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 learning (FL) enables multi-institutional MRI reconstruction without centralizing patient data, serving as a fundamental privacy-preserving technology for medical multimedia analysis and retrieval. However, practical deployment of FL-based reconstruction in clinical retrieval pipelines is hindered by three key obstacles: prohibitive edge computation costs, statistical heterogeneity that distorts diagnostic features, and the server blindness inherent in Secure Aggregation (SecAgg) which complicates quality auditing. In this paper, we propose FedLUR, a secure, lightweight, and intelligent federated framework designed to restore high-fidelity images for reliable downstream retrieval. FedLUR freezes a compact, physics-consistent deep unfolding backbone and communicates only lightweight visual prompts, reducing communication overhead compared to full-model baselines. To ensure the semantic consistency required for accurate image retrieval, we introduce a null-space constrained optimization strategy that stabilizes prompt tuning against client drift, thereby preserving global diagnostic structures. Furthermore, to reconcile SecAgg with utility-aware coordination, FedLUR utilizes a deterministic Large Language Model (LLM) agent to calibrate aggregation weights based on privacy-friendly metadata, ensuring robust convergence without exposing raw updates. Experiments demonstrate that FedLUR achieves state-of-the-art reconstruction quality. A proxy retrieval test shows highly consistent features between reconstructed and ground-truth images, supporting downstream medical multimedia retrieval.

Data collection and sharing have a tremendous impact on technology, business and society. Correspondingly, it brings in significant privacy and communication concerns. To this end, we present a hierarchical privacy-preserving and communication-efficient compression scheme via compressed sensing (CS) to address these two issues. In the encoding stage, the obfuscated sensitive regions and non-sensitive regions are compressed and encrypted simultaneously. Consequently, the semi-authorized users and authorized users are considered in the decoding stage. Additionally, the left annihilator matrices provide various kinds of recovery qualities for real-world requirements, which further achieves communication-efficient compression.

Compressed sensing (CS) has attracted wide attention in signal process field, which states that a sparse signal or a compressive signal can be reconstructed by using a small number of CS samples efficiently. In recent years, CS-based image coding methods have been investigated extensively. However, the ratio-distortion (R-D) performance of this method is unsatisfactory in comparison with traditional compression algorithms, such as JPEG and JPEG2000.

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.

Compressed sensing (CS), a popular signal processing technique, can achieve compression and encryption simultaneously. Therefore, it has extension applications in various fields. However, CS is vulnerable to cryptographic attacks for its linear encoding process. To solve this problem, a permutation-diffusion structure is designed and embedded to the CS encoding process. In addition, it can increase the key space while compressing. Since the permutation-diffusion structure reduces the sparseness, superior recovery performance cannot be achieved. Therefore, the multi-prior regularization recovery strategy is designed to improve the recovery performance, where the multi-prior regularization term denotes l1 norm, total variation (TV) and low rank. The simulation results and analyses demonstrate that the proposed encoding scheme can resist cryptographic attacks, increase the key space while compressing, and achieve 1.54dB PSNR gain on average in comparison with the existing schemes.

As a new signal processing technology, compressed sensing (CS) has been showed to be a promising solution for compressing cipher images. However, the previous CS-based schemes are unsatisfactory in terms of ratio-distortion (R-D) performance. In order to solve this problem, an image encryption-then-compression (ETC) scheme by using semi-tensor product CS (STP-CS) and pre-mapping is proposed in this paper. In the proposed scheme, the original image is encrypted by using the scrambling operation. After image encryption, the cipher image is compressed through three steps. Firstly, the original image is compressed by using STP-CS. Secondly, the CS samples are processed by using pre-mapping operation. Thirdly, the resultant CS samples are quantized and encoded into bits. For image signal recovery, an iterative bivariate shrinkage (IBS) algorithm is proposed. Compared with the existing CS-based image ETC schemes, the proposed scheme has better R-D performance.

In recent years, using compressed sensing (CS) as a cryptosystem has drawn more and more attention since this cryptosystem can perform compression and encryption simultaneously. However, this cryptosystem is vulnerable to known-plaintext attack (KPA) under multi-time-sampling (MTS) scenario due to the linearity of its encoding process. In this paper, a privacy-preserving CS scheme for image compression-encryption applications is proposed, which embeds a non-linear operation called noise injected negative-positive transformation (NINPT) in the CS encoding process with the purpose of withstanding KPA. The encoding procedures of the proposed scheme include three steps. First, the original image is encrypted by using NINPT operation. Second, the intermediate ciphertext is re-encrypted and compressed by using CS simultaneously. Third, the final compressed ciphertext is quantized into bits via scalar quantization (SQ). Since the introduction of NINPT operation in the CS encoding process breaks the linearity of the CS sampling process, the proposed scheme can withstand KPA under MTS scenario. For image signal reconstruction, a projected Landweber with embedding decryption (PL-ED) algorithm is proposed. Simulation results demonstrate that the proposed scheme can withstand KPA under MTS scenario at the cost of slightly sacrificing the compression performance.