p12
done · peak unseen_mismatch 68.9
α schedule
α 4.1934 ×25 α 4.1259 ×25 α 4.0685 ×25 α 4.0161 ×25 α 3.967 ×25 α 3.9202 ×25 α 3.8752 ×25 α 3.8317 ×25 α 3.7895 ×25 α 3.7484 ×25 α 3.7082 ×25 α 3.6689 ×25 α 3.6303 ×25 α 3.5924 ×25 α 3.5551 ×25 α 3.5183 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 3.5183
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 | 50.8 ± 1.2 | 50.5 ± 1.5 | 49.9 ± 1.5 | 49.5 ± 1.8 |
| 50 | 51.6 ± 1.6 | 50.4 ± 1.6 | 50.0 ± 1.0 | 49.8 ± 1.6 |
| 75 | 52.7 ± 2.2 | 50.3 ± 1.9 | 50.7 ± 1.8 | 49.4 ± 1.5 |
| 100 | 59.1 ± 8.1 | 56.9 ± 5.7 | 53.1 ± 5.0 | 51.2 ± 3.1 |
| 125 | 71.2 ± 10.9 | 67.8 ± 10.2 | 58.8 ± 10.0 | 56.9 ± 8.5 |
| 150 | 78.6 ± 6.3 | 75.8 ± 6.2 | 63.9 ± 8.4 | 61.3 ± 7.6 |
| 175 | 82.1 ± 4.7 | 80.1 ± 4.3 | 66.2 ± 8.5 | 64.0 ± 8.2 |
| 200 | 83.2 ± 3.3 | 81.3 ± 3.3 | 65.3 ± 6.0 | 63.2 ± 6.2 |
| 225 | 84.2 ± 3.1 | 82.0 ± 2.7 | 66.3 ± 5.7 | 64.3 ± 5.8 |
| 250 | 84.7 ± 2.5 | 82.9 ± 1.9 | 66.2 ± 4.9 | 64.4 ± 5.0 |
| 275 | 85.6 ± 2.8 | 84.1 ± 2.8 | 67.5 ± 5.5 | 65.5 ± 5.6 |
| 300 | 87.1 ± 2.7 | 85.5 ± 2.4 | 69.7 ± 5.8 | 67.5 ± 5.5 |
| 325 | 87.3 ± 2.2 | 85.8 ± 2.2 | 69.6 ± 4.3 | 67.4 ± 4.3 |
| 350 | 88.2 ± 2.1 | 86.9 ± 1.9 | 70.9 ± 4.3 | 68.4 ± 4.3 |
| 375 | 88.6 ± 1.9 | 86.8 ± 2.0 | 70.7 ± 4.6 | 67.9 ± 5.0 |
| 400 | 88.9 ± 2.2 | 87.5 ± 1.9 | 71.9 ± 4.0 | 68.9 ± 4.1 |