YHYun Hu

Yun Hu

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

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Yun Hu has a total of 27 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Open Access

    Euclidean Space Alignment Net: A Lightweight Model-Agnostic Framework for End-to-End Cross-Distribution EEG Alignment

    by Zhirui Chen, Jing Luo, Qi Mao, Xinhua Cai, Yun Hu, Wenyao Yan, Yunxiao Liu, Xiaofan Wang, Junhuai Li, Xinhong Hei

    The decoding accuracy of electroencephalography (EEG) signals is crucial to the performance of brain-computer interface (BCI) systems, yet it is severely bottlenecked by pervasive cross-distribution discrepancies in EEG signals. These discrepancies stem from two key sources—multi-subject/cross-session variability and intra-subject task-induced variability both eroding the generalization and robustness of decoding models. Although Euclidean-space alignment approaches offer an interpretable solution for reconciling such shifts, their integration into deep learning frameworks is constrained by two critical limitations: the lack of end-to-end trainability and adaptive adjustment mechanisms tailored to dynamic EEG characteristics. To address these gaps, this paper proposes a lightweight, model-agnostic Euclidean Space Alignment Net (ESAN) for cross-distribution EEG signal alignment, enabling task adaptability via flexible, end-to-end distribution matching. ESAN can be seamlessly embedded into mainstream EEG decoding backbones to optimize MI recognition and brainprint identification on the same dataset: it mitigates multi-subject EEG feature distribution shifts for MI recognition and intra-subject task-induced shifts for brainprint identification. Extensive experiments on three public MI-EEG datasets demonstrate that ESAN consistently improves the performance of diverse deep learning backbones: for multi-subject MI classification, the highest accuracy gains reach 4.01\%, 2.62\%, and 2.93\% on BCI IV-2a, BCI IV-2b, and High-Gamma, respectively; for cross-task brainprint identification, the highest accuracy improvements reach 10.28\%, 5.51\%, and 1.88\% on the three datasets, respectively, while also consistently outperforming representative alignment methods. This work enables efficient exploitation of cross-distribution EEG samples and advances the robustness and practical deployment of EEG-based BCI systems.

  • Open Access

    CAEBA:A Dynamic Hidden Backdoor Attack Framework in Federated Learning

    by Xiaojun Guo, Guoliang Li, Yun Hu

    Federated Learning (FL) is a privacy-preserving distributed learning framework, but its distributed data collection and client-side training pipeline expose the global model to backdoor attacks. Existing FL backdoor attacks often depend on fixed triggers or on update manipulations that are easier to isolate under robust aggregation and model-inspection defenses. We propose CAEBA (Conditional AutoEncoder Backdoor Attack), a dynamic hidden backdoor framework that uses a conditional autoencoder to generate target-aware and visually stealthy triggers while progressively implanting the backdoor through federated optimization. CAEBA separates the clean classifier from the trigger generator, formulates the attack as a dual-constrained objective, and updates the generator at a lower frequency than the classifier to stabilize local optimization. We evaluate CAEBA on MNIST, FashionMNIST, CIFAR-10, CIFAR-100, and Tiny-ImageNet under representative aggregation rules and defenses, including FL-Detector, RFLBAT, DeepSight, FoolsGold, and FLAME. The results show that CAEBA preserves main-task accuracy while maintaining persistent backdoor effectiveness.

  • Open Access

    Two-Stage Cross-Lingual Supervised Transfer for Low-Resource Tibetan Rumor Detection

    by Yongxin Ma, Haoqiang Du, Shuo Zhang, Jing He, Yun Hu, Peng Li

    Detecting rumors in low-resource languages such as Tibetan is challenging due to scarce labeled data and noisy social media writing. We propose a simple two-stage cross-lingual supervised transfer framework built on CINO, a multilingual pretrained model covering Chinese and several Chinese minority languages. In Stage S (source-supervised), we train on Weibo21 (9,128 Chinese rumor/non-rumor posts) to learn transferable rumor-discriminative representations. In Stage T (target-supervised), we fine-tune the resulting encoder on 7,997 labeled Tibetan short texts. Under five-fold stratified cross-validation, two-stage transfer improves Macro-F1 from 89.65±0.77 (direct Tibetan fine-tuning) to 90.77±0.79 (+1.12), outperforming XLM-R (67.52±1.66), mBERT (83.69±0.68), and TiBERT (85.76±1.19). We further examine whether inserting an intermediate masked language modeling (MLM) adaptation step (DAPT/TAPT) between the two supervised stages yields additional gains. Neither intermediate adaptation improves over the two-stage baseline, suggesting objective mismatch and representation drift between token-level MLM and sequence-level rumor classification. Our findings indicate that two-stage supervised transfer is a cost-effective recipe for low-resource Tibetan rumor detection.

  • A Novel Privacy-Enhanced Fine-Grained Access Control Scheme

    by Jiaju Liu, Yun Hu, Xiaojun Guo

    Access control technology ensures that information resources are not illegally used and accessed by restricting the ability and scope of users access to object resources. While, most access control models only focus on the optimization of access control policy, less attention is paid to the privacy of object resources. That is, authorized users have unlimited access to acquired object data. And it greatly increases the risk of privacy data being leaked. Therefore, our paper proposes a novel privacy-enhanced fine-grained access control scheme. The scheme introduces Differential Privacy (DP) into the role to resource allocation step of the Role-Based Access Control (RBAC) model. And the scheme realizes the privacy protection of the object data set and the fine-grained permissions control by assigning privacy budgets to different levels of roles. It enables users with different roles to obtain authorization object data sets with different accuracy. Finally, simulation results and performance analysis show that the scheme has high applicability of access control policy, strong privacy and high accuracy of object data set.