Table 3.

Performance comparison of Proposed Model with compressed version of competitive models.

Algorithm	Pruning	FLOPs (M)	Memory (MB)	Accuracy (%)	# of Channels Pruned/Total	Inference Time ($\mathsf{\mu }$s/Sample) (Container-NVIDIA Docker)
						Xavier	TX2	Nano
DeepConv GRU - Attention	0%	1.627	7.91	98.36	-	∼505	∼1175	∼2580
	8.50%	1.482	7.48	98.04	LSTM1(07/128) LSTM2(11/128)	∼469	∼1040	∼2270
	19%	1.314	6.64	96.89	LSTM1 (20/128) LSTM2(15/128)	∼452	∼997	∼2160
FCN-LSTM	0%	0.566	3.28	95.10	-	∼284	∼365	∼450
	5.30%	0.535	3.14	94.32	Conv1(10/128) Conv2(10/256)	∼253	∼342	∼416
	12.20%	0.496	3.07	93.92	Conv1(10/128) Conv2(20/256) Conv3(10/256)	∼241	∼333	∼371
Proposed DC-LSTM	0%	0.233	1.69	97.86	-	∼188	∼207	∼230
Proposed DC-GRU	0%	0.232	1.69	98.72	-	∼182	∼205	∼227