s2_e5
done · peak unseen_mismatch 89.1
α schedule
α 2.0 ×50 α 2.1304 ×50 α 2.2609 ×50 α 2.3913 ×50 α 2.5217 ×50 α 2.6522 ×50 α 2.7826 ×50 α 2.913 ×50 α 3.0435 ×50 α 3.1739 ×50 α 3.3043 ×50 α 3.4348 ×50 α 3.5652 ×50 α 3.6957 ×50 α 3.8261 ×50 α 3.9565 ×50 α 4.087 ×50 α 4.2174 ×50 α 4.3478 ×50 α 4.4783 ×50 α 4.6087 ×50 α 4.7391 ×50 α 4.8696 ×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 | 50.5 ± 1.9 | 49.9 ± 0.5 | 50.1 ± 1.9 | 50.1 ± 1.5 |
| 100 | 51.1 ± 2.6 | 50.4 ± 1.9 | 50.4 ± 1.6 | 49.0 ± 1.0 |
| 150 | 64.3 ± 15.9 | 59.6 ± 12.7 | 62.6 ± 15.1 | 56.5 ± 10.5 |
| 200 | 84.0 ± 17.0 | 77.7 ± 15.9 | 80.0 ± 19.5 | 72.7 ± 16.9 |
| 250 | 96.0 ± 4.8 | 93.5 ± 6.4 | 93.5 ± 6.4 | 89.1 ± 8.2 |
| 300 | 96.4 ± 1.8 | 94.6 ± 2.2 | 91.0 ± 5.4 | 87.8 ± 5.2 |
| 350 | 95.0 ± 1.5 | 93.0 ± 2.2 | 86.3 ± 5.1 | 81.9 ± 4.6 |
| 400 | 93.1 ± 1.0 | 91.0 ± 2.5 | 80.7 ± 4.4 | 76.1 ± 5.6 |
| 450 | 92.7 ± 0.8 | 90.8 ± 1.6 | 79.1 ± 3.1 | 74.8 ± 3.0 |
| 500 | 92.0 ± 1.2 | 90.1 ± 1.9 | 77.3 ± 3.4 | 72.6 ± 3.6 |
| 550 | 91.4 ± 1.9 | 90.1 ± 1.9 | 76.6 ± 3.5 | 72.8 ± 3.7 |
| 600 | 91.6 ± 1.2 | 89.8 ± 1.8 | 76.0 ± 4.0 | 71.9 ± 3.9 |
| 650 | 91.7 ± 1.6 | 90.2 ± 1.6 | 76.7 ± 5.7 | 73.2 ± 5.6 |
| 700 | 91.4 ± 1.7 | 90.1 ± 1.8 | 77.3 ± 4.5 | 72.2 ± 3.9 |
| 750 | 90.9 ± 1.8 | 89.7 ± 1.5 | 76.1 ± 5.5 | 72.3 ± 4.9 |
| 800 | 90.3 ± 1.6 | 89.2 ± 1.4 | 74.9 ± 4.3 | 71.6 ± 3.3 |
| 850 | 90.4 ± 1.5 | 89.5 ± 1.6 | 75.3 ± 5.1 | 71.5 ± 4.4 |
| 900 | 90.8 ± 1.5 | 89.7 ± 1.1 | 75.8 ± 4.6 | 72.2 ± 3.2 |
| 950 | 90.3 ± 1.4 | 88.9 ± 1.6 | 74.3 ± 5.0 | 71.6 ± 4.3 |
| 1000 | 90.4 ± 1.4 | 88.8 ± 1.5 | 73.7 ± 3.5 | 70.9 ± 3.6 |
| 1050 | 89.8 ± 1.8 | 88.7 ± 1.7 | 74.0 ± 5.1 | 70.5 ± 4.6 |
| 1100 | 90.0 ± 1.5 | 88.8 ± 2.0 | 74.3 ± 4.6 | 70.7 ± 4.6 |
| 1150 | 90.0 ± 1.6 | 88.8 ± 2.2 | 74.4 ± 4.3 | 71.4 ± 4.4 |
| 1200 | 89.5 ± 1.4 | 88.6 ± 1.5 | 73.2 ± 4.7 | 69.8 ± 3.8 |