Evolving Normalization-Activation Layers

TL;DR

This paper introduces EvoNorms, a novel class of normalization-activation layers discovered through automated evolution, which outperform traditional layers across various vision tasks by unifying normalization and activation in a flexible, evolved structure.

Contribution

The paper presents a unified, evolutionary approach to designing normalization-activation layers, resulting in EvoNorms with novel structures that surpass existing methods in multiple vision applications.

Findings

EvoNorms outperform BatchNorm and GroupNorm in image classification.

EvoNorms transfer effectively to segmentation and image synthesis tasks.

The method discovers layers with unconventional structures not based on sequential normalization and activation.

Abstract

Normalization layers and activation functions are fundamental components in deep networks and typically co-locate with each other. Here we propose to design them using an automated approach. Instead of designing them separately, we unify them into a single tensor-to-tensor computation graph, and evolve its structure starting from basic mathematical functions. Examples of such mathematical functions are addition, multiplication and statistical moments. The use of low-level mathematical functions, in contrast to the use of high-level modules in mainstream NAS, leads to a highly sparse and large search space which can be challenging for search methods. To address the challenge, we develop efficient rejection protocols to quickly filter out candidate layers that do not work well. We also use multi-objective evolution to optimize each layer's performance across many architectures to prevent overfitting. Our method leads to the discovery of EvoNorms, a set of new normalization-activation layers with novel, and sometimes surprising structures that go beyond existing design patterns. For example, some EvoNorms do not assume that normalization and activation functions must be applied sequentially, nor need to center the feature maps, nor require explicit activation functions. Our experiments show that EvoNorms work well on image classification models including ResNets, MobileNets and EfficientNets but also transfer well to Mask R-CNN with FPN/SpineNet for instance segmentation and to BigGAN for image synthesis, outperforming BatchNorm and GroupNorm based layers in many cases.