s4_e4
done · peak unseen_mismatch 69.8
α schedule
α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×50 α 4.0 ×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 | 51.7 ± 1.9 | 49.6 ± 1.0 | 49.9 ± 2.1 | 50.5 ± 1.2 |
| 100 | 58.6 ± 8.4 | 56.8 ± 7.1 | 54.0 ± 5.3 | 52.7 ± 4.4 |
| 150 | 77.6 ± 5.5 | 76.5 ± 5.5 | 64.2 ± 5.9 | 61.6 ± 6.1 |
| 200 | 82.4 ± 1.5 | 81.0 ± 1.4 | 66.1 ± 3.0 | 63.7 ± 3.6 |
| 250 | 85.1 ± 2.3 | 83.2 ± 2.0 | 67.0 ± 4.7 | 65.2 ± 3.5 |
| 300 | 87.0 ± 1.1 | 85.6 ± 1.2 | 69.6 ± 3.4 | 66.6 ± 3.9 |
| 350 | 88.2 ± 1.9 | 86.1 ± 1.6 | 70.2 ± 4.0 | 67.7 ± 5.0 |
| 400 | 87.8 ± 2.0 | 85.6 ± 1.9 | 70.0 ± 4.0 | 66.0 ± 3.7 |
| 450 | 88.2 ± 1.5 | 86.0 ± 2.0 | 70.5 ± 4.2 | 67.0 ± 4.8 |
| 500 | 89.5 ± 1.2 | 87.4 ± 1.8 | 72.1 ± 1.6 | 67.2 ± 3.1 |
| 550 | 89.0 ± 1.9 | 87.3 ± 2.2 | 71.6 ± 4.2 | 68.0 ± 4.5 |
| 600 | 89.2 ± 2.1 | 87.3 ± 2.0 | 71.7 ± 3.8 | 67.1 ± 2.8 |
| 650 | 89.9 ± 1.7 | 87.7 ± 1.1 | 72.4 ± 3.7 | 67.0 ± 3.1 |
| 700 | 89.1 ± 1.0 | 86.9 ± 1.9 | 71.3 ± 3.2 | 66.2 ± 3.2 |
| 750 | 89.8 ± 1.0 | 87.3 ± 1.2 | 73.7 ± 2.7 | 68.0 ± 3.5 |
| 800 | 89.2 ± 1.6 | 86.5 ± 1.4 | 70.9 ± 3.7 | 65.5 ± 3.8 |
| 850 | 90.2 ± 0.9 | 88.1 ± 1.5 | 72.4 ± 3.3 | 67.4 ± 4.1 |
| 900 | 90.4 ± 1.4 | 88.0 ± 1.4 | 73.7 ± 2.2 | 68.3 ± 3.0 |
| 950 | 90.5 ± 1.5 | 87.8 ± 0.7 | 74.1 ± 3.2 | 68.7 ± 3.1 |
| 1000 | 90.6 ± 1.7 | 88.5 ± 1.5 | 73.8 ± 2.2 | 69.0 ± 2.4 |
| 1050 | 90.6 ± 0.8 | 88.2 ± 1.7 | 74.3 ± 3.3 | 68.3 ± 4.5 |
| 1100 | 90.9 ± 1.3 | 88.2 ± 1.6 | 74.8 ± 3.5 | 68.5 ± 4.4 |
| 1150 | 90.9 ± 1.2 | 88.3 ± 1.3 | 75.8 ± 2.8 | 68.7 ± 4.1 |
| 1200 | 91.1 ± 1.0 | 88.4 ± 1.2 | 75.0 ± 3.0 | 69.8 ± 3.6 |