r2_c1
done · peak unseen_mismatch 91.2
α schedule
α 2.0413 ×25 α 2.0413 ×25 α 2.0414 ×25 α 2.0416 ×25 α 2.0419 ×25 α 2.0426 ×25 α 2.0437 ×25 α 2.0455 ×25 α 2.048 ×25 α 2.0516 ×25 α 2.0564 ×25 α 2.0628 ×25 α 2.0709 ×25 α 2.0811 ×25 α 2.0936 ×25 α 2.1089 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.1089
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 ± 1.6 | 50.2 ± 1.4 | 50.0 ± 1.4 | 49.6 ± 1.7 |
| 50 | 50.9 ± 1.6 | 50.5 ± 1.6 | 50.0 ± 1.5 | 50.0 ± 1.6 |
| 75 | 50.8 ± 1.8 | 49.5 ± 1.6 | 50.5 ± 1.4 | 49.5 ± 1.6 |
| 100 | 51.2 ± 2.7 | 50.5 ± 1.7 | 50.2 ± 1.5 | 50.0 ± 1.6 |
| 125 | 55.0 ± 5.0 | 49.7 ± 2.9 | 52.3 ± 4.6 | 48.9 ± 2.1 |
| 150 | 62.3 ± 16.1 | 53.6 ± 10.3 | 59.5 ± 15.9 | 51.4 ± 9.7 |
| 175 | 73.5 ± 19.7 | 62.1 ± 13.7 | 71.1 ± 19.6 | 60.3 ± 13.0 |
| 200 | 80.5 ± 20.1 | 71.8 ± 17.0 | 78.2 ± 20.8 | 68.9 ± 17.3 |
| 225 | 88.9 ± 14.5 | 80.4 ± 14.4 | 86.1 ± 16.8 | 76.9 ± 15.1 |
| 250 | 94.6 ± 9.5 | 90.9 ± 9.8 | 91.8 ± 12.5 | 87.2 ± 11.9 |
| 275 | 97.4 ± 5.1 | 93.3 ± 10.5 | 94.6 ± 7.3 | 89.2 ± 10.9 |
| 300 | 98.0 ± 2.3 | 95.8 ± 3.7 | 95.3 ± 4.5 | 91.2 ± 5.4 |
| 325 | 98.4 ± 1.1 | 96.4 ± 1.5 | 95.1 ± 3.9 | 91.2 ± 5.0 |
| 350 | 97.9 ± 1.3 | 96.2 ± 1.6 | 93.8 ± 4.0 | 90.3 ± 4.9 |
| 375 | 97.6 ± 1.1 | 96.1 ± 1.3 | 92.2 ± 3.9 | 88.5 ± 4.8 |
| 400 | 97.6 ± 1.0 | 95.9 ± 1.1 | 91.8 ± 3.6 | 88.0 ± 4.4 |