Federated collaborative filtering enables privacy-preserving recommendation systems but faces challenges in capturing high-order interactions, reducing communication overhead, and minimizing accuracy degradation. To address these issues, we propose Federated Compressive Collaborative Filtering (FCCF), a novel framework that leverages tensor compressive learning for cross-client predictions. FCCF employs a tensor-based model with GNN-based extraction to efficiently represent multi-type item and dual-role user nodes and introduces a sketch-to-embedding projection for feature analysis. It performs inference directly on compressed data, eliminating the need for precise signal reconstruction, and employs client-specific sampling matrices along with regularization to enhance privacy and preserve local representations. Experiments demonstrate FCCF’s effectiveness in improving prediction accuracy, robustness to privacy noise, and communication efficiency.

Federated learning (FL) enables participants to collaboratively train models without compromising the privacy of their local data. However, it remains vulnerable to threats such as Byzantine attacks and privacy leaks. These challenges are inherently interdependent: privacy-preserving mechanisms limit access to model updates to protect private information, whereas Byzantine-robust techniques need such access for thorough analysis. Existing research has developed FL for protecting privacy and resisting Byzantine attacks, but most of these methods are effective only under the honest majority premise or based on the non-collusion assumption among servers, and a few of them require the assistance of trusted third parties (TTPs). To address these challenges, we propose a Byzantine-robust and privacypreserving FL framework (BRPPFL), which identifies Byzantine attackers in the compressed domain of compressed sensing (CS) and has the advantage of low communication overhead. To mitigate privacy leaks, BRPPFL combines secure two-party computation (2PC) protocols with homomorphic encryption (HE) primitives to create a privacy-preserving secure aggregation protocol. Additionally, it minimizes computational overhead by offloading complex operations to the offline phase. Experimental results demonstrate that BRPPFL can accurately and efficiently identify Byzantine attackers while protecting privacy. Even at an extreme compression rate (CR) of 0.1, our framework maintains acceptable defense performance.

Federated Learning (FL) has received widespread attention as a collaborative learning paradigm. Clients can collaboratively train a global model with server by uploading parameters instead of sharing the raw data, which guarantees the basic data privacy. However, recent research has highlighted that sensitive information can be inferred from shared updates or gradients, resulting in serious privacy leakage. Moreover, transmitting the updates or gradients can result in communication bottlenecks. In order to solve the privacy and communication problems, we propose a federated multilinear compressed learning framework (FedMCL), which considers the tensor structure of the data and performs multidimensional compression on the client’s raw data through multilinear compressed learning. Compared with vector-based compressed learning, it has better learning performance on multi-dimensional data. We generate proxy images from the measurement domain for training, which effectively defends against gradient inversion attacks. In addition, we introduce low-rank approximation method to compress model updates and reduce the communication overhead by transmitting small matrices instead of the original model updates. Experimental results show that our scheme can resist gradient inversion attacks under different compression rates while obtaining good classification performance, which has advantages in terms of privacy, performance and communication.

Federated learning (FL) facilitates collaborative training of a global model without sharing the participants’ raw data. Nevertheless, existing FL approaches still face three major issues: 1) How to propose a more efficient and secure privacy-preserving method; 2) How to verify the identity of participants to ensure they are not impersonators; 3) How to reduce the significant communication cost. To address the aforementioned concerns, several schemes have been proposed. However, these schemes suffer from flaws in security, efficiency, and functionality. Furthermore, few researches have considered the possibility of adversaries impersonating legitimate participants to undermine the integrity and availability of the model or launch a free-riding attack. In this paper, we first combine the advantages of secret sharing, Diffie-Hellman key agreement, and functional encryption to develop an authenticable secure multi-party computing algorithm (SDF-ASMC). This algorithm can guarantee the security of transmitted data and provide authentication functionality in the absence of a trusted third party. Moreover, an efficient, secure, and authenticable FL algorithm (ESAFL), which leverages compressed sensing and all-or-nothing transform, is introduced to reduce the transmission and encryption of local gradients. Then, only the final element of the transformed measurements is encrypted by our proposed SDF-ASMC to protect all the measurements. This method effectively improves the efficiency of our algorithm. In addition, ESAFL also tolerates participants’ dropout. Security analysis demonstrates that our proposed algorithms can securely aggregate local gradients. Finally, the extensive experiments demonstrate the practical performance of our proposed algorithms.

Federated learning (FL) offers a solution for mitigating the issue of data silo. However, FL faces threats to both robustness and privacy, which hinder the widespread application of FL. Most existing approaches focus on one of these threats or require significant resources to tackle both simultaneously. To meet the requirements of robustness and privacy, we propose a robust and privacy-preserving FL (RPFL) based on random selection and lightweight sharing. Our random selection method effectively invalidates malicious models to protect the integrity of the global model. On the other hand, we employ the technique of multi-party computation (MPC) to enhance privacy. To mitigate additional communication overhead and computation overhead introduced by MPC, we propose lightweight sharing. Besides, we adopt compressed sensing and parameter-clipping to further improve the communication efficiency and robustness of RPFL. We prove the performance of RPFL in terms of robustness, privacy, as well as efficiency. The extensive experimental results demonstrate that RPFL effectively improves the robustness and privacy of FL with only a negligible performance penalty.