# Network Implosion: Effective Model Compression for ResNets via Static Layer Pruning and Retraining

## TL;DR

This paper introduces Network Implosion, a method for compressing ResNets by removing unimportant layers based on a priority measure, followed by retraining, achieving significant layer reduction without accuracy loss.

## Contribution

It proposes a novel layer pruning and retraining scheme for ResNets that preserves accuracy while substantially reducing model complexity.

## Key findings

- Reduces ResNet layers by up to 42.86% without accuracy loss
- Theoretical proof supports layer erasure without accuracy degradation
- Effective on CIFAR-10/100 and ImageNet datasets

## Abstract

Residual Networks with convolutional layers are widely used in the field of machine learning. Since they effectively extract features from input data by stacking multiple layers, they can achieve high accuracy in many applications. However, the stacking of many layers raises their computation costs. To address this problem, we propose Network Implosion, it erases multiple layers from Residual Networks without degrading accuracy. Our key idea is to introduce a priority term that identifies the importance of a layer; we can select unimportant layers according to the priority and erase them after the training. In addition, we retrain the networks to avoid critical drops in accuracy after layer erasure. A theoretical assessment reveals that our erasure and retraining scheme can erase layers without accuracy drop, and achieve higher accuracy than is possible with training from scratch. Our experiments show that Network Implosion can, for classification on Cifar-10/100 and ImageNet, reduce the number of layers by 24.00 to 42.86 percent without any drop in accuracy.