FPCC-GAN:CLUSTER CENTER AND GENERATIVE ADVERSARIAL LEARNING IN FILTER LEVEL PRUNING
The deep architecture and parameter redundancy of deep neural network will lead to high computational cost.Deep neural network compression and acceleration has become an important issue in recent years.To address the norm-criterion limitation and label dependence of current methods,we propose a structured filter pruning method based on cluster center and generative adversarial learning(FPCC-GAN).(1)The filters were clustered by K-means clustering algorithm for every convolution layer.(2)Filters closer to the cluster center were pruned proportionally,which extracted redundant features.(3)Generative adversarial learning was used for iteratively training.The experimental results show that compared with current mainstream methods,the proposed method has higher accuracy.
国家科技期刊平台
NSTL
万方数据
维普
