Research

FALCON-Net is a detector for AI-generated images that combines two modules: the Intrinsic Noise Pattern Isolation (INP) module, which isolates device-specific noise by analyzing high-frequency features in the frequency domain, and the Local Variation Pattern (LVP) module, which models relationships between local pixels to capture unnatural intensity variations. The source reports state-of-the-art detection performance and good generalization to unseen generative models, and the code is available at https://github.com/humiaomiaohaha/FALCON-Net.

Researchers show that screaming-channel side-channel leakage on mixed-signal devices appears at frequencies beyond the clock harmonics, not only at the harmonics where earlier work collected it. They combine multiple frequencies and demonstrate attacks at 15 and, for the first time, 30 meters in an RF-polluted environment. Frequency combination reduces the number of traces needed to reach a given attack performance by at least 2x and up to 3.76x.

The paper studies privacy risks in Wi-Fi-based wireless sensing, where signals that reveal movement status and sleep quality can be intercepted by unauthorized users. It proposes a multi-antenna signal processing strategy at the physical layer and models sensing as a communication task to define a sensing rate region. It characterizes two boundary points, P_OS and P_OPP, and a secure sensing rate R_PP, validated through numerical simulations.

PriLabel is a static auditor that checks whether Android apps omit disclosures from their Google Play privacy labels. It builds on Iterative Context Reconstruction (ICR), which rebuilds context from decompiled code to resolve semantic ambiguity, and maps transmitted data to Google's label taxonomy without source code or an ontology. Tested on 4,851 top-installed apps, it reported 91.5% precision and found 2,374 apps omitting at least one disclosure, 210 of which transmitted sensitive financial data such as credit card numbers without proper labeling.

This paper examines a smart jamming attack on the physical uplink shared channel (PUSCH) in 5G NR. The adversary sniffs the downlink control information (DCI) to learn the scheduled resources, then jams the PUSCH transmission precisely and covertly. The authors propose a suppression method that identifies the DCI-scheduled subset and reconstructs PUSCH resources, relying on spatial-domain differences between legitimate users and the attacker. Numerical results are reported to confirm its effectiveness.

The paper proposes the first task-adapted compressed sensing MRI (CS-MRI) method built on information-theoretic principles, which maximizes mutual information between undersampled k-space measurements and clinical tasks to enable probabilistic inference and uncertainty prediction. A single end-to-end trained model adapts to arbitrary sampling ratios and supports both joint task-and-reconstruction and reconstruction-suppressed settings for privacy protection. On large-scale MRI datasets, it reaches competitive Dice scores against a deterministic counterpart and better matches the ground-truth posterior distribution as measured by generalized energy distance (GED).

The source is a bibliographic record for a June 2026 article in the Journal of Information Security and Applications, Volume 99, by Jianzhong Wang, Lifei Wei, Jinjiao Zhang, Kai Zhang and Jianting Ning. The title indicates a study of efficient malicious-secure multi-party private set union that uses trusted execution environments. The provided text contains no abstract, method details, or findings.

Quadruplet Augmentation (QuadAug) is a method for Online Continual Learning, which learns from streaming data with unknown task boundaries. It targets two requirements, attribute invariance and structure invariance, to overcome the shortcut feature trap and limited plasticity of previous methods. The authors report significant improvements on four sequential datasets and three blurry datasets.

Researchers address node classification in a newly formed hypergraph, where labeled nodes are scarce and costly to obtain. They propose LHCCA (Local and High-order Consistency Coding and Adaptation), which transfers knowledge from a well-labeled source hypergraph by combining local and high-order consistency representations through attention, then aligns source and target features with adversarial domain adaptation and contrastive learning. Experiments on several real-world datasets are reported to show the model's effectiveness.

The paper, published in Computers & Security (Volume 167) in August 2026 by Zhaoxin Jin, Tianbo Lu, Hanrui Chen and Fangyi Yu, is titled "Enhancing website fingerprinting through combined data augmentation strategies." The source text provides only publication metadata, so the study's method and findings are not described here.

This paper proposes a hybrid detection method combining autoencoder and transformer models to catch the unresponsive ECN attack, a novel attack on low-latency network services. Compared with the current state-of-the-art on a large real-life network traffic dataset, it cuts detection error rate by more than 10%. At a false-alarm rate below 10−4, it reaches over 90% true-positive detection.