LayerMerge: Neural Network Depth Compression through Layer Pruning and Merging
This work addresses efficiency issues in neural network inference for applications like image processing, though it is incremental as it builds on prior depth compression techniques.
The paper tackles the problem of depth compression in convolutional neural networks, where existing methods increase kernel size and undermine latency reduction; the proposed LayerMerge method jointly prunes convolution and activation layers, achieving consistent performance improvements over existing methods on image classification and generation tasks.
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.