s1_e4_blk
done · peak unseen_mismatch 91.7
α schedule
α 1.0 ×300 α 2.0 ×300 α 3.0 ×300 α 4.0 ×300
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.0
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 ± 1.3 | 49.1 ± 0.7 | 50.0 ± 2.0 | 50.2 ± 1.2 |
| 100 | 50.1 ± 1.7 | 50.3 ± 2.2 | 50.3 ± 1.8 | 50.0 ± 1.8 |
| 150 | 51.2 ± 1.8 | 49.7 ± 1.9 | 51.1 ± 1.2 | 48.5 ± 1.1 |
| 200 | 57.8 ± 15.6 | 46.8 ± 8.9 | 58.3 ± 15.1 | 47.6 ± 9.7 |
| 250 | 71.9 ± 20.1 | 41.9 ± 10.9 | 72.1 ± 19.3 | 41.2 ± 9.8 |
| 300 | 88.1 ± 18.5 | 32.0 ± 21.6 | 88.1 ± 17.7 | 32.5 ± 20.6 |
| 350 | 99.9 ± 0.3 | 34.7 ± 20.7 | 99.9 ± 0.3 | 35.4 ± 21.4 |
| 400 | 100.0 ± 0.0 | 52.5 ± 30.4 | 100.0 ± 0.0 | 53.2 ± 30.3 |
| 450 | 100.0 ± 0.0 | 64.6 ± 24.9 | 100.0 ± 0.0 | 65.9 ± 22.6 |
| 500 | 99.8 ± 0.5 | 78.4 ± 20.4 | 99.4 ± 1.9 | 78.0 ± 19.2 |
| 550 | 99.7 ± 0.6 | 91.8 ± 11.6 | 99.4 ± 1.2 | 91.7 ± 11.1 |
| 600 | 99.7 ± 0.6 | 91.2 ± 15.0 | 98.8 ± 2.2 | 89.6 ± 13.6 |
| 650 | 98.5 ± 1.6 | 93.3 ± 10.6 | 96.8 ± 3.3 | 89.5 ± 11.1 |
| 700 | 97.5 ± 1.5 | 96.0 ± 1.8 | 94.2 ± 3.1 | 89.9 ± 3.0 |
| 750 | 96.5 ± 1.3 | 94.8 ± 2.1 | 91.0 ± 3.5 | 86.6 ± 3.6 |
| 800 | 95.2 ± 1.4 | 93.6 ± 2.0 | 87.9 ± 3.3 | 83.8 ± 4.0 |
| 850 | 93.6 ± 1.8 | 92.4 ± 1.9 | 84.6 ± 4.3 | 81.0 ± 5.0 |
| 900 | 93.0 ± 1.8 | 91.6 ± 2.1 | 81.6 ± 4.4 | 78.8 ± 4.8 |
| 950 | 91.3 ± 1.3 | 89.9 ± 1.9 | 79.1 ± 4.2 | 76.2 ± 4.2 |
| 1000 | 89.8 ± 1.7 | 88.8 ± 1.9 | 76.0 ± 4.4 | 73.1 ± 4.7 |
| 1050 | 89.0 ± 1.6 | 88.1 ± 2.4 | 74.3 ± 3.7 | 71.5 ± 4.1 |
| 1100 | 88.5 ± 1.4 | 87.5 ± 1.8 | 73.6 ± 3.6 | 70.5 ± 4.1 |
| 1150 | 88.5 ± 1.4 | 87.5 ± 1.5 | 72.2 ± 2.9 | 69.2 ± 3.5 |
| 1200 | 88.2 ± 1.3 | 87.1 ± 1.7 | 72.1 ± 3.5 | 68.9 ± 2.7 |