anneal_3_0
done · peak unseen_mismatch 90.5
Referenced by
α schedule
α 3.0 ×150 α 2.5714 ×150 α 2.1429 ×150 α 1.7143 ×150 α 1.2857 ×150 α 0.8571 ×150 α 0.4286 ×150 α 0.0 ×150
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 |
|---|---|---|---|---|
| 150 | 70.1 ± 16.1 | 62.2 ± 12.3 | 67.1 ± 15.6 | 60.3 ± 10.7 |
| 300 | 97.7 ± 1.9 | 96.4 ± 3.1 | 94.1 ± 4.6 | 90.5 ± 8.3 |
| 450 | 97.0 ± 1.3 | 94.7 ± 2.1 | 90.6 ± 5.4 | 84.5 ± 6.0 |
| 600 | 98.7 ± 0.6 | 96.0 ± 2.1 | 92.6 ± 2.6 | 83.9 ± 7.2 |
| 750 | 99.8 ± 0.2 | 98.2 ± 1.4 | 96.6 ± 3.0 | 88.2 ± 6.2 |
| 900 | 100.0 ± 0.0 | 91.3 ± 8.9 | 99.7 ± 0.4 | 76.8 ± 17.7 |
| 1050 | 100.0 ± 0.0 | 27.9 ± 12.6 | 99.9 ± 0.1 | 24.9 ± 12.4 |
| 1200 | 100.0 ± 0.0 | 7.3 ± 5.7 | 99.8 ± 0.3 | 10.4 ± 7.0 |