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

Sudhakar Sah, Nikhil Chhabra, Matthieu Durnerin
Proceedings of the The 39th Canadian Conference on Artificial Intelligence, PMLR 318:52-63, 2026.

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 8$\times$ with less than 4% accuracy drop; on COCO, we achieve 2$\times$ compression of YOLOv5s without any accuracy drop and 2$\times$ compression of YOLOv5n with a 2.5% drop.

Cite this Paper

BibTeX
@InProceedings{pmlr-v318-sah26b, title = {CompressNAS : A Fast and Efficient Technique for Model Compression using Decomposition}, author = {Sah, Sudhakar and Chhabra, Nikhil and Durnerin, Matthieu}, booktitle = {Proceedings of the The 39th Canadian Conference on Artificial Intelligence}, pages = {52--63}, year = {2026}, editor = {Bouzar-Benlabiod, Lydia and Leung, Carson}, volume = {318}, series = {Proceedings of Machine Learning Research}, month = {25--29 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v318/main/assets/sah26b/sah26b.pdf}, url = {https://proceedings.mlr.press/v318/sah26b.html}, 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 8$\times$ with less than 4% accuracy drop; on COCO, we achieve 2$\times$ compression of YOLOv5s without any accuracy drop and 2$\times$ compression of YOLOv5n with a 2.5% drop.} }
Endnote
%0 Conference Paper %T CompressNAS : A Fast and Efficient Technique for Model Compression using Decomposition %A Sudhakar Sah %A Nikhil Chhabra %A Matthieu Durnerin %B Proceedings of the The 39th Canadian Conference on Artificial Intelligence %C Proceedings of Machine Learning Research %D 2026 %E Lydia Bouzar-Benlabiod %E Carson Leung %F pmlr-v318-sah26b %I PMLR %P 52--63 %U https://proceedings.mlr.press/v318/sah26b.html %V 318 %X 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 8$\times$ with less than 4% accuracy drop; on COCO, we achieve 2$\times$ compression of YOLOv5s without any accuracy drop and 2$\times$ compression of YOLOv5n with a 2.5% drop.
APA
Sah, S., Chhabra, N. & Durnerin, M.. (2026). CompressNAS : A Fast and Efficient Technique for Model Compression using Decomposition. Proceedings of the The 39th Canadian Conference on Artificial Intelligence, in Proceedings of Machine Learning Research 318:52-63 Available from https://proceedings.mlr.press/v318/sah26b.html.

Related Material