Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FedHENet: A Frugal Federated Learning Framework for Heterogeneous Environments

Alejandro Dopico-Castro, Oscar Fontenla-Romero, Bertha Guijarro-Berdiñas, Amparo Alonso-Betanzos, Iván Pérez Digón

TL;DR

This work proposes FedHENet, extending the FedHEONN framework to image classification by using a fixed, pre-trained feature extractor and learning only a single output layer, removing the carbon footprint associated with hyperparameter tuning in standard FL.

Abstract

Federated Learning (FL) enables collaborative training without centralizing data, essential for privacy compliance in real-world scenarios involving sensitive visual information. Most FL approaches rely on expensive, iterative deep network optimization, which still risks privacy via shared gradients. In this work, we propose FedHENet, extending the FedHEONN framework to image classification. By using a fixed, pre-trained feature extractor and learning only a single output layer, we avoid costly local fine-tuning. This layer is learned by analytically aggregating client knowledge in a single round of communication using homomorphic encryption (HE). Experiments show that FedHENet achieves competitive accuracy compared to iterative FL baselines while demonstrating superior stability performance and up to 70\% better energy efficiency. Crucially, our method is hyperparameter-free, removing the carbon footprint associated with hyperparameter tuning in standard FL. Code available in https://github.com/AlejandroDopico2/FedHENet/

FedHENet: A Frugal Federated Learning Framework for Heterogeneous Environments

TL;DR

This work proposes FedHENet, extending the FedHEONN framework to image classification by using a fixed, pre-trained feature extractor and learning only a single output layer, removing the carbon footprint associated with hyperparameter tuning in standard FL.

Abstract

Federated Learning (FL) enables collaborative training without centralizing data, essential for privacy compliance in real-world scenarios involving sensitive visual information. Most FL approaches rely on expensive, iterative deep network optimization, which still risks privacy via shared gradients. In this work, we propose FedHENet, extending the FedHEONN framework to image classification. By using a fixed, pre-trained feature extractor and learning only a single output layer, we avoid costly local fine-tuning. This layer is learned by analytically aggregating client knowledge in a single round of communication using homomorphic encryption (HE). Experiments show that FedHENet achieves competitive accuracy compared to iterative FL baselines while demonstrating superior stability performance and up to 70\% better energy efficiency. Crucially, our method is hyperparameter-free, removing the carbon footprint associated with hyperparameter tuning in standard FL. Code available in https://github.com/AlejandroDopico2/FedHENet/
Paper Structure (9 sections, 3 equations, 1 figure, 3 tables)

This paper contains 9 sections, 3 equations, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Method
  3. Experimental Results and Discussion
  4. Experimental Setup
  5. Accuracy and Robustness Analysis
  6. Frugal Efficiency and Sustainability
  7. Accuracy-Energy Trade-off
  8. Homomorphic Encryption Overhead
  9. Conclusions

Figures (1)

  • Figure 1: Accuracy vs. energy consumption per training round for CIFAR-10 (left) and CIFAR-100 (right) for $\alpha = 0.1$ and single-class (sc) scenarios with 100 clients. FedHENet achieves peak accuracy in a single round, while baselines require multiple rounds. In CIFAR-10, the two circles overlap. For clarity, in CIFAR-100 only the last 30 baseline rounds are shown.