r4_c4
done · peak unseen_mismatch 64.1
α schedule
α 4.9823 ×25 α 4.974 ×25 α 4.9643 ×25 α 4.9537 ×25 α 4.9423 ×25 α 4.9304 ×25 α 4.918 ×25 α 4.9052 ×25 α 4.8919 ×25 α 4.8784 ×25 α 4.8645 ×25 α 4.8503 ×25 α 4.8358 ×25 α 4.821 ×25 α 4.8061 ×25 α 4.7909 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 4.7909
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 ± 1.3 | 50.5 ± 1.6 | 49.8 ± 1.3 | 49.7 ± 1.9 |
| 50 | 51.6 ± 2.0 | 50.5 ± 1.5 | 50.0 ± 1.4 | 50.0 ± 1.5 |
| 75 | 53.4 ± 2.5 | 50.6 ± 1.7 | 50.4 ± 1.4 | 49.5 ± 1.4 |
| 100 | 61.1 ± 9.0 | 58.7 ± 6.9 | 53.9 ± 6.2 | 51.8 ± 4.3 |
| 125 | 73.0 ± 8.2 | 69.8 ± 7.8 | 59.3 ± 8.2 | 56.9 ± 7.0 |
| 150 | 78.6 ± 3.4 | 76.6 ± 2.9 | 63.1 ± 7.0 | 60.8 ± 6.3 |
| 175 | 79.6 ± 3.5 | 77.6 ± 3.1 | 62.6 ± 5.8 | 60.5 ± 5.6 |
| 200 | 79.4 ± 2.9 | 77.9 ± 2.6 | 61.6 ± 4.3 | 60.3 ± 4.5 |
| 225 | 80.2 ± 3.0 | 78.5 ± 2.5 | 62.1 ± 4.6 | 60.9 ± 4.6 |
| 250 | 80.8 ± 2.5 | 79.1 ± 2.1 | 61.8 ± 4.0 | 60.8 ± 4.0 |
| 275 | 81.3 ± 2.5 | 80.0 ± 1.9 | 62.8 ± 3.8 | 61.4 ± 4.1 |
| 300 | 81.9 ± 2.7 | 80.5 ± 2.6 | 63.4 ± 4.2 | 62.2 ± 4.3 |
| 325 | 83.3 ± 2.8 | 81.5 ± 2.2 | 64.4 ± 4.6 | 62.7 ± 4.2 |
| 350 | 83.2 ± 2.1 | 81.5 ± 1.7 | 64.3 ± 4.1 | 62.9 ± 3.7 |
| 375 | 84.1 ± 2.2 | 82.2 ± 2.6 | 65.4 ± 4.9 | 63.8 ± 4.7 |
| 400 | 84.9 ± 2.5 | 83.0 ± 2.0 | 65.9 ± 5.0 | 64.1 ± 4.7 |