r4_c6
done · peak unseen_mismatch 98.5
α schedule
α 2.7429 ×25 α 2.7423 ×25 α 2.7391 ×25 α 2.7306 ×25 α 2.7133 ×25 α 2.6833 ×25 α 2.6361 ×25 α 2.5669 ×25 α 2.4704 ×25 α 2.3411 ×25 α 2.1731 ×25 α 1.9602 ×25 α 1.6957 ×25 α 1.373 ×25 α 0.9849 ×25 α 0.524 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.524
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 ± 1.3 | 50.4 ± 1.8 | 50.1 ± 1.3 | 49.4 ± 1.8 |
| 50 | 51.4 ± 1.4 | 50.3 ± 1.7 | 49.8 ± 1.2 | 50.0 ± 1.7 |
| 75 | 51.7 ± 1.6 | 49.8 ± 1.7 | 50.3 ± 1.6 | 49.5 ± 1.7 |
| 100 | 53.5 ± 5.0 | 51.3 ± 2.2 | 51.6 ± 3.7 | 49.7 ± 1.5 |
| 125 | 63.5 ± 12.7 | 58.3 ± 9.3 | 57.9 ± 11.7 | 53.2 ± 7.1 |
| 150 | 73.0 ± 16.1 | 69.7 ± 14.2 | 67.2 ± 15.4 | 62.9 ± 13.0 |
| 175 | 82.5 ± 11.4 | 78.6 ± 11.9 | 73.7 ± 12.5 | 69.1 ± 11.4 |
| 200 | 89.4 ± 7.6 | 86.4 ± 7.4 | 80.6 ± 9.5 | 76.5 ± 9.2 |
| 225 | 93.7 ± 3.4 | 90.9 ± 3.9 | 84.7 ± 7.8 | 80.7 ± 8.1 |
| 250 | 94.5 ± 2.7 | 92.5 ± 2.7 | 85.9 ± 6.8 | 82.5 ± 7.2 |
| 275 | 95.6 ± 1.9 | 93.8 ± 2.0 | 86.9 ± 5.7 | 83.5 ± 6.2 |
| 300 | 96.6 ± 1.5 | 95.0 ± 1.7 | 88.9 ± 5.2 | 85.3 ± 5.5 |
| 325 | 97.7 ± 1.0 | 96.4 ± 1.2 | 91.1 ± 4.3 | 86.9 ± 5.0 |
| 350 | 99.3 ± 0.4 | 98.4 ± 0.9 | 95.7 ± 2.4 | 92.6 ± 3.6 |
| 375 | 99.9 ± 0.1 | 99.7 ± 0.3 | 99.4 ± 0.5 | 97.6 ± 1.9 |
| 400 | 100.0 ± 0.0 | 99.3 ± 2.3 | 100.0 ± 0.1 | 98.5 ± 2.6 |