LayerMerge: Neural Network Depth Compression through Layer Pruning and Merging

TL;DR

LayerMerge introduces a joint pruning and merging approach for neural network depth compression, effectively reducing latency without performance loss by optimizing layer removal through a surrogate dynamic programming method.

Contribution

It proposes a novel joint pruning and merging technique with a surrogate optimization framework to improve depth compression efficiency in neural networks.

Findings

Outperforms existing depth compression methods on various architectures

Reduces inference latency while maintaining accuracy

Effective on both image classification and generation tasks

Abstract

Recent works show that reducing the number of layers in a convolutional neural network can enhance efficiency while maintaining the performance of the network. Existing depth compression methods remove redundant non-linear activation functions and merge the consecutive convolution layers into a single layer. However, these methods suffer from a critical drawback; the kernel size of the merged layers becomes larger, significantly undermining the latency reduction gained from reducing the depth of the network. We show that this problem can be addressed by jointly pruning convolution layers and activation functions. To this end, we propose LayerMerge, a novel depth compression method that selects which activation layers and convolution layers to remove, to achieve a desired inference speed-up while minimizing performance loss. Since the corresponding selection problem involves an exponential search space, we formulate a novel surrogate optimization problem and efficiently solve it via dynamic programming. Empirical results demonstrate that our method consistently outperforms existing depth compression and layer pruning methods on various network architectures, both on image classification and generation tasks. We release the code at https://github.com/snu-mllab/LayerMerge.