This paper proposes ROFCL, a robust one-bit federated compressive learning framework that mitigates compression distortion and channel noise over parallel binary symmetric channels. It features a client-side quantization-aware learnable coding to minimize compression distortion during training, and a server-side trust-weighted likelihood aggregation that performs robust information aggregation. In this method, quantization-aware learning within a single client, cross-client diversity within a single round, and accumulated history reputation across multiple rounds jointly achieve robustness in extreme one-bit compression.

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.

Asynchronous Federated Learning (AFL) has garnered significant attention in edge computing (EC) due to its adaptability. However, the frequent communication overhead inherent in AFL poses a critical challenge in resource-constrained EC settings. To address this challenge, we propose REAFL, an innovative framework that integrates the strategic selection of the optimal set of K devices with momentum gradient to reduce resource costs and enhance model performance. We begin by establishing a mathematical framework that characterizes the relationship between the selection of K devices and the global loss of AFL. Leveraging this theoretical insight, we formulate an optimization problem aimed at minimizing global loss under a resource budget, with a focus on determining the optimal device subset. Recognizing the dynamic nature of device selection, we incorporate Deep Reinforcement Learning (DRL) into REAFL to effectively learn the best device selection strategy while also considering client utility. Extensive experimental evaluations demonstrate the superior performance of REAFL over existing benchmarks, achieving up to a 24.03% improvement in model accuracy, a 61.07% reduction in energy consumption, and a 61.34% decrease in communication rounds.

Compressed sensing (CS) is a powerful signal pro-cessing technique that can simultaneously realize signal sampling, compression, and lightweight encryption. However, traditional CS reconstruction algorithms often face challenges such as high signal sparsity requirements, poor reconstruction accuracy, and low efficiency. Meta-learning, which optimizes parameter selection and update strategies during iterations, can enhance the efficiency and convergence speed of CS reconstruction. In this article, we present a Meta-learning-based Compressed Sensing Reconstruction method in the Cloud (MetaCSRC), an end-to-end imaging paradigm designed to address these challenges. We develop an adaptive sampling network for CS sampling on the local client and a deep meta-learning convolutional network for CS loss optimization on the cloud server. By leveraging the cloud's sufficient computing and storage resources, the computationally intensive reconstruction process is offloaded from the clients. Additionally, we incorporate local differential privacy noise into the measurements before uploading them to the cloud server to enhance the security of measurement transmission. Experimental results demonstrate that MetaCSRC excels in both reconstruction speed and accuracy, while also providing enhanced privacy protection.

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.

The rapid advances in deep learning have enabled the extraction of soft-biometric attributes from faces, raising privacy concerns regarding images collected for face recognition. To address this issue, we propose a privacy-preserving scheme for multiple soft-biometric attributes using Generative Adversarial Networks (GANs), aiming to effectively protect individual privacy while maintaining face recognition system performance. Our scheme incorporates two main modules: 1) The attribute encryption system, which identifies attributes to be hidden through chaotic cryptography and supports the recovery of original attributes. 2) Encoder features are adaptively transformed using Selective Transmission Units (STUs) within a generator featuring an encoder-decoder structure to generate perturbed images. The generated perturbed images are visually similar to the original images but differ in soft-biometric attributes. Experimental results show that our method maintains high face recognition accuracy while preserving privacy, and the quality of the generated images surpasses existing techniques.

Trust-aware recommendation plays an essential role in alleviating information overload by exploiting social relationships among users to build recommendation systems. However, recommendation systems in mobile social networks suffer from the sparse trust problem, which severely affects the reliability of trust propagation and the accuracy of the recommendation. The flourishing graph neural networks have revitalized Trust-aware Recommendation Systems. Therefore, we propose a sparse trust data mining method based on graph attention networks (TrustGAT) to mine potential trust information between entities in large-scale mobile social networks. First, the potential relationships between the trustors and the trustees are simulated in sparse data, and thereof an adaptive trust network is built. On this basis, features of trust-related information and items are learned in multi-head attention modules to enable the scheme stability. In addition, the implicit influence of trust is introduced to augment network node representation. Empirical results on three public benchmarks show that TrustGAT can make recommendation for users rapidly and accurately as well as alleviate the sparse trust problem effectively.

Recently, many biological studies reported that two groups of genes tend to show negatively correlated or opposite expression tendency in many biological processes or pathways. The negative correlation between genes may imply an important biological mechanism. In this study, we proposed a FCA-based negative correlation algorithm (NCFCA) that can effectively identify opposite expression tendency between two gene groups in gene expression data. After applying it to expression data of cell cycle-regulated genes in yeast, we found that six minichromosome maintenance family genes showed the opposite changing tendency with eight core histone family genes. Furthermore, we confirmed that the negative correlation expression pattern between these two families may be conserved in the cell cycle. Finally, we discussed the reasons underlying the negative correlation of six minichromosome maintenance (MCM) family genes with eight core histone family genes. Our results revealed that negative correlation is an important and potential mechanism that maintains the balance of biological systems by repressing some genes while inducing others. It can thus provide new understanding of gene expression and regulation, the causes of diseases, etc.