p15
done · peak unseen_mismatch 97.3
α schedule
α 3.6309 ×25 α 3.5934 ×25 α 3.5276 ×25 α 3.436 ×25 α 3.3205 ×25 α 3.1821 ×25 α 3.0218 ×25 α 2.8401 ×25 α 2.6377 ×25 α 2.4151 ×25 α 2.1727 ×25 α 1.9109 ×25 α 1.63 ×25 α 1.3304 ×25 α 1.0122 ×25 α 0.6758 ×25
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.6758
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 | 50.4 ± 1.6 | 49.9 ± 1.6 | 49.5 ± 1.9 |
| 50 | 51.4 ± 1.5 | 50.3 ± 1.6 | 50.1 ± 1.2 | 49.8 ± 1.8 |
| 75 | 52.2 ± 2.2 | 50.1 ± 1.6 | 50.5 ± 1.5 | 49.4 ± 1.4 |
| 100 | 57.2 ± 7.0 | 55.2 ± 4.8 | 52.3 ± 3.6 | 50.4 ± 2.0 |
| 125 | 68.3 ± 11.7 | 64.8 ± 10.6 | 59.3 ± 10.3 | 56.7 ± 8.4 |
| 150 | 79.0 ± 9.5 | 76.0 ± 9.3 | 66.9 ± 10.2 | 63.5 ± 9.9 |
| 175 | 85.6 ± 5.3 | 82.7 ± 5.3 | 71.8 ± 9.9 | 68.4 ± 9.4 |
| 200 | 88.0 ± 3.5 | 85.7 ± 3.4 | 73.4 ± 8.2 | 71.0 ± 7.8 |
| 225 | 90.0 ± 2.9 | 87.8 ± 2.6 | 75.2 ± 6.7 | 72.6 ± 7.1 |
| 250 | 91.7 ± 2.2 | 89.6 ± 2.0 | 76.7 ± 6.0 | 74.1 ± 6.2 |
| 275 | 94.2 ± 1.8 | 92.3 ± 1.8 | 80.9 ± 6.3 | 78.2 ± 5.6 |
| 300 | 95.8 ± 1.4 | 94.3 ± 1.6 | 85.2 ± 4.8 | 81.6 ± 5.1 |
| 325 | 97.4 ± 1.1 | 96.3 ± 1.2 | 89.3 ± 4.2 | 85.2 ± 5.2 |
| 350 | 99.3 ± 0.4 | 98.8 ± 0.4 | 95.6 ± 1.9 | 92.5 ± 2.7 |
| 375 | 99.9 ± 0.1 | 99.5 ± 0.5 | 98.5 ± 1.0 | 96.1 ± 2.2 |
| 400 | 100.0 ± 0.0 | 99.7 ± 0.6 | 99.6 ± 0.7 | 97.3 ± 4.3 |