Cheng Qiao

Graph neural networks (GNNs) are a prominent paradigm in dealing with graph structured data as they are able to explore rich relational information between nodes with attributes. That said, improving the robustness of GNNs becomes increasingly important since recent studies have shown that attackers can catastrophically ruin the performance of GNNs through injecting unnoticeable adversarial edges to graphs. Prior defense approaches focused on reducing the adversarial effect via deleting adversarial edges or designing a robust framework. However, this defense strategy is limited in (1) preserving the integrity of graph structure and (2) improving the expressive power of node representations. To address this limitation, we propose a lightweight purified graph neural network (Pure-GNN), which handles the adversarial effect by assigning importance weights to edges and utilizes a residual mechanism to improve the quality of node representations. With this design, it cannot only protect the intact information of graph topology but also learn the representations with significantly improved runtime efficiency. Extensive experiments over real-world datasets demonstrate that Pure-GNN outperforms the state-of-the-art approaches in defending against various adversarial attacks.

The growing heterogeneity of consumer-centric IoT and the continuous evolution of cyber threats are increasingly rendering traditional intrusion detection systems insufficient. Such systems suffer from data silos, severe class imbalance, and limited adaptability to novel attacks. To address these challenges, we propose Falcon, a federated incremental learning framework for intrusion detection in heterogeneous consumer-centric IoT traffic. Falcon enables collaborative training across multiple clients and remains effective under non-IID client data distributions and concept drift. It combines a hybrid data-balancing scheme with a CNN–BiLSTM–Attention backbone for learning rich, class-balanced representations, and a drift-triggered incremental module that integrates KL-based drift detection with a dual-head classifier, memory replay, and multi-teacher distillation to learn emerging attacks without catastrophic forgetting. Experiments on three benchmark datasets show that Falcon achieves absolute Precision gains of up to 20.76% over strong centralized and federated baselines in both binary and multi-class tasks.

Secure outsourced computation (SOC) provides secure computing services by taking advantage of the computation power of cloud computing and the technology of privacy computing (e.g., homomorphic encryption). Expanding computational operations on encrypted data (e.g., enabling complex calculations directly over ciphertexts) and broadening the applicability of SOC across diverse use cases remain critical yet challenging research topics in the field. Nevertheless, previous SOC solutions frequently lack the computational efficiency and adaptability required to fully meet evolving demands. To this end, in this paper, we propose a toolkit for TEE-assisted (Trusted Execution Environment) SOC over integers, named TRUST. In terms of system architecture, TRUST falls in a single TEE-equipped cloud server only through seamlessly integrating the computation of REE (Rich Execution Environment) and TEE. In consideration of TEE being difficult to permanently store data and being vulnerable to attacks, we introduce a (2, 2)-threshold homomorphic cryptosystem to fit the hybrid computation between REE and TEE. Additionally, we carefully design a suite of SOC protocols supporting unary, binary and ternary operations. To achieve applications, we present SEAT, secure data trading based on TRUST. Security analysis demonstrates that TRUST enables SOC, avoids collusion attacks among multiple cloud servers, and mitigates potential secret leakage risks within TEE (e.g., from side-channel attacks). Experimental evaluations indicate that TRUST outperforms the state-of-the-art and requires no alignment of data as well as any network communications. Furthermore, SEAT is as effective as the Baseline without any data protection.

Adversarial examples threaten the stability of Generative AI (GAI) in consumer electronics (CE), but existing attack strategies either rely solely on gradient information—yielding perturbations that are easily defended—or impose computational demands that far exceed the power budgets of CE devices. We propose a fast, entropy-guided framework that generates adversarial perturbations by focusing on an image’s most uncertain regions. This approach is extended to EDM-GAN, a lightweight generator that creates potent, stealthy attacks conditioned on an input image and its entropy map. Extensive experiments show our method boosts attack success rates by 4%, improves perceptual quality (SSIM +4.5%), and, when used for adversarial training, increases robust accuracy by 4.6%. Our work provides an efficient tool for both attacking and defending GAI models on CE devices.

This study investigates the security issues of networked control systems (NCSs) under distributed denial of service (DDoS) attacks. The DDoS attacks overwhelm a target system by flooding it with excessive fake traffic or requests, causing network congestion. This prevents the system from processing legitimate control requests and ultimately leads to system failure. First, to free up the substantial memory resources on the controller and mitigate the impact of distributed DDoS attacks on the performance of NCSs, we design a deep learning supervised DDoS attack detection and restart control mechanism that ensures that the duration of each attack is effectively limited to the tolerable interval of the system. In addition, a straightforward Lyapunov–Krasovskii functional (LKF) is formulated to analyze the stability of NCSs facing DDoS attacks under the restart control mechanism. Furthermore, it simplifies the stability analysis by avoiding complex calculations and diminishes the conservativeness of the stability criteria through the integration of sophisticated integral inequality techniques. Finally, simulations with an autonomous ground vehicle (AGV) model validate the practical benefits and effectiveness of the proposed theoretical advancements.

In the era of computational social systems (CSS), where AI is increasingly embedded in critical domains such as public services, healthcare, and social governance, concerns about system vulnerabilities have become paramount. Federated learning (FL) offers a promising solution by enabling privacy-preserving and secure AI model training in real-world applications. However, traditional FL approaches often assume that client data is either imbalanced or non-independent and identically distributed (non-IID), whereas in practice, these issues frequently coexist. In such complex scenarios, minority classes may appear only on a small subset of clients, significantly degrading global model performance and convergence. In this article, we propose a novel FL framework designed to tackle the intertwined challenges of data imbalance and non-IID data by a principled, decoupled approach. Unlike existing methods that address these issues in isolation, our approach introduces three synergistic innovations, each operating at a different level of the FL system. At the data level, a global rebalancing mechanism employs a privacy-preserving modified Z-score to rectify class imbalance. At the client level, a distribution-aware clustering strategy groups participants based on data subspace similarity. At the feature level, a novel contrastive loss function, leveraging the effective number of samples, simultaneously mitigates local imbalance and model drift. This cohesive framework ensures that interventions at each level reinforce each other, leading to robust performance. Extensive experiments show that FedScFc outperforms state-of-the-art baselines.

Unmanned aerial vehicle-based mobile crowdsensing (UAV-MCS) has emerged as a promising paradigm for large-scale data collection in low-altitude environments. Efficient task allocation is critical in UAV-MCS to reduce the flight distance of UAVs and ensure timely data acquisition. However, since UAVs typically depart from the locations of their operators, task allocation based on geographical coordinates can leak the location privacy of both UAV operators (called workers for simplicity) and task requesters. Existing privacy-preserving task allocation schemes often focus solely on protecting the privacy of UAV operators, while neglecting that of task requesters, and many fail to achieve accurate allocation results, particularly in multi-task scenarios. To address these challenges, we propose an efficient and accurate privacy-preserving task allocation scheme for UAV-MCS that leverages additive secret sharing to simultaneously protect the locations of both UAV operators and task requesters. To enable accurate allocation, we design a privacy-preserving comparison protocol (PAC) based on additive secret sharing and an optimized Paillier cryptosystem. Moreover, to support multi-task allocation under privacy constraints, we develop a privacy-preserving version of the Hungarian method. Experimental results on both real-world and synthetic datasets demonstrate that our scheme effectively reduces UAV travel distance while preserving location privacy, outperforming existing schemes in both efficiency and accuracy.

With the promising development of Internet of Things(IoT), many rich services configured in IoT devices exhibit high responsiveness to users' personalized requirements. However, recent studies have shown that distributed Denial of Service (DDoS) attacks pose a significant threat to IoT service via flooding the Internet Protocol (IP) and excessively consuming bandwidth, finally renders network services inaccessible. Machine learning-based approaches to mitigate DDoS attacks often depend on existing databases of DDoS incidents, yet the construction of such databases is non-trivial. Furthermore, solutions grounded in statistical analysis frequently encounter performance limitations. While the Economical Intelligent DDoS Demotivation (EID) algorithm, introduced in a study published at CCS'21, demonstrates effective defense against DDoS attacks by analyzing discrepancies in traffic patterns across different network nodes, it is marred by a high false positive rate, which can severely impact the quality of network services. To address these challenges, this paper introduces CRH4DDoS, a novel DDoS defense mechanism that employs conflict resolution strategies for the integration of heterogeneous data. Our proposed CRH4DDoS conceptualizes network traffic as heterogeneous data, framing the identification of malicious traffic as a conflict resolution problem. This approach does not necessitate a pre-existing DDoS database, achieves superior performance metrics, and markedly diminishes the false positive rate. Empirical assessments conducted on our real-world network platform reveal that the proposed CRH4DDoS is capable of defending against a variety of hybrid and dynamic DDoS attacks, occurring within seconds, with a false positive rate of merely 0.73%.