imagining syntax

← schedopt_v5_confirm

free-9a4f8259

done · peak unseen_mismatch 95.5

Referenced by

α schedule

α 5.0 ×10 α 4.8 ×10 α 4.9 ×10 α 5.0 ×10 α 4.1 ×20 α 4.2 ×10 α 4.4 ×10 α 4.6 ×10 α 4.8 ×10 α 4.4 ×10 α 4.7 ×10 α 4.9 ×10 α 4.1 ×10 α 4.8 ×10 α 5.0 ×20 α 2.4 ×10 α 0.0 ×20 α 0.4 ×10 α 0.0 ×10 α 3.0 ×10 α 4.8 ×10 α 4.6 ×10 α 3.5 ×10 α 2.3 ×10 α 1.0 ×10 α 0.4 ×10 α 0.5 ×110

direction
mixed
α range
0.0–5.0
stages
27

Figures

schedule.png
schedule.png

Results

accuracy by training iteration, mean ± sd over seeds · schedule ends at α = 0.5

seen_* probed at reference α = 1.4 · unseen_* = held-out pairings, uniform (α-independent) · 1000 pairs per condition

iterationseen_matchseen_mismatchunseen_matchunseen_mismatch
100 61.4 ± 7.957.8 ± 6.754.1 ± 5.951.8 ± 3.2
110 66.6 ± 9.262.9 ± 8.156.7 ± 7.554.1 ± 5.2
120 71.4 ± 9.267.8 ± 8.659.6 ± 8.456.6 ± 6.3
130 75.0 ± 7.871.7 ± 7.661.4 ± 8.158.4 ± 6.5
140 77.7 ± 6.574.8 ± 6.063.1 ± 7.660.1 ± 6.0
150 79.3 ± 5.076.8 ± 4.664.0 ± 7.061.1 ± 5.9
160 79.9 ± 4.277.5 ± 3.763.4 ± 6.560.9 ± 5.7
170 80.3 ± 3.778.1 ± 3.263.5 ± 6.061.2 ± 5.1
180 82.2 ± 3.580.2 ± 3.265.2 ± 6.262.9 ± 5.4
190 86.2 ± 5.584.1 ± 5.273.6 ± 9.270.6 ± 8.5
200 92.8 ± 6.790.7 ± 6.987.7 ± 9.684.4 ± 9.8
210 97.6 ± 4.095.6 ± 5.695.9 ± 6.092.9 ± 7.7
220 98.9 ± 1.593.8 ± 9.698.2 ± 2.591.4 ± 10.8
230 99.4 ± 0.890.0 ± 14.598.6 ± 1.787.2 ± 15.2
240 98.8 ± 0.595.2 ± 1.996.8 ± 2.089.7 ± 5.0
250 91.2 ± 2.588.3 ± 1.578.2 ± 7.774.5 ± 5.5
260 88.2 ± 0.786.6 ± 0.975.7 ± 3.273.5 ± 3.0
270 94.2 ± 0.792.1 ± 0.486.2 ± 0.382.1 ± 0.8
280 99.4 ± 0.398.2 ± 0.998.8 ± 1.195.5 ± 2.6

Downloads