Filter Bank Regularization of Convolutional Neural Networks

TL;DR

This paper introduces a filter bank-based regularization method for convolutional neural networks that enhances robustness and generalization by embedding structural priors while allowing flexible learning.

Contribution

It proposes a novel regularization approach using structured filter banks to mold CNN kernels, combining traditional image priors with deep learning flexibility.

Findings

Faster convergence of CNNs with the proposed regularization.

Improved generalization performance over existing methods.

Guided CNN training towards common image structures.

Abstract

Regularization techniques are widely used to improve the generality, robustness, and efficiency of deep convolutional neural networks (DCNNs). In this paper, we propose a novel approach of regulating DCNN convolutional kernels by a structured filter bank. Comparing with the existing regularization methods, such as $\ell_1$ or $\ell_2$ minimization of DCNN kernel weights and the kernel orthogonality, which ignore sample correlations within a kernel, the use of filter bank in regularization of DCNNs can mold the DCNN kernels to common spatial structures and features (e.g., edges or textures of various orientations and frequencies) of natural images. On the other hand, unlike directly making DCNN kernels fixed filters, the filter bank regularization still allows the freedom of optimizing DCNN weights via deep learning. This new DCNN design strategy aims to combine the best of two worlds: the inclusion of structural image priors of traditional filter banks to improve the robustness and generality of DCNN solutions and the capability of modern deep learning to model complex non-linear functions hidden in training data. Experimental results on object recognition tasks show that the proposed regularization approach guides DCNNs to faster convergence and better generalization than existing regularization methods of weight decay and kernel orthogonality.