s4_e1
done · peak unseen_mismatch 94.0
α schedule
α 4.0 ×50 α 3.8696 ×50 α 3.7391 ×50 α 3.6087 ×50 α 3.4783 ×50 α 3.3478 ×50 α 3.2174 ×50 α 3.087 ×50 α 2.9565 ×50 α 2.8261 ×50 α 2.6957 ×50 α 2.5652 ×50 α 2.4348 ×50 α 2.3043 ×50 α 2.1739 ×50 α 2.0435 ×50 α 1.913 ×50 α 1.7826 ×50 α 1.6522 ×50 α 1.5217 ×50 α 1.3913 ×50 α 1.2609 ×50 α 1.1304 ×50 α 1.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 1.0
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 | 51.4 ± 1.2 | 50.2 ± 1.3 | 49.9 ± 1.6 | 50.4 ± 1.8 |
| 100 | 56.2 ± 4.9 | 53.9 ± 4.7 | 52.4 ± 2.9 | 50.5 ± 2.2 |
| 150 | 81.6 ± 5.6 | 77.5 ± 7.4 | 70.2 ± 7.9 | 64.5 ± 7.2 |
| 200 | 85.2 ± 4.1 | 83.8 ± 4.1 | 70.8 ± 7.7 | 68.4 ± 6.7 |
| 250 | 86.4 ± 2.5 | 85.5 ± 2.3 | 71.4 ± 4.6 | 68.0 ± 5.3 |
| 300 | 89.1 ± 1.9 | 88.5 ± 1.7 | 75.2 ± 4.0 | 72.2 ± 4.4 |
| 350 | 90.0 ± 1.6 | 89.0 ± 1.6 | 73.6 ± 4.7 | 70.6 ± 5.0 |
| 400 | 91.4 ± 1.7 | 89.9 ± 2.3 | 77.3 ± 4.8 | 73.3 ± 5.3 |
| 450 | 91.5 ± 1.3 | 90.4 ± 1.5 | 76.7 ± 3.7 | 72.5 ± 4.4 |
| 500 | 92.8 ± 1.6 | 91.9 ± 1.6 | 78.4 ± 3.6 | 74.7 ± 4.1 |
| 550 | 93.9 ± 1.1 | 92.6 ± 1.2 | 81.0 ± 4.4 | 75.4 ± 4.8 |
| 600 | 94.7 ± 1.7 | 93.0 ± 1.5 | 81.4 ± 5.2 | 76.2 ± 5.1 |
| 650 | 95.4 ± 1.3 | 93.7 ± 1.3 | 82.6 ± 3.8 | 77.2 ± 4.7 |
| 700 | 96.6 ± 1.0 | 94.5 ± 1.4 | 85.4 ± 3.7 | 79.7 ± 5.6 |
| 750 | 96.8 ± 1.1 | 95.7 ± 1.0 | 88.5 ± 3.8 | 83.8 ± 3.3 |
| 800 | 97.5 ± 1.0 | 96.4 ± 0.7 | 89.7 ± 2.1 | 83.9 ± 4.8 |
| 850 | 98.1 ± 0.8 | 96.7 ± 0.9 | 90.8 ± 3.5 | 84.8 ± 5.1 |
| 900 | 98.6 ± 0.6 | 97.5 ± 0.7 | 92.9 ± 2.4 | 86.1 ± 4.4 |
| 950 | 99.4 ± 0.5 | 98.1 ± 0.7 | 94.9 ± 2.6 | 87.4 ± 4.2 |
| 1000 | 99.6 ± 0.3 | 98.7 ± 0.5 | 96.5 ± 2.0 | 90.1 ± 5.6 |
| 1050 | 99.9 ± 0.1 | 99.0 ± 0.9 | 97.9 ± 1.3 | 90.0 ± 6.5 |
| 1100 | 99.9 ± 0.1 | 99.4 ± 0.4 | 98.4 ± 0.8 | 91.0 ± 6.8 |
| 1150 | 100.0 ± 0.1 | 99.6 ± 0.4 | 99.0 ± 0.7 | 93.2 ± 5.2 |
| 1200 | 100.0 ± 0.0 | 99.6 ± 0.4 | 99.7 ± 0.2 | 94.0 ± 5.1 |