HWHan Wu

Han Wu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article

Previous Organizations / Institutes

The organizations / institutes from the scholar's other articles in the past ten years


Han Wu has a total of 20 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Revisit Location Privacy Protection: A Service Failure Perspective

    by Shaowei Wang, Han Wu, Yun Peng, Puning Zhao, Wenqi Ren, Changyu Dong, Jin Li, Jian Weng

  • Open Access

    Peeling the Layers of Privacy-Utility Onion on Tabular Data

    by Jiawei Wang, Zilong Zhao, Leonardo Aniello, Han Wu

    Machine Learning (ML) systems developed on sensitive data should be evaluated for unintended information leakage. Membership inference attacks (MIAs) serve as a standard auditing tool for quantifying unintended memorization. LiRA has emerged as the prevailing MIA methodology by evaluating true positive rates under strict false-positive constraints, thereby emphasizing worst-case leakage. Under LiRA, the privacy onion effect reveals that removing the most vulnerable samples and retraining can shift vulnerability to a different subset of the data. However, previous analyses focus primarily on image benchmarks, omit utility evaluation and do not systematically examine layer-by-layer removal. This paper introduces a structured evaluation protocol to analyze the privacy-utility onion effect in tabular machine learning models through iterative peeling. We evaluate the protocol across diverse tabular datasets and models. Our findings show that progressive layer-by-layer removal can yield stronger reductions in worst-case leakage than one-shot removal; also, they uncover consistent privacy-utility tradeoff, informing privacy auditing and defenses for tabular ML systems.

  • An Adaptive Mini-Batching Strategy for Reliable Streaming Data Delivery in Real-Time

    by Han Wu, Zhihao Shang, Huaming Wu, Katinka Wolter

    Modern applications show an increasing demand for continuously processing massive data streams in real time. Mini-batching technique is commonly used for transporting streaming data across these applications. However, although a larger mini-batch size increases throughput, it also raises the end-to-end latency and easily violate the latency constraint required by real-time applications. While existing work mostly studies this problem at the computation stage, this work explores it in the streaming data transportation stage. We show that selecting a proper mini-batch size is essential for efficient streaming data delivery. We further identify the shortcomings of existing latency measurements and introduce new Quality of Service (QoS) metrics: latency violation rate, timely throughput, message loss, and duplicate rate. To address the challenge of mini-batching under varying network conditions, we first develop prediction models for the proposed QoS metrics and then adaptively adjust the mini-batch size based on these predictions. In our experiments, the random forest regressor achieves an $R^{2}$ of 0.99 for performance metrics, and the multilayer perceptron achieves an MAE below 0.02 for reliability metrics. Using these predictions, the proposed adaptive strategy continuously updates the mini-batch size according to the observed network state. In a Kafka testbed with network packet loss rate approaching 25%, our strategy improves timely throughput by up to 30% compared with empirical mini-batch size selection. The results confirm the effectiveness of the adaptive mini-batching approach.

  • Impacts of Physical-Layer Information on Epidemic Spreading in Cyber-Physical Networked Systems

    by Xianglai Yuan, Yichao Yao, Han Wu, Minyu Feng

    Since Granell et al. proposed a multiplex network for information and epidemic propagation, researchers have explored how information propagation affects epidemic dynamics. However, the role of individuals acquiring information through physical interactions has received relatively less attention. In this work, we introduce a novel source of information: physical-layer information, and derive the epidemic outbreak threshold using the Microscopic Markov Chain Approach (MMCA). Our simulation results indicate that the outbreak threshold derived from the MMCA is consistent with the Monte Carlo (MC) simulation results, thereby confirming the accuracy of the theoretical model. Furthermore, we find that the physical-layer information effectively increases the population’s awareness density and the infection threshold βc, while reducing the population’s infection density, thereby suppressing the spreading of the epidemic. Another interesting finding is that when the density of 2-simplex information is relatively high, the 2-simplex plays a role similar to pairwise interaction, significantly enhancing the population’s awareness density and effectively preventing large-scale epidemic outbreaks. In addition, our model works equally well for cyber-physical systems with similar interaction mechanisms, while we simulate and validate it in a real grid system.


  • The Face of Deception: The Impact of AI-Generated Photos on Malicious Social Bots

    by Maxim Kolomeets, Han Wu, Lei Shi, Aad van Moorsel

    In this research, we investigate the influence of utilizing artificial intelligence (AI)-generated photographs on malicious bots that engage in disinformation, fraud, reputation manipulation, and other types of malicious activity on social networks. Our research aims to compare the performance metrics of social bots that employ AI photos with those that use other types of photographs. To accomplish this, we analyzed a dataset with 13 748 measurements of 11 423 bots from the VK social network and identified 73 cases where bots employed generative adversarial network (GAN)-photos and 84 cases where bots employed diffusion or transformers photos. We conducted a qualitative comparison of these bots using metrics such as price, survival rate, quality, speed, and human trust. Our study findings indicate that bots that use AI-photos exhibit less danger and lower levels of sophistication compared to other types: AI-enhanced bots are less expensive, less popular on exchange platforms, of inferior quality, less likely to be operated by humans, and, as a consequence, faster and more susceptible to being blocked by social networks. We also did not observe any significant difference between GAN-based and diffusion/transformers-based bots, indicating that diffusion/transformers models did not contribute to increased bot sophistication compared to GAN models. Our contributions include a proposed methodology for evaluating the impact of photos on bot sophistication, along with a publicly available dataset for other researchers to study and analyze bots. Our research findings suggest a contradiction to theoretical expectations: in practice, bots using AI-generated photos pose less danger.


  • Constrained Multiobjective Optimization for IoT-Enabled Computation Offloading in Collaborative Edge and Cloud Computing

    by Guang Peng, Huaming Wu, Han Wu, Katinka Wolter

    Internet-of-Things (IoT) applications are becoming more resource-hungry and latency-sensitive, which are severely constrained by limited resources of current mobile hardware. Mobile cloud computing (MCC) can provide abundant computation resources, while mobile-edge computing (MEC) aims to reduce the transmission latency by offloading complex tasks from IoT devices to nearby edge servers. It is still challenging to satisfy the quality of service with different constraints of IoT devices in a collaborative MCC and MEC environment. In this article, we propose three constrained multiobjective evolutionary algorithms (CMOEAs) for solving IoT-enabled computation offloading problems in collaborative edge and cloud computing networks. First of all, a constrained multiobjective computation offloading model considering time and energy consumption is established in the mobile environment. Inspired by the push and pull search framework, three CMOEAs are developed by combing the advantages of population-based search algorithms with flexible constraint handling mechanisms. On one hand, three popular and challenging constrained benchmark suites are selected to test the performance of the proposed algorithms by comparing them to the other seven state-of-the-art CMOEAs. On the other hand, a multiserver multiuser multitask computation offloading experimental scenario with a different number of IoT devices is used to evaluate the performance of three proposed algorithms and other compared algorithms as well as representative offloading schemes. The experimental results of the benchmark suites and computation offloading problems demonstrate the effectiveness and superiority of the proposed algorithms.


  • Dynamic Load Balancing in the Control Plane of Software-Defined Networks

    by Zhihao Shang, Han Wu, Guang Peng, Katinka Wolter

    This electronic Because of the limited capacity of a single controller, multiple controllers need to be deployed in a large software defined-network. The assignment of controllers and switches is a significant problem that may affect the flow setup time. Since the network traffic is dynamic, static controller assignment may lead to high flow setup time. To reduce the flow setup time and improve the performance of networks, we focus on the dynamic controller assignment problem in this paper. We propose a controller assignment scheme for multiple controllers. It adjusts the assignment among controllers and switches dynamically based on the load of switches and the capacity of controllers. We model each controller as an M/PH/1 queue and use a heuristic called LANS to optimize the controller assignment based on the model. We compare our method with DCP-SA in two cases. The results show that LANS can balance the load of controllers better than DCP-SA. The flow setup time under LANS is 8% and 17% lower than DCP-SA and LANS makes 0.8 and 1 fewer migrations on average than DCP-SA in the two cases.