s0_e3
done · peak unseen_mismatch 86.8
α schedule
α 0.0 ×50 α 0.1304 ×50 α 0.2609 ×50 α 0.3913 ×50 α 0.5217 ×50 α 0.6522 ×50 α 0.7826 ×50 α 0.913 ×50 α 1.0435 ×50 α 1.1739 ×50 α 1.3043 ×50 α 1.4348 ×50 α 1.5652 ×50 α 1.6957 ×50 α 1.8261 ×50 α 1.9565 ×50 α 2.087 ×50 α 2.2174 ×50 α 2.3478 ×50 α 2.4783 ×50 α 2.6087 ×50 α 2.7391 ×50 α 2.8696 ×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 | 50.4 ± 1.6 | 49.9 ± 1.1 | 49.8 ± 1.9 | 49.6 ± 1.9 |
| 100 | 50.5 ± 1.5 | 50.2 ± 1.7 | 49.8 ± 1.4 | 49.6 ± 1.0 |
| 150 | 50.1 ± 1.8 | 49.6 ± 2.1 | 50.1 ± 2.5 | 48.9 ± 1.8 |
| 200 | 60.8 ± 17.0 | 44.4 ± 8.4 | 60.3 ± 17.2 | 44.3 ± 8.4 |
| 250 | 70.2 ± 23.3 | 38.3 ± 17.8 | 69.8 ± 23.7 | 38.6 ± 17.9 |
| 300 | 76.9 ± 23.2 | 28.9 ± 21.0 | 76.4 ± 23.7 | 30.6 ± 22.3 |
| 350 | 83.3 ± 21.9 | 23.8 ± 21.0 | 83.0 ± 22.1 | 23.3 ± 20.4 |
| 400 | 91.7 ± 17.0 | 19.7 ± 19.3 | 91.7 ± 16.9 | 19.8 ± 19.0 |
| 450 | 95.1 ± 14.8 | 19.9 ± 18.8 | 95.4 ± 13.9 | 20.4 ± 19.2 |
| 500 | 95.9 ± 12.3 | 14.3 ± 12.6 | 95.8 ± 12.7 | 13.9 ± 12.1 |
| 550 | 100.0 ± 0.0 | 23.6 ± 25.8 | 100.0 ± 0.0 | 23.9 ± 26.3 |
| 600 | 100.0 ± 0.0 | 28.7 ± 16.4 | 100.0 ± 0.0 | 28.5 ± 16.4 |
| 650 | 100.0 ± 0.0 | 27.4 ± 30.2 | 100.0 ± 0.0 | 27.2 ± 30.1 |
| 700 | 100.0 ± 0.0 | 36.0 ± 30.0 | 100.0 ± 0.0 | 36.1 ± 29.7 |
| 750 | 100.0 ± 0.0 | 56.2 ± 33.1 | 100.0 ± 0.0 | 55.9 ± 33.0 |
| 800 | 100.0 ± 0.0 | 67.3 ± 30.7 | 100.0 ± 0.0 | 66.6 ± 30.6 |
| 850 | 99.9 ± 0.1 | 76.6 ± 33.5 | 99.8 ± 0.5 | 75.7 ± 33.1 |
| 900 | 99.6 ± 0.6 | 85.0 ± 22.5 | 99.4 ± 0.9 | 82.7 ± 23.2 |
| 950 | 99.2 ± 0.7 | 87.8 ± 16.8 | 98.5 ± 1.3 | 85.5 ± 16.3 |
| 1000 | 98.6 ± 0.7 | 89.5 ± 24.2 | 96.5 ± 2.2 | 86.2 ± 22.9 |
| 1050 | 97.6 ± 1.2 | 90.9 ± 15.9 | 94.2 ± 3.1 | 86.8 ± 14.6 |
| 1100 | 96.7 ± 1.1 | 89.4 ± 16.8 | 91.5 ± 3.3 | 82.9 ± 15.4 |
| 1150 | 95.4 ± 1.5 | 92.2 ± 6.2 | 88.4 ± 4.3 | 83.9 ± 6.3 |
| 1200 | 94.2 ± 1.6 | 90.4 ± 6.5 | 85.5 ± 4.6 | 79.8 ± 7.2 |