A Cooperative Lightweight Translation Algorithm Combined with Sparse-ReLU

Table 9

Experimental results of the combination of Sparse-ReLU and pruning algorithm.

Pruning algorithm	Set parameters	Relative activations sparsity	Relative loss value	Weights sparsity
No pruning + Sparse-ReLU	\	100% (Baseline)	100% (Baseline)	0.00% (Baseline)
L1Unstructured + Sparse-ReLU	0.3	121.22%	173.60%	22.58%
	0.5	122.97%	195.12%	68.47%
RandomUnstructured + Sparse-ReLU	0.3	120.18%	120.42%	65.99%
	0.2	120.55%	110.79%	48.70%
	0.4	120.70%	288.05%	78.26%