s0_e5_blk
done · peak unseen_mismatch 91.9
α schedule
α 0.0 ×200 α 1.0 ×200 α 2.0 ×200 α 3.0 ×200 α 4.0 ×200 α 5.0 ×200
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 | 50.4 ± 1.6 | 49.9 ± 1.1 | 49.8 ± 1.9 | 49.6 ± 1.9 |
| 100 | 50.3 ± 1.4 | 50.1 ± 2.2 | 49.8 ± 1.5 | 50.0 ± 1.9 |
| 150 | 52.9 ± 4.9 | 48.8 ± 3.2 | 53.3 ± 5.4 | 47.9 ± 2.6 |
| 200 | 57.3 ± 16.0 | 44.2 ± 13.4 | 57.9 ± 15.6 | 45.0 ± 14.2 |
| 250 | 67.0 ± 21.6 | 42.8 ± 11.2 | 66.8 ± 21.8 | 42.5 ± 11.0 |
| 300 | 84.6 ± 20.4 | 27.9 ± 18.2 | 83.9 ± 20.9 | 27.9 ± 18.1 |
| 350 | 91.4 ± 17.2 | 29.1 ± 28.1 | 91.3 ± 17.5 | 29.2 ± 28.0 |
| 400 | 99.5 ± 1.5 | 22.6 ± 22.6 | 99.4 ± 1.6 | 22.4 ± 23.1 |
| 450 | 100.0 ± 0.0 | 27.8 ± 28.6 | 100.0 ± 0.0 | 28.3 ± 28.2 |
| 500 | 100.0 ± 0.0 | 39.9 ± 26.6 | 100.0 ± 0.0 | 40.5 ± 26.7 |
| 550 | 100.0 ± 0.0 | 59.6 ± 40.2 | 100.0 ± 0.0 | 60.3 ± 40.2 |
| 600 | 100.0 ± 0.0 | 69.9 ± 34.1 | 100.0 ± 0.0 | 70.0 ± 33.8 |
| 650 | 99.7 ± 0.5 | 84.5 ± 19.6 | 99.3 ± 1.2 | 83.8 ± 17.9 |
| 700 | 99.3 ± 0.9 | 92.3 ± 14.3 | 98.3 ± 2.4 | 90.2 ± 15.9 |
| 750 | 98.8 ± 0.9 | 95.2 ± 6.7 | 96.8 ± 2.9 | 91.9 ± 8.6 |
| 800 | 97.9 ± 1.5 | 95.3 ± 3.0 | 94.2 ± 3.9 | 90.1 ± 6.1 |
| 850 | 95.4 ± 1.7 | 93.1 ± 2.5 | 89.2 ± 4.3 | 85.3 ± 3.9 |
| 900 | 93.7 ± 2.2 | 91.9 ± 2.6 | 85.7 ± 5.2 | 81.8 ± 5.3 |
| 950 | 92.0 ± 2.4 | 90.6 ± 2.0 | 81.9 ± 4.4 | 78.9 ± 4.0 |
| 1000 | 90.3 ± 1.6 | 89.2 ± 1.6 | 78.6 ± 2.9 | 75.5 ± 3.4 |
| 1050 | 89.1 ± 1.8 | 87.8 ± 2.2 | 75.4 ± 1.9 | 72.8 ± 2.9 |
| 1100 | 87.5 ± 2.2 | 86.0 ± 1.9 | 72.4 ± 2.7 | 70.5 ± 2.9 |
| 1150 | 86.7 ± 1.8 | 85.5 ± 1.8 | 70.8 ± 3.0 | 69.0 ± 2.8 |
| 1200 | 85.8 ± 1.8 | 84.1 ± 1.9 | 68.8 ± 4.0 | 66.3 ± 2.7 |