r5_c5
done · peak unseen_mismatch 88.4
α schedule
α 1.583 ×25 α 1.6855 ×25 α 1.788 ×25 α 1.8905 ×25 α 1.9929 ×25 α 2.0954 ×25 α 2.1979 ×25 α 2.3003 ×25 α 2.4028 ×25 α 2.5052 ×25 α 2.6077 ×25 α 2.7101 ×25 α 2.8125 ×25 α 2.915 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.915
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 | 50.5 ± 1.6 | 50.0 ± 1.4 | 49.9 ± 1.8 | 49.5 ± 2.0 |
| 50 | 51.0 ± 1.5 | 50.1 ± 1.6 | 50.2 ± 1.5 | 49.7 ± 1.6 |
| 75 | 50.6 ± 1.4 | 49.7 ± 1.8 | 50.3 ± 1.4 | 49.7 ± 1.8 |
| 100 | 50.6 ± 2.2 | 50.3 ± 1.6 | 50.2 ± 1.4 | 49.8 ± 1.6 |
| 125 | 52.1 ± 2.8 | 49.7 ± 2.5 | 51.2 ± 3.4 | 49.4 ± 2.4 |
| 150 | 57.5 ± 12.7 | 51.2 ± 6.6 | 56.1 ± 12.8 | 49.8 ± 6.5 |
| 175 | 65.9 ± 16.7 | 55.1 ± 9.6 | 64.3 ± 16.8 | 53.5 ± 9.5 |
| 200 | 78.8 ± 19.9 | 64.0 ± 16.5 | 77.0 ± 20.2 | 62.1 ± 16.1 |
| 225 | 84.1 ± 20.2 | 73.0 ± 17.9 | 82.4 ± 20.6 | 71.5 ± 16.9 |
| 250 | 88.5 ± 16.0 | 82.5 ± 16.0 | 85.7 ± 17.3 | 79.7 ± 15.4 |
| 275 | 93.3 ± 11.6 | 89.5 ± 12.7 | 90.4 ± 12.8 | 85.8 ± 12.9 |
| 300 | 95.2 ± 9.0 | 92.4 ± 8.5 | 91.8 ± 10.3 | 88.0 ± 9.2 |
| 325 | 95.8 ± 6.2 | 94.1 ± 6.1 | 91.6 ± 6.9 | 88.4 ± 7.2 |
| 350 | 95.4 ± 4.7 | 93.4 ± 4.7 | 89.9 ± 6.3 | 86.4 ± 6.2 |