phi4star_ext
done · peak unseen_mismatch 95.2
α schedule
α 5.0 ×25 α 5.0 ×25 α 5.0 ×25 α 1.063 ×25 α 1.054 ×25 α 5.0 ×25 α 5.0 ×25 α 5.0 ×25 α 0.9043 ×25 α 0.8215 ×25 α 0.7122 ×25 α 3.084 ×25 α 2.128 ×25 α 0.1755 ×25 α 0.1755 ×50
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.1755
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 |
|---|---|---|---|---|
| 100 | 56.3 ± 6.7 | 51.7 ± 4.2 | 53.3 ± 6.2 | 49.3 ± 2.1 |
| 110 | 59.1 ± 9.8 | 51.8 ± 3.8 | 56.0 ± 9.5 | 49.4 ± 1.8 |
| 120 | 62.0 ± 14.4 | 52.7 ± 5.5 | 59.9 ± 14.6 | 50.6 ± 4.1 |
| 130 | 66.3 ± 17.5 | 55.5 ± 9.0 | 64.2 ± 17.3 | 54.3 ± 8.0 |
| 140 | 72.4 ± 17.7 | 61.7 ± 13.9 | 69.4 ± 17.5 | 59.4 ± 12.1 |
| 150 | 79.7 ± 15.0 | 67.2 ± 13.6 | 75.1 ± 16.6 | 64.2 ± 12.5 |
| 160 | 84.9 ± 13.1 | 77.1 ± 13.2 | 80.4 ± 14.7 | 72.5 ± 13.8 |
| 170 | 87.1 ± 10.6 | 79.9 ± 10.0 | 81.8 ± 12.5 | 73.9 ± 11.0 |
| 180 | 88.9 ± 7.8 | 84.3 ± 7.6 | 82.2 ± 11.6 | 77.0 ± 10.4 |
| 190 | 90.2 ± 5.9 | 87.0 ± 5.3 | 82.8 ± 9.5 | 78.1 ± 8.8 |
| 200 | 89.3 ± 5.3 | 86.4 ± 5.5 | 81.7 ± 8.5 | 77.3 ± 8.5 |
| 210 | 91.6 ± 5.8 | 89.1 ± 5.9 | 85.3 ± 7.6 | 81.3 ± 7.5 |
| 220 | 97.6 ± 2.5 | 95.3 ± 3.0 | 95.5 ± 3.8 | 91.7 ± 5.0 |
| 230 | 98.6 ± 1.7 | 94.7 ± 3.6 | 97.1 ± 3.5 | 91.1 ± 4.9 |
| 240 | 99.8 ± 0.2 | 97.7 ± 1.6 | 99.2 ± 0.6 | 95.2 ± 1.4 |