notesum.ai

Published at November 19

mDAE : modified Denoising AutoEncoder for missing data imputation

cs.AI

Released Date: November 19, 2024

Authors: Mariette Dupuy¹, Marie Chavent², Remi Dubois

Aff.: ¹Liryc Institute University of Bordeaux; ²Inria Bordeaux

Arxiv: http://arxiv.org/abs/2411.12847v1

Refer to caption

Method	breast	climate	sonar	iono	seeds	wine	blood
mDAE	0.466 $\pm$ 0.016	1.007 $\pm$ 0.007	0.656 $\pm$ 0.007	0.776 $\pm$ 0.018	0.496 $\pm$ 0.022	0.790 $\pm$ 0.030	0.701 $\pm$ 0.059

mDAE w/o	0.685 $\pm$ 0.036	1.005 $\pm$ 0.008	0.988 $\pm$ 0.013	0.808 $\pm$ 0.020	0.587 $\pm$ 0.028	0.828 $\pm$ 0.034	0.755 $\pm$ 0.058

modified loss	(46.996%)	(-0.199%)	(50.610%)	(4.124%)	(18.347%)	(4.810%)	(7.703%)
mDAE w/o	0.501 $\pm$ 0.043	1.030 $\pm$ 0.013	0.682 $\pm$ 0.049	0.802 $\pm$ 0.039	0.514 $\pm$ 0.054	0.853 $\pm$ 0.033	0.710 $\pm$ 0.055

optimal $\mu$	(7.511%)	(2.284%)	(3.963%)	(3.351%)	(3.629%)	(7.975%)	(1.284%)
mDAE w/o	0.500 $\pm$ 0.011	1.147 $\pm$ 0.013	0.699 $\pm$ 0.008	0.808 $\pm$ 0.025	0.671 $\pm$ 0.209	0.932 $\pm$ 0.045	0.960 $\pm$ 0.140

overcomplete	(7.296%)	(13.903%)	(6.555%)	(4.124%)	(35.282%)	(17.975%)	(36.947%)