cand1
done · peak unseen_mismatch 97.7
α schedule
α 3.1128 ×25 α 3.0863 ×25 α 3.0348 ×25 α 2.9587 ×25 α 2.8585 ×25 α 2.7343 ×25 α 2.5864 ×25 α 2.4148 ×25 α 2.2199 ×25 α 2.0016 ×25 α 1.7601 ×25 α 1.4955 ×25 α 1.2078 ×25 α 0.8972 ×25 α 0.5636 ×25 α 0.2073 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.2073
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 ± 1.6 | 50.1 ± 1.8 | 49.8 ± 1.7 |
| 50 | 51.6 ± 1.8 | 50.2 ± 1.5 | 50.3 ± 1.7 | 49.9 ± 1.7 |
| 75 | 52.2 ± 2.5 | 50.3 ± 2.0 | 50.7 ± 2.1 | 49.9 ± 1.6 |
| 100 | 57.8 ± 7.4 | 53.4 ± 4.9 | 53.7 ± 6.1 | 50.9 ± 3.2 |
| 125 | 69.6 ± 12.3 | 63.8 ± 10.3 | 62.6 ± 12.5 | 57.3 ± 8.5 |
| 150 | 82.2 ± 11.2 | 77.4 ± 10.3 | 73.0 ± 13.2 | 67.3 ± 10.7 |
| 175 | 88.8 ± 7.7 | 85.1 ± 7.7 | 79.0 ± 10.7 | 74.2 ± 9.7 |
| 200 | 91.9 ± 5.8 | 89.2 ± 5.6 | 81.9 ± 9.3 | 77.9 ± 8.5 |
| 225 | 93.7 ± 4.7 | 91.5 ± 4.5 | 83.7 ± 8.0 | 79.8 ± 7.6 |
| 250 | 95.3 ± 4.0 | 93.6 ± 4.0 | 86.2 ± 6.7 | 82.5 ± 6.6 |
| 275 | 96.9 ± 3.6 | 95.3 ± 3.4 | 89.7 ± 5.2 | 85.6 ± 5.7 |
| 300 | 98.4 ± 1.9 | 97.1 ± 2.7 | 93.1 ± 3.6 | 88.9 ± 4.7 |
| 325 | 99.6 ± 0.5 | 98.7 ± 1.3 | 97.0 ± 2.2 | 93.2 ± 4.4 |
| 350 | 99.9 ± 0.1 | 99.5 ± 1.3 | 99.4 ± 0.7 | 97.2 ± 3.0 |
| 375 | 100.0 ± 0.0 | 99.2 ± 3.0 | 100.0 ± 0.1 | 97.7 ± 4.9 |
| 400 | 100.0 ± 0.0 | 90.0 ± 16.0 | 100.0 ± 0.1 | 86.8 ± 18.3 |