s2_e0
done · peak unseen_mismatch 98.1
α schedule
α 2.0 ×50 α 1.913 ×50 α 1.8261 ×50 α 1.7391 ×50 α 1.6522 ×50 α 1.5652 ×50 α 1.4783 ×50 α 1.3913 ×50 α 1.3043 ×50 α 1.2174 ×50 α 1.1304 ×50 α 1.0435 ×50 α 0.9565 ×50 α 0.8696 ×50 α 0.7826 ×50 α 0.6957 ×50 α 0.6087 ×50 α 0.5217 ×50 α 0.4348 ×50 α 0.3478 ×50 α 0.2609 ×50 α 0.1739 ×50 α 0.087 ×50 α 0.0 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.0
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 ± 2.0 | 50.1 ± 2.2 | 50.6 ± 1.7 | 49.1 ± 1.1 |
| 150 | 61.2 ± 16.1 | 54.3 ± 7.0 | 60.6 ± 14.8 | 52.9 ± 4.6 |
| 200 | 83.5 ± 17.3 | 70.4 ± 16.1 | 81.5 ± 19.2 | 67.6 ± 15.4 |
| 250 | 96.7 ± 9.6 | 86.9 ± 14.4 | 96.1 ± 10.8 | 85.5 ± 15.2 |
| 300 | 100.0 ± 0.1 | 91.1 ± 15.0 | 100.0 ± 0.1 | 90.9 ± 14.6 |
| 350 | 99.9 ± 0.1 | 92.8 ± 12.8 | 99.7 ± 0.6 | 91.6 ± 13.1 |
| 400 | 99.9 ± 0.1 | 97.9 ± 4.2 | 99.6 ± 0.9 | 96.1 ± 5.4 |
| 450 | 99.9 ± 0.1 | 99.4 ± 0.6 | 99.7 ± 0.4 | 98.0 ± 1.6 |
| 500 | 99.9 ± 0.1 | 98.7 ± 2.5 | 99.7 ± 0.3 | 96.4 ± 4.9 |
| 550 | 100.0 ± 0.1 | 99.5 ± 0.6 | 99.5 ± 0.6 | 97.3 ± 2.1 |
| 600 | 100.0 ± 0.1 | 99.5 ± 0.7 | 99.9 ± 0.2 | 98.1 ± 2.8 |
| 650 | 100.0 ± 0.0 | 98.4 ± 2.9 | 99.9 ± 0.2 | 96.5 ± 4.4 |
| 700 | 100.0 ± 0.0 | 99.2 ± 1.4 | 99.8 ± 0.3 | 96.6 ± 4.9 |
| 750 | 100.0 ± 0.0 | 97.4 ± 3.4 | 99.9 ± 0.1 | 93.1 ± 6.2 |
| 800 | 100.0 ± 0.0 | 88.0 ± 16.7 | 99.9 ± 0.2 | 83.6 ± 20.2 |
| 850 | 100.0 ± 0.0 | 86.7 ± 12.7 | 100.0 ± 0.1 | 82.9 ± 14.5 |
| 900 | 100.0 ± 0.0 | 74.8 ± 17.1 | 100.0 ± 0.1 | 70.2 ± 18.6 |
| 950 | 100.0 ± 0.0 | 64.5 ± 26.3 | 99.9 ± 0.1 | 61.7 ± 23.5 |
| 1000 | 100.0 ± 0.0 | 42.1 ± 24.8 | 100.0 ± 0.1 | 41.5 ± 21.7 |
| 1050 | 100.0 ± 0.0 | 37.3 ± 10.0 | 100.0 ± 0.1 | 37.3 ± 7.8 |
| 1100 | 100.0 ± 0.0 | 22.0 ± 12.0 | 99.9 ± 0.2 | 24.3 ± 10.4 |
| 1150 | 100.0 ± 0.0 | 20.7 ± 16.1 | 99.8 ± 0.4 | 24.0 ± 13.4 |
| 1200 | 100.0 ± 0.0 | 15.6 ± 12.3 | 99.1 ± 1.5 | 19.8 ± 11.9 |