Introduction

FHEON is a configurable framework for building privacy-preserving convolutional neural networks (CNNs) using Homomorphic Encryption (HE). At its core, FHEON leverages the Cheon–Kim–Kim–Song (CKKS) scheme, a widely adopted approximate homomorphic encryption method designed for efficient computation on real-valued data. It enables computations directly on encrypted data, ensuring data confidentiality while performing inference in the encrypted domain. This capability allows users to run complete inference tasks in the encrypted domain without ever exposing the underlying inputs, thereby ensuring strong data confidentiality. In doing so, FHEON enables secure deployment of machine learning models in sensitive environments, bridging the gap between utility and privacy in encrypted inference.

Key features of FHEON include:

  • Optimized HE CNN Layers: Provide users with multiple varients of secure convolution, average pooling, ReLU, and fully connected layers.

  • Configurable Architecture: All CNN Layers can be customized with standard paramters such input/output channels, kernel size, stride, and padding.

  • Versatile Evaluation: Tested on multiple architectures including VGG-11, VGG-16, ResNet-20, and ResNet-34. It has also been tested on MNIST, CIFAR-10, and CIFAR-100

FHEON provides a flexible and efficient platform for researchers and developers to build HE-friendly neural networks without sacrificing accuracy or privacy.

Citation

If you use FHEON in your work, please cite the following paper:

@misc{njungle2025fheonconfigurableframeworkdeveloping,
   title={FHEON: A Configurable Framework for Developing Privacy-Preserving Neural Networks Using Homomorphic Encryption},
   author={Nges Brian Njungle and Eric Jahns and Michel A. Kinsy},
   year={2025},
   eprint={2510.03996},
   archivePrefix={arXiv},
   primaryClass={cs.CR},
   url={https://arxiv.org/abs/2510.03996}
}

Link: Read the paper