r8_c5
done · peak unseen_mismatch 99.7
α schedule
α 5.0 ×25 α 5.0 ×25 α 5.0 ×25 α 0.0718 ×25 α 0.0723 ×25 α 5.0 ×25 α 5.0 ×25 α 5.0 ×25 α 0.0776 ×25 α 0.0801 ×25 α 0.0831 ×25 α 5.0 ×25 α 5.0 ×25 α 0.0964 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.0964
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 | 50.9 ± 1.3 | 50.5 ± 1.6 | 49.9 ± 1.4 | 49.7 ± 2.0 |
| 50 | 51.5 ± 2.0 | 50.5 ± 1.5 | 50.0 ± 1.3 | 50.0 ± 1.6 |
| 75 | 53.4 ± 2.6 | 50.7 ± 1.7 | 50.4 ± 1.3 | 49.6 ± 1.5 |
| 100 | 54.6 ± 7.8 | 51.2 ± 2.2 | 52.9 ± 6.8 | 50.1 ± 2.3 |
| 125 | 61.0 ± 14.8 | 48.2 ± 7.1 | 58.9 ± 15.5 | 47.1 ± 7.8 |
| 150 | 74.7 ± 19.4 | 57.1 ± 14.5 | 73.5 ± 19.9 | 54.7 ± 13.8 |
| 175 | 87.2 ± 12.5 | 77.0 ± 12.5 | 84.5 ± 13.8 | 73.5 ± 12.7 |
| 200 | 92.7 ± 5.5 | 89.3 ± 6.4 | 88.6 ± 8.4 | 84.3 ± 9.1 |
| 225 | 99.8 ± 0.6 | 88.2 ± 17.1 | 99.6 ± 0.7 | 87.5 ± 17.0 |
| 250 | 100.0 ± 0.0 | 87.5 ± 16.1 | 100.0 ± 0.0 | 87.0 ± 16.5 |
| 275 | 100.0 ± 0.0 | 80.6 ± 18.9 | 100.0 ± 0.0 | 80.9 ± 18.4 |
| 300 | 99.9 ± 0.5 | 95.5 ± 8.4 | 99.7 ± 1.4 | 95.3 ± 8.2 |
| 325 | 98.5 ± 2.2 | 96.7 ± 3.0 | 97.1 ± 4.1 | 93.9 ± 6.0 |
| 350 | 100.0 ± 0.0 | 99.8 ± 0.6 | 100.0 ± 0.0 | 99.7 ± 0.6 |