federated homomorphic encryption

本文提出基于RLWE的多密钥同态加密协议，解决联邦学习中模型更新泄露敏感信息的问题，实现私有平均聚合。

To address the high computational overhead and ciphertext expansion caused by fully homomorphic encryption (FHE) in federated learning, this paper proposes FedBit, a software–hardware co-optimization framework. Methodologically, FedBit introduces (i) bit-interleaved packing—a novel technique that embeds multiple model parameters into individual coefficients of BFV ciphertexts, overcoming conventional word-level packing constraints—and (ii) an FPGA-specific FHE accelerator with a memory-aware dataflow, enabling fine-grained parallelism and efficient hardware resource utilization. Experimental results demonstrate that FedBit achieves a 100× speedup in encryption throughput and reduces communication overhead by 60.7% compared to baseline FHE-based federated learning systems, while preserving model accuracy. To the best of our knowledge, FedBit is the first end-to-end system for privacy-preserving federated learning that jointly supports bit-level ciphertext compression and hardware-native acceleration.

In federated learning (FL), model updates are vulnerable to gradient reconstruction and membership inference attacks, while pure fully homomorphic encryption (FHE)—e.g., BFV—suffers from ciphertext expansion and prohibitive computational overhead, hindering deployment on resource-constrained edge clients. To address this, we propose the first efficient hybrid homomorphic FL framework: lightweight symmetric encryption (PASTA) at clients for fast update encryption, coupled with BFV-based FHE for secure key transmission and server-side aggregation, implemented atop Flower for cross-device secure aggregation. Our design balances communication efficiency and strong privacy: it achieves 97.6% accuracy on MNIST, reduces upload bandwidth by over 2000×, cuts client runtime by 30%, and prevents gradient leakage. The trade-off is increased server-side computation. The core contribution is the first practical, edge-aware hybrid homomorphic FL architecture, overcoming key deployment bottlenecks of FHE in real-world FL systems.

To address the high online encryption overhead in Fully Homomorphic Encryption (FHE) systems—which critically limits throughput in high-load scenarios such as outsourced databases—this paper proposes a compile-time ciphertext synthesis framework. It shifts ciphertext generation entirely to compilation time via precomputed basis vectors, zero-encryption reuse, and composition of homomorphic addition and scalar multiplication, enabling runtime-zero encryption during data ingestion. We formally define “random-mode homomorphism” for the first time and prove its IND-CPA security via a hybrid game, rigorously characterizing the security boundaries of basis reuse and structured noise injection. The scheme remains compatible with standard FHE APIs while preserving layout semantics for downstream homomorphic operations. Experimental results demonstrate substantial improvements in batch encoding throughput, establishing an efficient, secure, and deployable paradigm for ciphertext injection in high-throughput FHE pipelines.

In federated learning, sharing gradients exposes models to gradient inversion attacks, compromising data privacy; while fully homomorphic encryption (FHE) offers strong protection, its computational overhead is prohibitive. Existing selective encryption approaches rely on heuristic parameter tuning and lack theoretical foundations. This paper establishes the first theoretical framework for selective gradient encryption, revealing an intrinsic relationship between gradient spectral characteristics and privacy preservation capability. It formally characterizes how encryption ratio, model complexity, and exposed gradient volume jointly govern defense efficacy. Methodologically, we integrate gradient spectral analysis with lightweight homomorphic encryption to achieve a Pareto-optimal trade-off between privacy and efficiency. Extensive experiments validate the effectiveness of our key design principles. Our work provides both a rigorous theoretical foundation and a systematic design methodology for practical, scalable privacy-preserving federated learning.

This work addresses the high computational and memory overheads that hinder the practical deployment of fully homomorphic encryption (FHE) in privacy-preserving machine learning, which stem from the inherent complexity of cryptographic operations and inefficient ciphertext packing. To overcome these limitations, the authors propose FEnc², a fragment-based unified encoding framework that treats encrypted tensor layout as a first-class design dimension in FHE systems. By jointly optimizing spatial locality and feature grouping through convolution-aware encoding and architecture-aware ciphertext compression, FEnc² substantially improves slot utilization while reducing rotation complexity and ciphertext count. Built upon the CKKS scheme and compatible with NTT and key-switching accelerations, FEnc² achieves up to 228.83× (GPU) and 226.06× (CPU) speedup for LeNet inference on MNIST, and 4.55× (GPU) and 9.43× (CPU) acceleration for MobileNet inference on ImageNet.

As security demands increase, the importance of secure computation technologies grows, yet these technologies can often seem overwhelming to practitioners. Furthermore, many approaches focus only on a single technology, potentially overlooking superior alternatives. This work aims to address the issue of selecting the right technology for secure computation by presenting a comparative analysis of two highly relevant cryptographic methods and their software implementations, with a particular focus on machine learning. Firstly, we provide a theoretical summary and comparison of the secure computation paradigms of secure multi-party computation (SMPC) and fully homomorphic encryption (FHE). We outline the advantages and limitations of the protocols, as well as the relevant open-source software implementations. Secondly, we present the results of extensive benchmarking of the main software frameworks identified for machine learning operations and models. Regarding the current state of the art in FHE, we observe that it outperforms SMPC for regressions. Additionally it may be faster for simple dense networks using GPUs or Hybrid Models. Conversely, SMPC showed superior performance for complex models such as CNNs. Our results should pave the way for more technology-agnostic benchmarking of secure computation technologies for machine learning, providing guidance for practitioners looking to adopt these technologies.

This work addresses the lack of systematic parameter configuration guidelines for CKKS homomorphic encryption in privacy-preserving personalized federated learning, where balancing security, accuracy, and efficiency remains challenging. The authors propose pFedCKKS, the first framework to establish a complete set of CKKS parameter constraints tailored to personalized federated learning under 128-bit security, reducing complex parameter selection to determining the inner and outer ciphertext primes. Implemented using Flower and TenSEAL and integrated with algorithms such as FedPer, Ditto, and FedFinetune, extensive experiments on FEMNIST, CelebA, and Sentiment140 empirically reveal the trade-offs between model accuracy and computational/communication overheads induced by parameter choices, yielding practical configuration guidelines for real-world deployment.

This work addresses the dual privacy risks in cloud-based AI inference—exposure of both user inputs and model weights—and tackles the impractical computational overhead of fully homomorphic encryption (FHE). To bridge this gap, the authors propose a co-design paradigm that integrates FHE with AI inference through a novel “meet-in-the-middle” optimization framework. This approach jointly tailors cryptographic primitives and neural network architectures: on one side, it customizes an FHE scheme and compiler to align with the static structure of the inference circuit; on the other, it imposes architectural constraints on the AI model to minimize dominant homomorphic operations. The resulting synergy substantially reduces FHE inference costs, offering a practical and efficient pathway toward privacy-preserving AI inference.

This work addresses the privacy risks in federated learning, where model updates may inadvertently leak sensitive user information. To mitigate this, the authors propose a novel federated learning framework that integrates homomorphic encryption with differential privacy. Specifically, homomorphic encryption enables secure aggregation of model updates in encrypted form, while differential privacy introduces calibrated noise to these updates, thereby providing dual-layer privacy protection without requiring clients to upload raw local data. The efficacy of the approach is empirically validated on real-world datasets—including Framingham, Pima Indians Diabetes, and Bank Marketing—demonstrating its ability to maintain high model accuracy while ensuring strong privacy guarantees in sensitive domains such as healthcare and finance. The study also systematically investigates the impact of data heterogeneity on performance and presents corresponding optimization strategies.