p14
done · peak unseen_mismatch 74.8
α schedule
α 1.5169 ×25 α 2.0614 ×25 α 2.4156 ×25 α 2.6947 ×25 α 2.9309 ×25 α 3.1385 ×25 α 3.3254 ×25 α 3.4964 ×25 α 3.6548 ×25 α 3.8028 ×25 α 3.9421 ×25 α 4.074 ×25 α 4.1994 ×25 α 4.3192 ×25 α 4.434 ×25 α 4.5444 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.5444
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 | 50.5 ± 1.4 | 50.1 ± 1.6 | 49.8 ± 1.5 | 49.3 ± 2.0 |
| 50 | 50.8 ± 1.7 | 50.1 ± 1.7 | 49.7 ± 1.4 | 50.0 ± 1.6 |
| 75 | 50.5 ± 1.4 | 49.5 ± 1.6 | 50.2 ± 1.4 | 49.5 ± 1.7 |
| 100 | 50.7 ± 2.2 | 50.3 ± 1.6 | 50.0 ± 1.3 | 50.2 ± 1.4 |
| 125 | 53.4 ± 2.9 | 50.3 ± 2.2 | 51.6 ± 3.1 | 49.5 ± 2.2 |
| 150 | 60.4 ± 11.3 | 55.1 ± 7.3 | 56.8 ± 10.7 | 51.9 ± 7.6 |
| 175 | 72.3 ± 15.4 | 67.3 ± 12.9 | 66.6 ± 15.2 | 61.8 ± 12.1 |
| 200 | 78.1 ± 15.4 | 75.4 ± 14.0 | 72.0 ± 15.6 | 68.0 ± 14.6 |
| 225 | 83.2 ± 12.2 | 80.4 ± 12.3 | 74.7 ± 14.5 | 71.1 ± 13.8 |
| 250 | 86.7 ± 7.8 | 84.2 ± 7.2 | 77.2 ± 10.7 | 74.1 ± 10.0 |
| 275 | 88.6 ± 5.4 | 86.7 ± 5.3 | 77.7 ± 9.8 | 74.8 ± 9.1 |
| 300 | 88.8 ± 4.6 | 87.0 ± 4.5 | 76.9 ± 9.4 | 74.5 ± 8.9 |
| 325 | 88.0 ± 4.5 | 86.3 ± 4.5 | 74.8 ± 8.8 | 72.4 ± 8.5 |
| 350 | 87.6 ± 3.6 | 85.9 ± 3.4 | 73.5 ± 6.9 | 71.1 ± 6.8 |
| 375 | 87.9 ± 3.1 | 85.6 ± 2.9 | 72.9 ± 6.4 | 70.3 ± 6.5 |
| 400 | 87.5 ± 2.4 | 85.5 ± 2.5 | 72.5 ± 5.1 | 70.0 ± 5.1 |