p23
done · peak unseen_mismatch 65.4
α schedule
α 2.3173 ×25 α 2.0592 ×25 α 1.8863 ×25 α 1.7478 ×25 α 1.6294 ×25 α 1.5244 ×25 α 1.4292 ×25 α 1.3415 ×25 α 1.26 ×25 α 1.1834 ×25 α 1.1111 ×25 α 1.0423 ×25 α 0.9767 ×25 α 0.9139 ×25 α 0.8534 ×25 α 0.7952 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.7952
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 | 50.7 ± 1.4 | 50.2 ± 1.6 | 49.9 ± 1.5 | 49.5 ± 1.7 |
| 50 | 50.9 ± 1.5 | 50.2 ± 1.7 | 49.9 ± 1.5 | 49.9 ± 1.8 |
| 75 | 51.1 ± 1.8 | 49.7 ± 1.6 | 50.4 ± 1.5 | 49.5 ± 1.4 |
| 100 | 51.3 ± 2.7 | 50.2 ± 1.4 | 50.7 ± 1.8 | 49.8 ± 1.4 |
| 125 | 56.2 ± 7.6 | 49.4 ± 2.4 | 54.2 ± 7.4 | 49.0 ± 2.4 |
| 150 | 64.1 ± 17.3 | 48.8 ± 6.7 | 62.3 ± 17.8 | 48.1 ± 6.2 |
| 175 | 74.7 ± 21.7 | 52.4 ± 15.5 | 73.7 ± 22.2 | 51.3 ± 14.9 |
| 200 | 80.6 ± 21.8 | 56.0 ± 17.8 | 80.1 ± 22.1 | 55.4 ± 17.3 |
| 225 | 88.1 ± 18.1 | 62.1 ± 21.1 | 88.0 ± 17.8 | 61.3 ± 20.4 |
| 250 | 92.5 ± 17.1 | 62.5 ± 24.7 | 92.1 ± 17.7 | 61.9 ± 24.8 |
| 275 | 93.7 ± 15.5 | 64.0 ± 26.2 | 93.6 ± 15.6 | 62.9 ± 26.0 |
| 300 | 97.2 ± 9.4 | 61.7 ± 29.8 | 97.1 ± 9.5 | 61.1 ± 29.4 |
| 325 | 100.0 ± 0.0 | 65.5 ± 27.9 | 100.0 ± 0.0 | 65.4 ± 27.6 |
| 350 | 100.0 ± 0.0 | 63.8 ± 29.2 | 100.0 ± 0.1 | 63.1 ± 29.2 |
| 375 | 100.0 ± 0.0 | 63.7 ± 29.7 | 100.0 ± 0.0 | 64.0 ± 29.8 |
| 400 | 100.0 ± 0.0 | 62.4 ± 30.6 | 100.0 ± 0.0 | 62.5 ± 30.5 |