s0_e5
done · peak unseen_mismatch 82.4
α schedule
α 0.0 ×50 α 0.2174 ×50 α 0.4348 ×50 α 0.6522 ×50 α 0.8696 ×50 α 1.087 ×50 α 1.3043 ×50 α 1.5217 ×50 α 1.7391 ×50 α 1.9565 ×50 α 2.1739 ×50 α 2.3913 ×50 α 2.6087 ×50 α 2.8261 ×50 α 3.0435 ×50 α 3.2609 ×50 α 3.4783 ×50 α 3.6957 ×50 α 3.913 ×50 α 4.1304 ×50 α 4.3478 ×50 α 4.5652 ×50 α 4.7826 ×50 α 5.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 5.0
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 | 50.2 ± 1.4 | 50.0 ± 1.9 | 49.3 ± 1.4 | 49.6 ± 0.9 |
| 150 | 50.5 ± 3.6 | 49.0 ± 3.0 | 51.4 ± 3.5 | 48.4 ± 2.6 |
| 200 | 57.1 ± 14.6 | 45.0 ± 12.2 | 56.5 ± 14.7 | 44.7 ± 12.4 |
| 250 | 66.8 ± 20.6 | 43.4 ± 7.8 | 66.2 ± 20.8 | 45.0 ± 8.3 |
| 300 | 71.7 ± 23.3 | 40.8 ± 16.9 | 71.5 ± 23.5 | 42.0 ± 17.8 |
| 350 | 82.7 ± 22.3 | 44.6 ± 13.8 | 82.3 ± 22.6 | 45.3 ± 13.7 |
| 400 | 92.0 ± 12.3 | 37.2 ± 19.1 | 91.3 ± 13.4 | 39.1 ± 18.8 |
| 450 | 100.0 ± 0.1 | 38.5 ± 29.2 | 100.0 ± 0.1 | 39.2 ± 28.4 |
| 500 | 99.9 ± 0.2 | 41.4 ± 37.6 | 99.9 ± 0.2 | 42.1 ± 37.4 |
| 550 | 99.8 ± 0.5 | 71.8 ± 33.6 | 99.6 ± 1.3 | 71.7 ± 32.8 |
| 600 | 99.7 ± 1.0 | 69.9 ± 44.0 | 99.2 ± 2.2 | 68.8 ± 43.3 |
| 650 | 99.4 ± 1.2 | 81.1 ± 29.0 | 97.9 ± 4.0 | 78.5 ± 28.3 |
| 700 | 99.0 ± 1.2 | 80.9 ± 31.1 | 96.5 ± 4.7 | 77.5 ± 30.0 |
| 750 | 98.2 ± 1.6 | 84.3 ± 23.3 | 94.7 ± 4.5 | 80.5 ± 21.1 |
| 800 | 96.5 ± 2.5 | 84.1 ± 27.5 | 91.5 ± 5.8 | 77.8 ± 25.5 |
| 850 | 95.1 ± 2.3 | 91.0 ± 4.9 | 88.1 ± 5.8 | 82.4 ± 5.1 |
| 900 | 93.8 ± 2.4 | 89.7 ± 7.5 | 84.4 ± 6.7 | 79.7 ± 5.4 |
| 950 | 92.2 ± 2.7 | 90.9 ± 2.7 | 81.7 ± 6.6 | 78.9 ± 5.7 |
| 1000 | 90.9 ± 2.5 | 90.0 ± 2.8 | 78.1 ± 6.5 | 77.2 ± 5.9 |
| 1050 | 89.8 ± 2.6 | 89.2 ± 3.1 | 75.6 ± 6.0 | 74.9 ± 6.3 |
| 1100 | 88.2 ± 1.7 | 87.7 ± 1.9 | 72.7 ± 5.2 | 72.2 ± 4.1 |
| 1150 | 87.1 ± 1.3 | 87.2 ± 1.8 | 71.5 ± 4.7 | 70.9 ± 4.3 |
| 1200 | 86.1 ± 1.0 | 86.1 ± 1.5 | 69.3 ± 3.9 | 68.7 ± 3.5 |