Federated learning aims to complete model training without private data sharing, but many privacy risks remain. Recent studies have shown that federated learning is vulnerable to membership inference attacks. The weight as an important parameter in neural networks has been proven effective for membership inference attacks, but it leads to significant overhead. Facing this issue, in this paper, we propose a bias-based method for efficient membership inference attacks against federated learning. Different from the weight that determines the direction of the decision surface, the bias also plays an important role in determining the distance to move along the direction. Moreover, the number of bias is way less than the weight. We consider two types of attacks: local attack and global attack, corresponding to two possible types of insiders: participant and central aggregator. For the local attack, we design a neural network-based inference, which fully learns the vertical bias changes of the member data and non-member data. For the global attack, we design a difference comparison-based inference to determine the data source. Extensive experimental results on four public datasets show that the proposed method achieves state-of-the-art inference accuracy. Moreover, experiments prove the effectiveness of the proposed method to resist some commonly used defenses.
12 January 2026 in IEEE Internet of Things Journal
Deepfake (DF) technology poses a significant threat to the Internet of Things (IoT) by enabling identity spoofing and the dissemination of misinformation. Although numerous face forgery detectors have been developed to counter the risks of facial DFs, detecting forgeries across varying quality levels, particularly under extreme degradations, remains a critical challenge. Current face forgery detection methods largely rely on identifying low-level artifacts, which are highly susceptible to distortions introduced by image degradation. Recognizing that restoration can recover critical forensic cues from severely degraded forgeries, this study proposes a quality-agnostic DF detection framework that leverages a blind face restoration (BFR) model to enhance robustness. The framework incorporates an auxiliary restoration branch alongside the original detection pathway. While the original branch operates directly on the degraded input, the restoration branch performs detection on facial images restored by a BFR model. In addition, a saliency-guided restoration (SGR) objective is introduced to enhance alignment between the restoration and detection tasks. A restoration similarity-aware fusion mechanism is further designed to adaptively integrate predictions from both branches based on input quality. To assess the robustness of our approach under extreme degradation, we establish a specialized, real-world-inspired benchmark that simulates diverse degradation scenarios. Comprehensive experiments on both the proposed and existing benchmarks demonstrate that our method consistently achieves superior robustness in various degradation scenarios.
Federated learning provides a privacy-enhancing paradigm for distributed collaborative learning across diverse application scenarios, yet it inherently contains vulnerabilities that can lead to information leakage. As critical parameters in deep neural networks, biases play a decisive role in shaping decision boundaries. In this paper, we reveal that these biases introduce susceptibility within federated learning, making it vulnerable to membership inference attacks. Specifically, we examine two distinct manifestations of bias-based membership inference attacks (BMIA), distinguished by the roles in federated learning: a local attack launched by individual participants, and a global attack initiated by the central server. For both scenarios, we develop efficient BMIA strategies. We employ a shallow neural network to identify whether a given data point is part of the training dataset, and a straightforward difference-comparison mechanism to identify the specific participant’s dataset from which the data is derived. Furthermore, we introduce a novel adaptive noise pruning defense (ANPD) mechanism designed to effectively mitigate BMIA without substantially degrading the performance of the target model. Comprehensive experiments conducted on four public datasets covering location services, retail, healthcare, and image recognition demonstrate the superiority of our attacks, which achieve higher accuracy than state-of-the-art methods and are robust against prevalent defenses. Additionally, the experiments demonstrate that our proposed defense strategy significantly counteracts the proposed BMIA and other prevalent attack techniques.