ascend3
done · peak unseen_mismatch 81.5
α schedule
α 1.7754 ×25 α 1.7881 ×25 α 1.8076 ×25 α 1.8326 ×25 α 1.8623 ×25 α 1.8963 ×25 α 1.9343 ×25 α 1.976 ×25 α 2.0212 ×25 α 2.0698 ×25 α 2.1215 ×25 α 2.1763 ×25 α 2.2341 ×25 α 2.2947 ×25 α 2.2947 ×50
Figures
Results
accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 2.2947
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 |
|---|---|---|---|---|
| 100 | 51.8 ± 3.6 | 49.7 ± 1.6 | 50.7 ± 3.4 | 49.6 ± 1.6 |
| 110 | 52.4 ± 4.7 | 49.6 ± 1.7 | 51.4 ± 4.5 | 49.4 ± 1.9 |
| 120 | 53.5 ± 6.9 | 49.6 ± 2.2 | 52.3 ± 6.9 | 49.5 ± 2.2 |
| 130 | 55.1 ± 9.2 | 50.0 ± 3.4 | 53.9 ± 9.1 | 49.5 ± 3.0 |
| 140 | 58.3 ± 12.6 | 50.4 ± 4.4 | 56.7 ± 12.6 | 49.5 ± 3.9 |
| 150 | 62.0 ± 15.4 | 51.8 ± 6.6 | 60.3 ± 15.6 | 50.7 ± 5.9 |
| 160 | 66.3 ± 18.0 | 54.0 ± 9.3 | 64.4 ± 18.2 | 52.8 ± 8.7 |
| 170 | 71.1 ± 19.4 | 57.0 ± 12.2 | 69.3 ± 19.8 | 55.5 ± 11.4 |
| 180 | 75.1 ± 20.1 | 60.1 ± 14.3 | 73.4 ± 20.8 | 58.2 ± 13.4 |
| 190 | 79.0 ± 19.8 | 62.4 ± 15.5 | 77.0 ± 20.7 | 60.0 ± 14.5 |
| 200 | 82.9 ± 18.7 | 67.0 ± 17.1 | 81.0 ± 19.8 | 64.5 ± 16.3 |
| 210 | 86.0 ± 17.1 | 70.1 ± 16.7 | 84.0 ± 18.4 | 67.6 ± 16.1 |
| 220 | 89.1 ± 15.5 | 72.6 ± 17.7 | 87.1 ± 17.1 | 69.8 ± 17.2 |
| 230 | 91.1 ± 14.1 | 73.8 ± 18.0 | 89.2 ± 15.6 | 71.1 ± 17.2 |
| 240 | 92.7 ± 13.0 | 75.9 ± 18.6 | 90.9 ± 14.4 | 73.0 ± 17.6 |
| 250 | 93.2 ± 12.3 | 76.9 ± 19.5 | 91.1 ± 14.2 | 73.7 ± 18.7 |
| 260 | 93.4 ± 12.0 | 77.1 ± 19.0 | 90.5 ± 14.4 | 73.3 ± 17.7 |
| 270 | 93.5 ± 11.4 | 77.6 ± 20.2 | 90.6 ± 13.8 | 74.0 ± 18.7 |
| 280 | 93.9 ± 10.5 | 78.1 ± 17.9 | 90.9 ± 12.8 | 74.1 ± 16.6 |
| 290 | 94.7 ± 8.9 | 82.5 ± 15.1 | 91.7 ± 11.5 | 78.5 ± 13.5 |
| 300 | 95.5 ± 6.2 | 83.6 ± 13.7 | 91.6 ± 9.4 | 79.2 ± 12.0 |
| 310 | 95.8 ± 5.3 | 83.1 ± 16.4 | 91.8 ± 8.2 | 78.3 ± 14.8 |
| 320 | 96.1 ± 3.9 | 83.2 ± 19.8 | 91.6 ± 7.2 | 77.6 ± 17.5 |
| 330 | 95.8 ± 3.8 | 85.8 ± 18.7 | 91.5 ± 6.6 | 80.0 ± 17.1 |
| 340 | 96.0 ± 3.1 | 87.1 ± 19.6 | 90.5 ± 6.9 | 80.4 ± 18.2 |
| 350 | 96.7 ± 2.5 | 88.0 ± 18.7 | 91.4 ± 6.2 | 81.5 ± 17.3 |
| 360 | 96.2 ± 2.3 | 86.3 ± 21.2 | 89.5 ± 5.7 | 78.4 ± 18.1 |
| 370 | 96.2 ± 2.1 | 86.4 ± 19.0 | 88.7 ± 6.5 | 77.5 ± 16.0 |
| 380 | 96.0 ± 2.2 | 85.5 ± 18.6 | 89.2 ± 6.5 | 77.0 ± 15.4 |
| 390 | 96.4 ± 2.2 | 86.4 ± 18.8 | 89.3 ± 6.0 | 77.5 ± 15.7 |
| 400 | 95.9 ± 2.0 | 87.2 ± 15.3 | 88.5 ± 6.7 | 77.9 ± 12.8 |