RGP: Neural Network Pruning through Its Regular Graph Structure

Pruning
DOI: 10.48550/arxiv.2110.15192 Publication Date: 2021-01-01
ABSTRACT
Lightweight model design has become an important direction in the application of deep learning technology, pruning is effective mean to achieve a large reduction parameters and FLOPs. The existing neural network methods mostly start from importance parameters, parameter evaluation metrics perform iteratively. These are not studied perspective topology, may be but efficient, requires completely different for datasets. In this paper, we study graph structure network, propose regular based (RGP) one-shot pruning. We generate graph, set node degree value meet ratio, reduce average shortest path length by swapping edges obtain optimal edge distribution. Finally, obtained mapped into realize Experiments show that negatively correlated with classification accuracy corresponding proposed RGP shows strong precision retention capability extremely high (more than 90%) FLOPs 90%).
SUPPLEMENTAL MATERIAL
Coming soon ....
REFERENCES ()
CITATIONS ()
EXTERNAL LINKS
PlumX Metrics
RECOMMENDATIONS
FAIR ASSESSMENT
Coming soon ....
JUPYTER LAB
Coming soon ....