Neural Network Panning: Screening the Optimal Sparse Network Before Training

Pruning on neural networks before training not only compresses the original models, but also accelerates network phase, which has substantial application value. The current work focuses fine-grained pruning, uses metrics to calculate weight scores for screening, and extends from initial single-order pruning iterative pruning. Through these works, we argue that can be summarized as an expressive force transfer process of weights, where reserved weights will take removed ones purpose maintaining performance networks. In order achieve optimal scheduling, propose a scheme called Neural Network Panning guides through multi-index multi-process steps, designs kind panning agent based reinforcement learning automate processes. Experimental results show performs better than various available methods. Our code is made public at: https://github.com/kangxiatao/RLPanning.