CompressNAS : A Fast and Efficient Technique for Model Compression using Decomposition

TL;DR

CompressNAS introduces a global search framework for optimal tensor rank selection in CNN compression, significantly reducing model size with minimal accuracy loss, suitable for deployment on resource-constrained devices.

Contribution

The paper presents CompressNAS, a novel global search approach for tensor rank selection in model compression, improving efficiency and accuracy trade-offs over existing local methods.

Findings

Compresses ResNet-18 by 8x with less than 4% accuracy drop on ImageNet.

Achieves 2x compression of YOLOv5s without accuracy loss on COCO.

Develops STResNet, a new family of efficient compressed models.

Abstract

Deep Convolutional Neural Networks (CNNs) are increasingly difficult to deploy on microcontrollers (MCUs) and lightweight NPUs (Neural Processing Units) due to their growing size and compute demands. Low-rank tensor decomposition, such as Tucker factorization, is a promising way to reduce parameters and operations with reasonable accuracy loss. However, existing approaches select ranks locally and often ignore global trade-offs between compression and accuracy. We introduce CompressNAS, a MicroNAS-inspired framework that treats rank selection as a global search problem. CompressNAS employs a fast accuracy estimator to evaluate candidate decompositions, enabling efficient yet exhaustive rank exploration under memory and accuracy constraints. In ImageNet, CompressNAS compresses ResNet-18 by 8x with less than 4% accuracy drop; on COCO, we achieve 2x compression of YOLOv5s without any accuracy drop and 2x compression of YOLOv5n with a 2.5% drop. Finally, we present a new family of compressed models, STResNet, with competitive performance compared to other efficient models.