Collaborative Intelligence: Accelerating Deep Neural Network Inference via Device-Edge Synergy

Table 2

Comparisons of pruned parameter ratio and latency speedup of different strategies.

Prune process	Accuracy (%)	Pruned (%)	Parameter (M)	Pruned (%)	Mult-Adds (M)	Time-32
VGGNET (baseline)	94.64	—	20.3	—	398.14	69.0641 ms
VGGNET (status)	93.34	67.5	6.51	37.2	250.03	24.8501 ms ()
VGGNET (cogent)	93.82	88.5	2.31	50.8	195.87	7.7703 ms ()