s4_e4
done · peak unseen_mismatch 72.1
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 | 51.4 ± 1.2 | 50.2 ± 1.3 | 49.9 ± 1.6 | 50.4 ± 1.8 |
| 100 | 56.3 ± 4.5 | 53.9 ± 4.3 | 52.0 ± 2.7 | 50.4 ± 2.2 |
| 150 | 80.7 ± 5.4 | 77.4 ± 6.5 | 70.0 ± 7.5 | 64.9 ± 6.8 |
| 200 | 83.8 ± 3.4 | 82.4 ± 3.3 | 69.0 ± 6.1 | 66.3 ± 5.4 |
| 250 | 84.5 ± 2.9 | 84.0 ± 2.8 | 68.7 ± 5.6 | 65.9 ± 5.8 |
| 300 | 86.4 ± 2.2 | 85.8 ± 2.4 | 71.3 ± 4.6 | 69.2 ± 5.1 |
| 350 | 87.5 ± 1.4 | 86.3 ± 1.6 | 71.0 ± 3.8 | 68.5 ± 3.9 |
| 400 | 88.3 ± 2.1 | 87.0 ± 2.6 | 72.4 ± 3.9 | 69.6 ± 4.8 |
| 450 | 88.0 ± 1.7 | 86.8 ± 2.0 | 71.1 ± 3.9 | 68.4 ± 5.0 |
| 500 | 89.0 ± 2.0 | 87.2 ± 1.9 | 72.5 ± 4.7 | 68.2 ± 5.1 |
| 550 | 88.7 ± 1.5 | 86.9 ± 2.3 | 72.7 ± 2.3 | 67.8 ± 3.2 |
| 600 | 89.3 ± 1.7 | 87.5 ± 2.1 | 73.3 ± 5.2 | 69.5 ± 4.9 |
| 650 | 89.4 ± 1.2 | 88.0 ± 1.6 | 73.4 ± 3.6 | 68.1 ± 4.4 |
| 700 | 89.6 ± 2.2 | 87.8 ± 1.7 | 74.7 ± 4.3 | 68.8 ± 4.9 |
| 750 | 89.5 ± 2.0 | 88.0 ± 1.7 | 74.2 ± 4.3 | 70.2 ± 4.5 |
| 800 | 90.5 ± 1.9 | 89.0 ± 2.1 | 75.9 ± 3.8 | 71.1 ± 4.8 |
| 850 | 90.2 ± 2.3 | 88.6 ± 2.3 | 75.7 ± 4.6 | 71.2 ± 5.7 |
| 900 | 90.5 ± 1.7 | 89.2 ± 1.8 | 75.0 ± 2.6 | 70.8 ± 2.7 |
| 950 | 90.5 ± 2.1 | 88.8 ± 1.8 | 74.9 ± 4.0 | 70.0 ± 4.7 |
| 1000 | 90.9 ± 1.6 | 88.9 ± 1.7 | 75.6 ± 3.6 | 70.3 ± 4.7 |
| 1050 | 90.8 ± 1.8 | 88.9 ± 1.8 | 74.8 ± 4.7 | 69.1 ± 5.4 |
| 1100 | 91.0 ± 2.0 | 89.7 ± 2.0 | 76.0 ± 4.7 | 70.7 ± 6.2 |
| 1150 | 91.1 ± 1.8 | 89.6 ± 1.8 | 76.8 ± 2.9 | 72.1 ± 4.0 |
| 1200 | 91.2 ± 1.7 | 89.5 ± 1.5 | 76.0 ± 3.5 | 71.4 ± 5.4 |