s2_e4
done · peak unseen_mismatch 90.3
α schedule
α 2.0 ×50 α 2.087 ×50 α 2.1739 ×50 α 2.2609 ×50 α 2.3478 ×50 α 2.4348 ×50 α 2.5217 ×50 α 2.6087 ×50 α 2.6957 ×50 α 2.7826 ×50 α 2.8696 ×50 α 2.9565 ×50 α 3.0435 ×50 α 3.1304 ×50 α 3.2174 ×50 α 3.3043 ×50 α 3.3913 ×50 α 3.4783 ×50 α 3.5652 ×50 α 3.6522 ×50 α 3.7391 ×50 α 3.8261 ×50 α 3.913 ×50 α 4.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.0
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 ± 2.7 | 50.3 ± 1.8 | 50.6 ± 2.0 | 49.0 ± 1.1 |
| 150 | 62.8 ± 16.8 | 59.2 ± 12.2 | 62.0 ± 16.2 | 56.1 ± 9.9 |
| 200 | 83.4 ± 18.2 | 77.2 ± 18.0 | 80.1 ± 20.0 | 73.9 ± 18.3 |
| 250 | 97.3 ± 2.5 | 93.9 ± 6.9 | 95.0 ± 4.0 | 90.1 ± 6.6 |
| 300 | 97.6 ± 1.7 | 96.4 ± 1.6 | 93.6 ± 5.0 | 90.3 ± 4.9 |
| 350 | 96.2 ± 1.7 | 94.8 ± 1.8 | 88.2 ± 5.2 | 85.1 ± 5.3 |
| 400 | 94.3 ± 1.1 | 92.9 ± 1.7 | 83.7 ± 5.0 | 79.7 ± 5.0 |
| 450 | 93.4 ± 1.2 | 92.4 ± 1.2 | 79.8 ± 5.3 | 75.2 ± 5.1 |
| 500 | 93.3 ± 1.1 | 91.8 ± 1.3 | 79.7 ± 3.9 | 75.4 ± 4.3 |
| 550 | 92.8 ± 1.4 | 92.0 ± 1.6 | 79.2 ± 4.9 | 75.4 ± 5.0 |
| 600 | 92.7 ± 1.4 | 91.4 ± 1.4 | 78.5 ± 4.3 | 74.4 ± 4.8 |
| 650 | 92.8 ± 1.4 | 91.9 ± 1.6 | 78.7 ± 4.2 | 75.2 ± 4.4 |
| 700 | 92.3 ± 1.5 | 91.3 ± 1.4 | 77.2 ± 4.5 | 73.5 ± 4.7 |
| 750 | 92.0 ± 1.4 | 91.4 ± 2.1 | 77.4 ± 5.0 | 74.7 ± 5.6 |
| 800 | 92.2 ± 1.5 | 91.2 ± 1.5 | 78.2 ± 3.6 | 74.7 ± 3.7 |
| 850 | 92.1 ± 1.3 | 91.5 ± 1.7 | 77.7 ± 4.8 | 74.0 ± 5.4 |
| 900 | 91.9 ± 1.7 | 91.2 ± 1.7 | 78.0 ± 4.8 | 74.7 ± 5.1 |
| 950 | 92.0 ± 1.3 | 91.5 ± 1.6 | 77.1 ± 4.6 | 74.2 ± 5.2 |
| 1000 | 91.7 ± 1.7 | 91.2 ± 1.5 | 76.9 ± 3.9 | 74.0 ± 4.4 |
| 1050 | 91.7 ± 1.3 | 91.2 ± 1.6 | 77.7 ± 5.0 | 74.1 ± 5.1 |
| 1100 | 91.0 ± 0.7 | 90.5 ± 1.6 | 76.6 ± 4.2 | 73.2 ± 4.9 |
| 1150 | 91.5 ± 1.1 | 91.0 ± 1.4 | 77.3 ± 3.7 | 75.3 ± 4.1 |
| 1200 | 91.1 ± 1.2 | 90.2 ± 1.2 | 75.8 ± 4.6 | 73.3 ± 5.0 |