Learnable Heterogeneous Convolution: Learning both topology and strength

TL;DR

This paper introduces Learnable Heterogeneous Convolution, a biologically inspired method that jointly learns kernel topology and weights, significantly reducing computation while maintaining or improving accuracy.

Contribution

It unifies existing convolution techniques into a data-driven approach that learns both structure and strength, enabling efficient and accurate neural network models.

Findings

Reduces VGG16/19 and ResNet34/50 computation by nearly 5x on CIFAR10

Achieves 2x speedup on ImageNet without performance loss

Compresses weights by 10x and 4x respectively

Abstract

Existing convolution techniques in artificial neural networks suffer from huge computation complexity, while the biological neural network works in a much more powerful yet efficient way. Inspired by the biological plasticity of dendritic topology and synaptic strength, our method, Learnable Heterogeneous Convolution, realizes joint learning of kernel shape and weights, which unifies existing handcrafted convolution techniques in a data-driven way. A model based on our method can converge with structural sparse weights and then be accelerated by devices of high parallelism. In the experiments, our method either reduces VGG16/19 and ResNet34/50 computation by nearly 5x on CIFAR10 and 2x on ImageNet without harming the performance, where the weights are compressed by 10x and 4x respectively; or improves the accuracy by up to 1.0% on CIFAR10 and 0.5% on ImageNet with slightly higher efficiency. The code will be available on www.github.com/Genera1Z/LearnableHeterogeneousConvolution.