0.7 | 50.0 ± 2.0 | 50.2 ± 1.2 |
| 100 | 50.4 ± 0.9 | 49.9 ± 1.8 | 49.9 ± 1.1 | 49.3 ± 0.9 |
| 150 | 51.8 ± 5.4 | 48.6 ± 3.5 | 52.4 ± 5.3 | 48.1 ± 3.3 |
| 200 | 64.3 ± 20.1 | 41.9 ± 13.4 | 62.5 ± 20.9 | 40.8 ± 13.3 |
| 250 | 72.0 ± 21.3 | 40.1 ± 16.6 | 71.3 ± 21.9 | 40.8 ± 16.7 |
| 300 | 90.7 ± 18.7 | 23.5 ± 18.7 | 90.9 ± 18.2 | 23.6 ± 19.5 |
| 350 | 94.6 ± 10.8 | 29.5 ± 18.5 | 94.4 ± 11.4 | 29.4 ± 19.2 |
| 400 | 100.0 ± 0.0 | 37.3 ± 28.3 | 100.0 ± 0.0 | 37.3 ± 27.3 |
| 450 | 100.0 ± 0.0 | 43.6 ± 27.7 | 100.0 ± 0.0 | 43.7 ± 28.0 |
| 500 | 100.0 ± 0.0 | 67.1 ± 29.7 | 100.0 ± 0.0 | 66.7 ± 29.7 |
| 550 | 100.0 ± 0.0 | 76.7 ± 19.5 | 100.0 ± 0.0 | 76.8 ± 19.3 |
| 600 | 100.0 ± 0.0 | 75.4 ± 27.9 | 100.0 ± 0.0 | 75.6 ± 27.8 |
| 650 | 100.0 ± 0.0 | 99.5 ± 0.6 | 100.0 ± 0.0 | 99.6 ± 0.4 |
| 700 | 99.9 ± 0.2 | 95.4 ± 8.0 | 100.0 ± 0.1 | 95.4 ± 8.0 |
| 750 | 100.0 ± 0.1 | 95.6 ± 12.5 | 99.4 ± 0.9 | 95.0 ± 11.5 |
| 800 | 99.6 ± 0.6 | 99.0 ± 1.3 | 99.2 ± 0.9 | 98.5 ± 1.2 |
| 850 | 99.5 ± 0.6 | 99.1 ± 0.8 | 98.7 ± 1.9 | 98.1 ± 2.1 |
| 900 | 99.2 ± 0.8 | 94.0 ± 14.4 | 97.5 ± 2.7 | 91.9 ± 13.5 |
| 950 | 99.2 ± 0.7 | 98.8 ± 0.7 | 97.4 ± 3.0 | 96.7 ± 2.7 |
| 1000 | 98.8 ± 0.9 | 98.0 ± 1.4 | 96.6 ± 3.0 | 94.9 ± 3.5 |
| 1050 | 98.6 ± 0.7 | 98.2 ± 0.9 | 95.7 ± 3.4 | 94.5 ± 3.3 |
| 1100 | 98.5 ± 0.8 | 98.0 ± 0.8 | 94.5 ± 2.8 | 93.8 ± 3.1 |
| 1150 | 98.3 ± 0.7 | 97.7 ± 1.0 | 93.2 ± 3.4 | 91.9 ± 4.1 |
| 1200 | 97.6 ± 1.2 | 97.2 ± 1.1 | 92.3 ± 3.0 | 90.8 ± 3.4 |