s3_e3
done · peak unseen_mismatch 76.6
α schedule
α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50 α 3.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.0
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 | 51.3 ± 2.2 | 49.6 ± 0.8 | 50.0 ± 1.7 | 50.3 ± 1.6 |
| 100 | 56.5 ± 6.6 | 53.8 ± 4.2 | 53.7 ± 5.8 | 50.7 ± 2.5 |
| 150 | 76.1 ± 14.0 | 74.2 ± 13.1 | 68.9 ± 14.3 | 64.5 ± 11.9 |
| 200 | 89.4 ± 3.4 | 87.3 ± 3.9 | 77.6 ± 5.8 | 74.1 ± 6.6 |
| 250 | 90.1 ± 2.0 | 89.0 ± 2.0 | 76.2 ± 3.8 | 73.2 ± 3.8 |
| 300 | 92.2 ± 1.4 | 90.7 ± 1.1 | 77.6 ± 3.2 | 74.2 ± 3.5 |
| 350 | 92.1 ± 2.3 | 90.5 ± 1.9 | 77.0 ± 5.7 | 72.9 ± 4.9 |
| 400 | 91.8 ± 1.9 | 89.8 ± 1.8 | 76.8 ± 5.3 | 72.0 ± 4.6 |
| 450 | 92.9 ± 1.6 | 90.9 ± 1.6 | 78.4 ± 4.0 | 72.6 ± 4.1 |
| 500 | 93.9 ± 1.3 | 91.6 ± 1.3 | 78.6 ± 3.2 | 73.6 ± 3.4 |
| 550 | 93.3 ± 1.5 | 91.4 ± 1.3 | 79.0 ± 4.4 | 74.5 ± 4.7 |
| 600 | 93.3 ± 1.4 | 91.0 ± 1.4 | 79.0 ± 4.4 | 73.1 ± 4.9 |
| 650 | 93.9 ± 1.1 | 91.6 ± 1.1 | 79.9 ± 4.9 | 74.3 ± 6.1 |
| 700 | 93.8 ± 1.2 | 91.6 ± 1.5 | 78.9 ± 3.7 | 73.7 ± 4.6 |
| 750 | 93.8 ± 0.8 | 91.8 ± 1.0 | 80.3 ± 3.1 | 73.8 ± 4.5 |
| 800 | 93.8 ± 1.0 | 91.9 ± 1.2 | 79.2 ± 3.9 | 74.0 ± 4.5 |
| 850 | 93.8 ± 0.9 | 92.2 ± 1.4 | 80.3 ± 2.7 | 75.0 ± 3.9 |
| 900 | 94.5 ± 1.1 | 92.2 ± 1.4 | 81.6 ± 3.8 | 74.7 ± 5.3 |
| 950 | 94.3 ± 1.3 | 92.0 ± 1.2 | 80.6 ± 4.3 | 74.8 ± 4.4 |
| 1000 | 94.2 ± 0.8 | 92.2 ± 1.2 | 80.9 ± 2.7 | 75.1 ± 3.0 |
| 1050 | 94.5 ± 0.9 | 92.5 ± 1.7 | 82.1 ± 2.6 | 75.9 ± 4.6 |
| 1100 | 94.7 ± 0.6 | 92.5 ± 1.5 | 81.7 ± 3.6 | 75.4 ± 4.7 |
| 1150 | 94.8 ± 1.1 | 92.5 ± 1.5 | 82.7 ± 4.3 | 76.6 ± 5.5 |
| 1200 | 94.7 ± 0.9 | 92.6 ± 1.0 | 82.0 ± 3.1 | 75.2 ± 5.2 |