Yongzhao Zhang

Intrusion detection systems (IDS) can effectively monitor network traffic and accurately detect malicious behaviors. In Internet of Things (IoT) environments, the massive influx of heterogeneous, resource-constrained devices introduces more complex security challenges, making IDS even more crucial for maintaining network security. However, the presence of redundant or irrelevant features in network traffic can significantly degrade the detection performance of IDS. To address this issue, this paper proposes a Feature distribution and Gini Impurity Filter-based intrusion feature selection method (FGIF). It combines the cardinality and Gini impurity distributions of features within a dataset to construct a multi-parameter evaluation framework, which is used to define efficient feature filtering rules that eliminate redundant and irrelevant features. Theoretical analysis demonstrates that, compared to entropy-based methods, FGIF mitigates selection bias during the feature selection process and significantly reduces computational overhead. Experiments conducted on six widely used IDS benchmark datasets and five commonly adopted classification models further confirm its effectiveness. FGIF significantly reduces feature dimensionality while maintaining detection performance comparable to that of the full feature set. Moreover, compared to existing state-of-the-art methods, FGIF achieves a superior balance between dimensionality reduction and model performance.

Cyber Threat Intelligence (CTI) and serverless computing are two emerging technologies that have significantly impacted their respective domains in recent years. However, their interaction remains surprisingly underexplored. In this work, through in-depth semi-structured interviews with cybersecurity experts, we identify the trust issues within the CTI ecosystem that can be exploited to introduce fake CTI manipulation, enabling indirect attacks against entities with dynamic IP allocation, such as those in serverless computing. Furthermore, these attacks can be amplified by commercial CTI platforms due to their widespread adoption and sharing mechanisms. Based on these insights, we propose Ares, a novel attack strategy that leverages fake CTI manipulation to enable large-scale, stealthy indirect denial-of-service attacks against serverless infrastructures. We demonstrate the feasibility and impact of Ares through extensive evaluations in a controlled experimental environment. Our results show that Ares can rapidly and widely disseminate fake CTI within the CTI ecosystem, leading to an overall average reject rate of 23.03% and a high reject rate of up to 45.42% when accessing top websites in certain industries, while maintaining a low detection rate across state-of-the-art serverless security systems. These findings underscore the urgent need for more frequent communication and collaboration among CTI platforms and related stakeholders to develop a more robust trustworthiness model across the ecosystem.

Privacy-Preserving Federated Learning (PPFL) enables multiple clients to collaboratively train models by submitting secreted model updates. Nonetheless, PPFL is vulnerable to data poisoning attacks due to its distributed training paradigm in cross-silo scenarios. Existing solutions have struggled to improve the performance of PPFL under poisoned Non-Independent and Identically Distributed (Non-IID) data. To address the issues, this paper proposes a privacy-preserving federated prototype learning framework, named PPFPL, which enhances the crosssilo FL performance against poisoned Non-IID data while protecting client privacy. Specifically, we adopt prototypes as clientsubmitted model updates to eliminate the impact of poisoned data distributions. In addition, we design a secure aggregation protocol utilizing homomorphic encryption to achieve Byzantine-robust aggregation on two servers, significantly reducing the impact of malicious clients. Theoretical analyses confirm the convergence and privacy of PPFPL. Experimental results on public datasets show that PPFPL effectively resists data poisoning attacks under Non-IID settings.

Real-time audio streaming transmission and processing play a crucial role in time-sensitive applications such as food delivery services and ride-hailing platforms, where rapid response is essential. However, existing server-based audio streaming architectures struggle to handle the high concurrency of massive mobile devices efficiently. Traditional compression methods like MP3 and AAC offer limited compression ratios, while deep learning-based approaches often fail to meet the real-time transmission demands of edge computing environments. In this paper, we propose a novel edge-to-server audio streaming architecture that leverages Mel filter bank spectral features to achieve ultra-high compression efficiency. Our system integrates audio denoising, Mel feature extraction, and quantization-based compression at the edge, effectively suppressing environmental and device-induced noise while achieving an extreme compression ratio of 0.39% relative to the original uncompressed audio. Compared to conventional methods like MP3, our approach further reduces the file size by 96.1%. The decompressed Mel features remain task-independent, enabling seamless support for various general-purpose audio processing tasks in the server. We evaluate our system across three key audio tasks: speech recognition, speech emotion recognition, and audio classification. Extensive experiments on five different mobile devices demonstrate a 93.10% reduction in transmission latency at 1 Mbps bandwidth compared to 64 kbps MP3 audio, while maintaining task performance within a 5% deviation from state-of-the-art (SOTA) models across six mainstream audio datasets. These results highlight the efficiency, robustness, and scalability of our approach for real-time edge-to-server audio processing.

Federated learning (FL), a distributed computing paradigm, is vulnerable to poisoning attacks that impair model performance and privacy attacks that leak participant information. Existing FL defense schemes struggle to counter poisoning attacks under data heterogeneity and high privacy computation overhead, limiting the practicality of federated learning. To address these issues, this paper proposes a model-contrastive federated learning framework with lightweight privacy preservation and poisoning attack detection, named MCFL. Specifically, we design a novel model-contrastive term by aligning intermediate-layer representations of models in the local optimization function to promote consistency of model updates among benign participants. Additionally, we design a secure aggregation protocol that adopts two-server aggregation instead of the single server to resist poisoning attacks with lightweight privacy protection. The proposed MCFL is theoretically proven in terms of convergence, robustness, and privacy. Extensive experiments demonstrate the superiority of MCFL compared to existing FL defense schemes.