Accepted in the 22nd EAI International Conference on Security and Privacy in Communication Networks (SecureComm 2026), July 2026

Abstract: 

Machine Learning as a Service (MLaaS) platforms are increasingly adopted in regulated domains such as finance and healthcare, where protecting sensitive data and model parameters is paramount. Fully Homomorphic Encryption (FHE) offers rigorous privacy guarantees by enabling computation directly on encrypted data; however, its high computational overhead has limited practical deployment. Prior research has primarily focused on optimizing encrypted deep learning inference, leaving efficient FHE-based training largely unexplored. This paper addresses this gap by presenting a GPU-accelerated CUDA framework for FHE logistic regression models with full 128-bit security. Logistic regression is selected as a foundational case study because it incorporates a nonlinear activation function, making it an essential building block for extending FHE operations to neural networks. Our implementation executes all computations directly over ciphertexts, offloading heavy FHE operations, such as homomorphic addition, multiplication, and bootstrapping, to optimized CUDA GPU kernels. By leveraging GPU parallelism, the framework achieves significant efficiency gains while maintaining strong privacy guarantees. Our open-source design abstracts the complexities of FHE and CUDA, lowering the technical barrier and enabling practitioners to perform privacy-preserving model training without deep cryptographic or GPU programming expertise.

Implementation: Click Here!

There are no restrictions placed on the usage of this implementation except for a commercial use, but please cite our paper:

D. Davidson, M. Kang, Y. Lee, Z. Xiong, and J. Son, "GPU Accelerated Fully Homomorphic Encryption Framework for End-to-End Logistic Regression," The 22nd EAI International Conference on Security and Privacy in Communication Networks (SecureComm 2026), pp. 1-19, July 2026. 

@INPROCEEDINGS{DillonSecurecomm26,
author={Dillon Davidson and Minjae Kang and Yeonjoon Lee and Zuobin Xiong and Junggab Son },
booktitle={The 22nd EAI International Conference on Security and Privacy in Communication Networks (SecureComm)},
title={GPU Accelerated Fully Homomorphic Encryption Framework for End-to-End Logistic Regression},
year={2026},
month={July},
pages={1-19},}

Published in Security and Communication networks (Hindawi)

Data Availability Statement:

This data is bit-rate data from collected by streaming videos from YouTube. The data was collected using WireShark, and it measure the number of bits sent from server to client at 0.25 seconds intervals. Every video was streamed for exactly 20 minutes, so there are 4800 datapoints (a recording taken every 0.25 seconds at 20-minute intervals). The data is in CSV format and was read/written to and from CSV using the Python package PANDAS. In the CSV file, there are two columns: Time and Bytes. The time column increases at 0.25 increments and is a record of the time of the video at these increments. The Bytes column measures the number of bits at each time interval. For example (Time: 0.25, Bytes: 1701) denotes that 1401 bits were sent from YouTube to the client 0.25 seconds into the video stream. In this dataset, we recorded data from 41 videos and each video was streamed 100+ times (some videos had over slightly 100 streams while others had exactly 100) so that the dataset has 4345 CSV files. The name of each file is the name of the video as it appeared on YouTube, without any spaces.

The collected dataset for this research: Videos bitstream.zip

There are no restrictions placed on the usage of this data, but please cite the following paper: https://www.hindawi.com/journals/scn/2021/4702469/

Presented and published in the IEEE International Conference on Systems, Man, and Cybernetics (IEEE SMC 2020) A full version was submitted to IEEE Transactions on Dependable and Secure Computing (IEEE TDSC).

The proposed scheme has been implemented using Solidity version 0.5.16 and deployed on Rinkeby Testnet (an Ethereum test network for test operations of smart contracts) with Truffle. The contract ID is "0xC48f9bb74ebaD1EEf59967b6e2Ba245f4D37F89C".
The nodes in a group are implemented using Node.js 10.16.3 including web3.js and Crypto modules to communicate with Blockchain and to utilize cryptographic functions.
The file below is the source code of the proposed scheme:
AGKA_SC.zip
The solidity code has successfully passed formal verification process by means of the automated analysis tool OYNTE.
1. Test Result_OYENTE, 2. Test Result_OYENTE IPFS.

Presented and published in the 29th International Conference on Computer Communications and Networks (ICCCN 2020)

A Camera-ready version: ICCCN20.pdf
Source Codes and Datasets: IIS lab data for website.7z