cand2
done · peak unseen_mismatch 96.8
α schedule
α 4.2358 ×25 α 4.0116 ×25 α 3.7763 ×25 α 3.5344 ×25 α 3.2878 ×25 α 3.0374 ×25 α 2.7839 ×25 α 2.5278 ×25 α 2.2694 ×25 α 2.009 ×25 α 1.7467 ×25 α 1.4827 ×25 α 1.2172 ×25 α 0.9502 ×25 α 0.682 ×25 α 0.4124 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.4124
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 |
|---|---|---|---|---|
| 25 | 51.0 ± 1.6 | 50.0 ± 1.5 | 49.9 ± 1.8 | 49.9 ± 1.8 |
| 50 | 51.8 ± 1.9 | 50.3 ± 1.6 | 50.3 ± 1.8 | 49.9 ± 1.7 |
| 75 | 53.2 ± 3.0 | 50.9 ± 2.3 | 51.0 ± 2.2 | 50.0 ± 1.7 |
| 100 | 61.5 ± 8.4 | 56.9 ± 6.3 | 55.0 ± 6.3 | 51.8 ± 4.1 |
| 125 | 75.0 ± 10.5 | 70.2 ± 9.5 | 64.1 ± 10.8 | 59.7 ± 7.9 |
| 150 | 83.0 ± 7.7 | 79.7 ± 7.3 | 69.7 ± 10.1 | 65.7 ± 8.4 |
| 175 | 86.8 ± 5.3 | 84.0 ± 5.0 | 73.1 ± 8.2 | 69.6 ± 7.4 |
| 200 | 89.6 ± 3.6 | 87.2 ± 3.2 | 75.9 ± 7.3 | 72.2 ± 6.6 |
| 225 | 92.1 ± 2.7 | 90.1 ± 2.6 | 78.7 ± 6.5 | 75.2 ± 5.8 |
| 250 | 94.5 ± 2.0 | 92.8 ± 2.0 | 83.0 ± 5.4 | 79.4 ± 4.8 |
| 275 | 96.8 ± 1.2 | 95.3 ± 1.3 | 87.9 ± 3.9 | 83.9 ± 4.2 |
| 300 | 98.5 ± 0.7 | 97.2 ± 1.0 | 92.4 ± 3.0 | 88.3 ± 3.9 |
| 325 | 99.5 ± 0.4 | 98.7 ± 0.8 | 96.5 ± 2.2 | 92.9 ± 3.7 |
| 350 | 99.9 ± 0.1 | 99.4 ± 0.7 | 99.0 ± 1.0 | 96.1 ± 3.6 |
| 375 | 100.0 ± 0.0 | 99.2 ± 3.5 | 99.8 ± 0.4 | 96.8 ± 5.9 |
| 400 | 100.0 ± 0.0 | 96.2 ± 9.4 | 100.0 ± 0.1 | 92.6 ± 13.0 |