SSIMLayer: Towards Robust Deep Representation Learning via Nonlinear Structural Similarity

TL;DR

This paper introduces SSIMLayer, a nonlinear layer inspired by the human visual system that enhances deep neural network robustness and convergence by focusing on structural similarity, reducing the need for additional nonlinearities.

Contribution

The paper proposes a novel SSIMLayer that evaluates structural similarity directly, improving training convergence and robustness without extra nonlinear transformations.

Findings

Better convergence than traditional convolutional layers

Increased robustness against noise and adversarial attacks

Reduces need for subsequent nonlinear transformations

Abstract

Deeper convolutional neural networks provide more capacity to approximate complex mapping functions. However, increasing network depth imposes difficulties on training and increases model complexity. This paper presents a new nonlinear computational layer of considerably high capacity to the deep convolutional neural network architectures. This layer performs a set of comprehensive convolution operations that mimics the overall function of the human visual system (HVS) via focusing on learning structural information in its input. The core of its computations is evaluating the components of the structural similarity metric (SSIM) in a setting that allows the kernels to learn to match structural information. The proposed SSIMLayer is inherently nonlinear and hence, it does not require subsequent nonlinear transformations. Experiments conducted on CIFAR-10 benchmark demonstrates that the SSIMLayer provides better convergence than the traditional convolutional layer, bypasses the need for nonlinear transformations and shows more robustness against noise perturbations and adversarial attacks.