Recursive Least Squares for Training and Pruning Convolutional Neural Networks

TL;DR

This paper introduces a recursive least squares-based method for training and pruning CNNs, effectively reducing model complexity while maintaining performance, and applicable to feedforward networks as well.

Contribution

The paper presents a novel RLS-based algorithm that prunes CNNs layer by layer with performance recovery, outperforming existing pruning methods in efficiency and effectiveness.

Findings

Achieves more effective pruning than four popular algorithms.

Maintains full network performance after pruning.

Applicable to both CNNs and FNNs.

Abstract

Convolutional neural networks (CNNs) have succeeded in many practical applications. However, their high computation and storage requirements often make them difficult to deploy on resource-constrained devices. In order to tackle this issue, many pruning algorithms have been proposed for CNNs, but most of them can't prune CNNs to a reasonable level. In this paper, we propose a novel algorithm for training and pruning CNNs based on the recursive least squares (RLS) optimization. After training a CNN for some epochs, our algorithm combines inverse input autocorrelation matrices and weight matrices to evaluate and prune unimportant input channels or nodes layer by layer. Then, our algorithm will continue to train the pruned network, and won't do the next pruning until the pruned network recovers the full performance of the old network. Besides for CNNs, the proposed algorithm can be used for feedforward neural networks (FNNs). Three experiments on MNIST, CIFAR-10 and SVHN datasets show that our algorithm can achieve the more reasonable pruning and have higher learning efficiency than other four popular pruning algorithms.