r8_c1
done · peak unseen_mismatch 95.8
α schedule
α 3.4651 ×25 α 3.464 ×25 α 3.4579 ×25 α 0.9384 ×25 α 0.9288 ×25 α 3.3433 ×25 α 3.2438 ×25 α 3.0962 ×25 α 0.7877 ×25 α 0.7109 ×25 α 0.6104 ×25 α 1.7673 ×25 α 1.1785 ×25 α 0.1241 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.1241
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 ± 1.4 | 50.3 ± 1.6 | 50.0 ± 1.5 | 49.4 ± 1.7 |
| 50 | 51.3 ± 1.4 | 50.3 ± 1.5 | 49.9 ± 1.2 | 50.0 ± 1.8 |
| 75 | 52.5 ± 2.0 | 50.0 ± 1.6 | 50.7 ± 1.5 | 49.6 ± 1.6 |
| 100 | 53.8 ± 5.2 | 50.9 ± 1.6 | 51.8 ± 3.4 | 49.9 ± 1.7 |
| 125 | 60.6 ± 13.3 | 50.5 ± 2.7 | 57.9 ± 13.2 | 49.4 ± 3.2 |
| 150 | 72.7 ± 19.2 | 61.3 ± 12.3 | 70.2 ± 19.4 | 58.8 ± 11.0 |
| 175 | 83.6 ± 16.3 | 74.8 ± 15.4 | 80.6 ± 17.1 | 71.5 ± 15.2 |
| 200 | 89.9 ± 10.9 | 86.2 ± 10.3 | 86.3 ± 12.3 | 80.7 ± 11.9 |
| 225 | 98.7 ± 3.1 | 89.2 ± 17.1 | 98.5 ± 3.3 | 88.3 ± 16.6 |
| 250 | 100.0 ± 0.0 | 83.3 ± 21.4 | 99.9 ± 0.3 | 82.7 ± 21.3 |
| 275 | 100.0 ± 0.0 | 83.5 ± 19.8 | 100.0 ± 0.0 | 82.9 ± 19.9 |
| 300 | 99.9 ± 0.5 | 93.9 ± 9.4 | 99.5 ± 1.9 | 92.1 ± 9.5 |
| 325 | 99.9 ± 0.3 | 96.8 ± 9.4 | 99.8 ± 0.5 | 95.8 ± 9.0 |
| 350 | 100.0 ± 0.0 | 90.4 ± 17.6 | 100.0 ± 0.0 | 89.3 ± 18.0 |