s3_e3
done · peak unseen_mismatch 79.4
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 | 50.6 ± 1.2 | 49.8 ± 1.4 | 49.8 ± 1.6 | 51.1 ± 1.2 |
| 100 | 52.9 ± 2.7 | 51.0 ± 2.3 | 52.1 ± 1.3 | 50.1 ± 1.3 |
| 150 | 77.7 ± 8.1 | 70.5 ± 10.3 | 71.5 ± 10.1 | 64.1 ± 10.0 |
| 200 | 90.8 ± 4.6 | 88.3 ± 3.6 | 81.6 ± 9.1 | 77.0 ± 7.5 |
| 250 | 91.3 ± 2.7 | 90.1 ± 2.5 | 80.7 ± 6.4 | 76.6 ± 6.1 |
| 300 | 91.5 ± 1.9 | 90.5 ± 1.6 | 80.0 ± 4.5 | 76.2 ± 5.0 |
| 350 | 91.4 ± 1.4 | 89.9 ± 1.8 | 77.2 ± 3.1 | 72.8 ± 4.5 |
| 400 | 93.2 ± 1.0 | 90.9 ± 1.8 | 79.6 ± 3.7 | 75.3 ± 5.3 |
| 450 | 92.0 ± 0.9 | 90.5 ± 1.3 | 77.4 ± 3.5 | 72.7 ± 4.9 |
| 500 | 92.8 ± 1.3 | 90.8 ± 2.0 | 78.2 ± 4.0 | 73.9 ± 5.5 |
| 550 | 92.8 ± 0.9 | 90.7 ± 1.2 | 79.2 ± 3.2 | 73.3 ± 4.9 |
| 600 | 93.5 ± 0.8 | 91.6 ± 1.4 | 79.6 ± 3.3 | 74.5 ± 4.6 |
| 650 | 93.7 ± 0.9 | 91.6 ± 1.4 | 80.6 ± 2.9 | 74.2 ± 4.9 |
| 700 | 94.0 ± 1.2 | 92.3 ± 1.7 | 81.5 ± 3.7 | 77.2 ± 4.3 |
| 750 | 93.3 ± 1.0 | 91.7 ± 1.3 | 80.2 ± 2.9 | 74.8 ± 4.8 |
| 800 | 93.8 ± 1.2 | 92.6 ± 1.0 | 81.7 ± 3.5 | 77.1 ± 4.4 |
| 850 | 94.1 ± 1.0 | 92.7 ± 1.2 | 82.0 ± 3.5 | 76.3 ± 5.2 |
| 900 | 93.7 ± 1.2 | 92.2 ± 1.0 | 80.3 ± 4.1 | 75.0 ± 5.8 |
| 950 | 94.5 ± 0.9 | 92.7 ± 1.5 | 81.1 ± 2.5 | 76.0 ± 4.2 |
| 1000 | 94.3 ± 0.8 | 93.0 ± 1.1 | 81.2 ± 3.5 | 76.8 ± 4.4 |
| 1050 | 94.3 ± 0.7 | 92.8 ± 1.4 | 80.3 ± 3.2 | 75.0 ± 4.7 |
| 1100 | 94.4 ± 0.6 | 92.8 ± 1.1 | 82.1 ± 2.6 | 77.4 ± 4.0 |
| 1150 | 94.3 ± 0.8 | 93.1 ± 1.4 | 82.5 ± 2.7 | 78.3 ± 4.8 |
| 1200 | 95.4 ± 0.9 | 93.8 ± 1.4 | 83.6 ± 2.4 | 79.4 ± 3.8 |