Back to Search Start Over

FLoCoRA: Federated learning compression with low-rank adaptation

Authors :
Ribeiro, Lucas Grativol
Leonardon, Mathieu
Muller, Guillaume
Fresse, Virginie
Arzel, Matthieu
Source :
32nd European Signal Processing Conference EUSIPCO, Aug 2024, Lyon, France
Publication Year :
2024

Abstract

Low-Rank Adaptation (LoRA) methods have gained popularity in efficient parameter fine-tuning of models containing hundreds of billions of parameters. In this work, instead, we demonstrate the application of LoRA methods to train small-vision models in Federated Learning (FL) from scratch. We first propose an aggregation-agnostic method to integrate LoRA within FL, named FLoCoRA, showing that the method is capable of reducing communication costs by 4.8 times, while having less than 1% accuracy degradation, for a CIFAR-10 classification task with a ResNet-8. Next, we show that the same method can be extended with an affine quantization scheme, dividing the communication cost by 18.6 times, while comparing it with the standard method, with still less than 1% of accuracy loss, tested with on a ResNet-18 model. Our formulation represents a strong baseline for message size reduction, even when compared to conventional model compression works, while also reducing the training memory requirements due to the low-rank adaptation.

Details

Database :
arXiv
Journal :
32nd European Signal Processing Conference EUSIPCO, Aug 2024, Lyon, France
Publication Type :
Report
Accession number :
edsarx.2406.14082
Document Type :
Working Paper