p04
done · peak unseen_mismatch 64.7
α schedule
α 4.7706 ×25 α 4.7737 ×25 α 4.7773 ×25 α 4.7814 ×25 α 4.7857 ×25 α 4.7903 ×25 α 4.7951 ×25 α 4.8001 ×25 α 4.8052 ×25 α 4.8105 ×25 α 4.8159 ×25 α 4.8214 ×25 α 4.8271 ×25 α 4.8329 ×25 α 4.8387 ×25 α 4.8447 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.8447
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 ± 1.6 | 49.8 ± 1.5 | 49.7 ± 1.9 |
| 50 | 51.6 ± 1.9 | 50.5 ± 1.4 | 50.0 ± 1.3 | 49.9 ± 1.5 |
| 75 | 53.0 ± 2.3 | 50.5 ± 1.6 | 50.4 ± 1.3 | 49.5 ± 1.5 |
| 100 | 60.6 ± 8.7 | 58.5 ± 6.7 | 53.6 ± 5.6 | 51.8 ± 3.6 |
| 125 | 72.7 ± 8.8 | 69.0 ± 8.2 | 59.3 ± 8.3 | 57.2 ± 7.4 |
| 150 | 78.9 ± 4.7 | 76.5 ± 4.3 | 63.3 ± 7.9 | 60.8 ± 7.2 |
| 175 | 80.0 ± 3.4 | 77.9 ± 3.2 | 63.0 ± 6.1 | 60.7 ± 5.6 |
| 200 | 80.1 ± 3.0 | 78.4 ± 2.6 | 62.5 ± 5.0 | 61.1 ± 5.1 |
| 225 | 80.7 ± 2.8 | 79.2 ± 2.5 | 62.7 ± 5.0 | 61.2 ± 4.8 |
| 250 | 80.9 ± 2.4 | 79.5 ± 2.0 | 62.7 ± 4.1 | 61.3 ± 4.2 |
| 275 | 81.5 ± 2.5 | 80.3 ± 2.0 | 63.2 ± 4.4 | 61.8 ± 4.4 |
| 300 | 82.1 ± 2.9 | 81.0 ± 2.6 | 64.2 ± 5.3 | 63.0 ± 4.9 |
| 325 | 83.2 ± 2.5 | 81.8 ± 1.8 | 65.0 ± 4.2 | 63.5 ± 3.9 |
| 350 | 83.8 ± 2.3 | 82.3 ± 2.0 | 65.5 ± 4.5 | 64.1 ± 4.3 |
| 375 | 83.8 ± 2.7 | 82.2 ± 2.4 | 65.4 ± 5.1 | 63.9 ± 5.5 |
| 400 | 84.7 ± 2.7 | 83.0 ± 2.8 | 66.4 ± 5.2 | 64.7 ± 5.4 |