Despite federated learning (FL)'s potential in collaborative learning, its performance has deteriorated due to the data heterogeneity of distributed users. Recently, clustered federated learning (CFL) has emerged to address this challenge by partitioning users into clusters according to their similarity. However, CFL faces difficulties in training when users are unwilling to share their cluster identities due to privacy concerns. To address these issues, we present an innovative Efficient and Robust Secure Aggregation scheme for CFL, dubbed EBS-CFL. The proposed EBS-CFL supports effectively training CFL while maintaining users' cluster identity confidentially. Moreover, it detects potential poisonous attacks without compromising individual client gradients by discarding negatively correlated gradients and aggregating positively correlated ones using a weighted approach. The server also authenticates correct gradient encoding by clients. EBS-CFL has high efficiency with client-side overhead O(ml + m^2) for communication and O(m^2l) for computation, where m is the number of cluster identities, and l is the gradient size. When m = 1, EBS-CFL's computational efficiency of client is at least O(log(n)) times better than comparison schemes, where n is the number of clients. In addition, we validate the scheme through extensive experiments. Finally, we theoretically prove the scheme's security.

Federated Learning (FL) is a distributed machine learning framework that allows for model training across multiple clients without requiring access to their local data. However, FL poses some risks, for example, curious clients might conduct inference attacks (e.g., membership inference attacks, model-inversion attacks) to extract sensitive information from other participants. Existing solutions typically fail to strike a good balance between performance and privacy, or are only applicable to specific FL scenarios. To address these challenges, we propose a universal and efficient privacy-preserving FL framework based on matrix theory. Specifically, we design the Improved Extended Hill Cryptosystem (IEHC), which efficiently encrypts model parameters while supporting the secure ReLU function. To accommodate different training tasks, we design the Secure Loss Function Computation (SLFC) protocol, which computes derivatives of various loss functions while maintaining data privacy of both client and server. And we implement SLFC specifically for three classic loss functions, including MSE, Cross Entropy, and L1. Extensive experimental results demonstrate that our approach robustly defends against various inference attacks. Furthermore, model training experiments conducted in various FL scenarios indicate that our method shows significant advantages across most metrics.

Federated learning (FL) is a promising privacy-preserving distributed machine learning paradigm. However, data privacy leakage and Byzantine clients are common challenges in the FL aggregation phase. While extensive research has been conducted to explore defenses for these risks independently, there is a notable lack of scholarly work on integrated defense strategies to address both challenges simultaneously. To bridge this gap, we propose a novel two-party secure robust aggregation (TPSRA) framework. The critical insight of TPSRA is to couple client-side gradient obfuscation with server-side secure two-party computation to achieve robust and private FL aggregation. Specifically, clients obfuscate and split local gradients using matrix theory, while servers utilize a novel secure multiparty computation protocol based on mutually orthogonal matrices to preserve the privacy of local gradients. Additionally, TPSRA designs and integrates state-of-the-art robust aggregation algorithms into compatible subprotocols, enabling efficient parallel computation. This establishes a highly efficient and versatile secure robust aggregation framework for FL. Experiments demonstrate that our TPSRA framework not only effectively resists gradient leakage attacks and detects malicious gradients, but also exhibits superior computational and communication efficiency. We also prove theoretically that TPSRA is secure under the semi-honest adversary model.