PRoVeFL: Private Robust and Verifiable Aggregation in Federated Learning

NETFL: Network Efficient Fully Secure Federated Learning at Scale

A Secure Aggregation Scheme Based on Additive Secret Sharing in Federated Learning

Federated learning alleviates data silos through a “data-local, model-global” paradigm, but transmitting plaintext gradients exposes clients to reconstruction attacks from malicious servers. Existing secure aggregation methods face trade-offs among privacy, accuracy, and efficiency: homomorphic encryption incurs high overhead, differential privacy sacrifices accuracy, and lightweight secret-sharing schemes often lack weighted aggregation support and suffer accuracy degradation as client numbers grow. To address these limitations, we propose SecAGG, a lossless secure weighted aggregation scheme based on additive secret sharing. SecAGG adopts a three-tier architecture consisting of client clusters, cooperative servers, and a super server. Clients split weighted model parameters into random shares and distribute them to cooperative servers, which perform encrypted partial aggregation before the super server securely reconstructs the global model. Experimental results demonstrate that SecAGG achieves strict security against up to M-1 colluding servers under the semi-honest model while preserving FedAvg-equivalent accuracy with minimal computation and communication overhead, effectively balancing privacy, accuracy, and efficiency.

TriVer: a lightweight and client-verifiable secure aggregation with dropout tolerance for federated learning

Federated learning relies on a central server to aggregate client updates, making secure aggregation (SA) essential for privacy protection. Masking-based SA schemes dominate large-scale deployments due to their low overhead, yet they face a fundamental tension: efficient schemes assume a semi-honest server and provide no client-side cryptographic verification, while existing verifiable schemes admit algebraic degrees of freedom that allow a malicious server to forge results that pass all client-side checks. We present TriVer, a lightweight SA protocol that resolves this tension by augmenting a single-mask HPRF-based framework with a triple-binding verification mechanism grounded in elliptic curve cryptography. Each client independently verifies the aggregated result through three non-interactive checks: threshold signature verification, commitment aggregation consistency, and model-hash binding. Passing all three checks simultaneously is computationally equivalent to solving the elliptic curve discrete logarithm problem (ECDLP). We prove that TriVer satisfies client data privacy, aggregation correctness, and aggregation-result non-forgeability in the Random Oracle Model under ECDLP hardness, HPRF pseudorandomness, and hash collision resistance, against a fully malicious server that may collude with a subset of aggregators and clients. Experimentally, TriVer achieves a per-round latency of 1.96 s, approximately six times lower than LightVeriFL (11.68 s) and roughly twice the cost of Aion (0.97 s), with no forged result accepted across all evaluated attack types in our simulations, including the algebraic bypass attack; under the same attack, LightVeriFL’s single-constraint verification equation did not prevent forgery in our evaluated scenario. TriVer tolerates client dropout rates up to 40% with negligible recovery overhead, whereas LightVeriFL fails to complete aggregation at a 20% dropout rate.

Information-Theoretically Secure Aggregation for Lightweight Federated Learning: Resilient to Dropouts and Adversaries

On-device federated learning (FL) enables privacy-preserving and personalized model training on resource-constrained devices such as smartphones and IoT nodes. To reduce communication cost, sign-based methods (e.g., signSGD) transmit one-bit gradients. However, exposing gradient signs makes them vulnerable to inference attacks, while existing secure aggregation schemes are often incompatible with such methods or incur significant computational and communication overhead. We propose a lightweight and information-theoretically secure aggregation framework tailored for sign-based FL. The framework securely computes the majority vote (MV) polynomial through single-round secure multiplication, ensuring end-to-end information-theoretic security under the honest-majority assumption while revealing only the final aggregated sign to the server. To enhance efficiency and scalability, we introduce two key techniques. First, inverse-form exponent reduction halves the effective MV polynomial degree, reducing both communication and computation costs. Second, we propose single-round secure multiplication, achieving linear offline complexity and storage with only a single online communication. Together, these techniques reduce online communication by up to 99.5% and latency by up to 85.7% compared to conventional approaches. Also, by leveraging inherent MDS-code-based decoding, the framework achieves robustness against both dropouts and adversarial behaviors, yielding accuracy gains of up to 20.65% and 10.74%, respectively. Overall, the proposed framework establishes a practical foundation for large-scale, low-latency, and information-theoretically secure aggregation in sign-based FL.

A Byzantine-Robust Secure Federated Learning Scheme in Heterogeneous Data

Secure Federated Learning protects user privacy by keeping local gradients within the ciphertext space, thereby preventing direct exposure of raw gradients during aggregation. Paradoxically, this very protection introduces a fundamental vulnerability: the opacity of encrypted gradients blinds the server, rendering honest and malicious gradients indistinguishable. An attacker can exploit this opacity to inject manipulated gradients directly into the secure aggregation process, compromising the integrity of the global model. To address this issue, we propose SFLBR, a novel Byzantine-robust Secure Federated Learning framework. Specifically, we adopt the gradient median as a robust benchmark and introduce a layer-wise cosine similarity metric to differentiate malicious gradients. Furthermore, considering the complexity of non-independent and identically distributed (Non-IID) scenarios, we further design a proactive defense strategy that constrains the divergence among honest gradients, and augment it with a trust score function to improve robustness against potential misjudgments. Finally, we construct a series of secure communication protocols based on a lightweight dual-masking encryption mechanism, which enable efficient robust aggregation directly within the ciphertext space. Extensive experiments validate the effectiveness of SFLBR, demonstrating state-of-the-art (SOTA) defense performance across diverse attack scenarios.

DMSA-FL: secure and robust federated learning via double-masked secure aggregation with multiparty homomorphic encryption

Privacy preserving federated learning (FL) faces a crucial trade-off between privacy and efficiency. Specifically, traditional multi-party homomorphic encryption incurs high computational and communication costs, while schemes relying on a single mask are susceptible to selective aggregation attacks. To address these limitations, this paper proposes DMSA-FL as a practical and robust FL framework leveraging double-masked secure aggregation integrated with multiparty homomorphic encryption. We mitigate efficiency bottlenecks by designing a strategy that encrypts only low-dimensional mask seeds rather than high-dimensional gradients. This method utilizes a homomorphic pseudorandom generator for mask expansion. Furthermore, a robustness protocol based on threshold secret sharing over sparse sketches is implemented to ensure aggregation continuity during user dropouts. Theoretical analysis demonstrates that DMSA-FL is secure against collusion attacks and selective aggregation attacks. Extensive experiments on multiple datasets indicate that DMSA-FL maintains model accuracy comparable to plaintext baselines. Compared with representative federated learning schemes based on CKKS multiparty homomorphic encryption, our approach achieves improved computational efficiency and significantly reduces communication overhead. Moreover, the computational advantage becomes increasingly evident as the number of model parameters grows.