Table 4.

Neural network model specific details comparison. [FLOPs] denotes the floating-point operations of the model compression and reconstruction. [CP-time] denotes the compression time. [FP-time] denotes the compression and reconstruction time. [TR-time] denotes the training iteration time. These results are obtained when the number of mini-batches is 1, and the time results are the corresponding unit time.

Model	FLOPs	No. of Parameters	CP-Time (ms)	FP-Time (ms)	TR-Time (ms)
Stacked RBM-AE	35,850	19,032	0.2113	0.3873	1.0211
Stacked RBM-VAE	36,150	19,234	0.2817	0.4577	1.6549
CNN-AE	772,008	9245	0.9859	1.5493	2.6761
CBN-VAE	27,564	2415	0.1761	1.162	1.7606