Table 2.

Comparison with the baseline. The ATD and accuracy were computed for all noise and sensors cases, both with a prior and without a prior, and then averaged. Our framework is denoted by “Bayes”, followed by the neural networks utilized. The vertical arrows denote the desired direction of the metric. The best results are highlighted in boldface.

	WDN1		WDN2		WDN3
	ATD ↓	Acc ↑	ATD ↓	Acc ↑	ATD ↓	Acc ↑
Classifier-ResNet	2.94	26.5	5.81	13.25	10.01	7.85
Classifier-Transformer	2.06	31.67	3.60	21.5	4.88	27.92
Bayes-WAE, ResNet	1.23	45.70	2.91	25.03	3.49	29.50
Bayes-WAE, Transformer	1.07	50.00	1.71	39.28	10.38	4.54