r4_c0
done · peak unseen_mismatch 96.2
α schedule
α 4.9959 ×25 α 4.9951 ×25 α 4.9909 ×25 α 4.9787 ×25 α 4.9525 ×25 α 4.905 ×25 α 4.8277 ×25 α 4.7111 ×25 α 4.5444 ×25 α 4.316 ×25 α 4.0132 ×25 α 3.6223 ×25 α 3.1289 ×25 α 2.5175 ×25 α 1.7718 ×25 α 0.8747 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.8747
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 | 49.8 ± 1.4 | 49.7 ± 2.0 |
| 50 | 51.6 ± 2.0 | 50.5 ± 1.5 | 50.0 ± 1.4 | 50.0 ± 1.5 |
| 75 | 53.4 ± 2.6 | 50.7 ± 1.7 | 50.4 ± 1.4 | 49.6 ± 1.5 |
| 100 | 60.8 ± 8.9 | 58.5 ± 7.0 | 53.8 ± 6.0 | 51.7 ± 4.2 |
| 125 | 72.9 ± 8.6 | 69.6 ± 8.1 | 59.2 ± 8.1 | 56.8 ± 6.9 |
| 150 | 78.3 ± 3.8 | 76.2 ± 3.7 | 62.8 ± 7.0 | 60.6 ± 6.3 |
| 175 | 79.8 ± 3.6 | 77.7 ± 3.4 | 62.7 ± 6.1 | 60.5 ± 5.6 |
| 200 | 80.0 ± 2.8 | 78.2 ± 2.3 | 62.0 ± 4.4 | 60.5 ± 4.6 |
| 225 | 80.6 ± 3.2 | 78.9 ± 2.8 | 62.5 ± 4.7 | 61.1 ± 4.7 |
| 250 | 81.3 ± 2.2 | 79.8 ± 2.0 | 62.5 ± 4.2 | 61.5 ± 4.3 |
| 275 | 82.9 ± 2.6 | 81.8 ± 2.4 | 64.8 ± 4.4 | 63.0 ± 4.4 |
| 300 | 85.0 ± 2.5 | 83.7 ± 2.3 | 67.1 ± 4.9 | 65.5 ± 4.9 |
| 325 | 87.9 ± 2.0 | 86.2 ± 1.6 | 69.7 ± 4.8 | 68.0 ± 4.7 |
| 350 | 91.3 ± 1.2 | 89.7 ± 1.5 | 74.1 ± 4.5 | 72.5 ± 4.3 |
| 375 | 96.5 ± 1.1 | 95.3 ± 1.6 | 85.7 ± 4.6 | 82.6 ± 4.1 |
| 400 | 99.9 ± 0.1 | 99.0 ± 3.4 | 99.2 ± 0.7 | 96.2 ± 8.0 |