Symmetric Convolutional Filters: A Novel Way to Constrain Parameters in CNN

TL;DR

This paper introduces symmetric convolutional filters to constrain CNN parameters, improving generalization and reducing redundancy, with comparative analysis against pruning methods.

Contribution

The paper presents a new symmetric filter constraint technique for CNNs, enhancing parameter efficiency and generalization compared to existing methods.

Findings

Models with symmetric filters show improved generalization.

Symmetric constraints reduce parameter redundancy.

Comparable or better performance than pruning techniques.

Abstract

We propose a novel technique to constrain parameters in CNN based on symmetric filters. We investigate the impact on SOTA networks when varying the combinations of symmetricity. We demonstrate that our models offer effective generalisation and a structured elimination of redundancy in parameters. We conclude by comparing our method with other pruning techniques.