检索结果-内蒙古大学图书馆

Uncertainty-Aware Graph Structure Learning

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Han, Shen Zhou, Zhiyao Chen, Jiawei Hao, Zhezheng Zhou, Sheng Wang, Gang Feng, Yan Chen, Chun Wang, Can The State Key Laboratory of Blockchain and Data Security Zhejiang University Hangzhou China Northwestern Polytechnical University Xi’an China Zhejiang University Hangzhou China Bangsun Technology Hangzhou China College of Computer Science and Technology Zhejiang University China Institute of Blockchain and Data Security China

Graph Neural Networks (GNNs) have become a prominent approach for learning from graph-structured data. However, their effectiveness can be significantly compromised when the graph structure is suboptimal. To address this issue, Graph Structure Learning (GSL) has emerged as a promising technique that refines node connections adaptively. Nevertheless, we identify two key limitations in existing GSL methods: 1) Most methods primarily focus on node similarity to construct relationships, while overlooking the quality of node information. Blindly connecting low-quality nodes and aggregating their ambiguous information can degrade the performance of other nodes. 2) The constructed graph structures are often constrained to be symmetric, which may limit the model’s flexibility and effectiveness. To overcome these limitations, we propose an Uncertainty-aware Graph Structure Learning (UnGSL) strategy. UnGSL estimates the uncertainty of node information and utilizes it to adjust the strength of directional connections, where the influence of nodes with high uncertainty is adaptively reduced. Importantly, UnGSL serves as a plug-in module that can be seamlessly integrated into existing GSL methods with minimal additional computational cost. In our experiments, we implement UnGSL into six representative GSL methods, demonstrating consistent performance improvements. The code is available at https://***/UnHans/UnGSL. Copyright © 2025, The Authors. All rights reserved.

关键词： Graph neural networks

blockchain-Empowered Asynchronous Federated Reinforcement Learning for IoT-Based Traffic Trajectory Prediction

学校读者我要写书评

暂无评论

IEEE Internet of Things Journal 2025年第11期12卷 17095-17109页

作者： Wang, Bin Tian, Zhao Tang, Fengxiao Pan, Heng She, Wei Liu, Wei Zhengzhou University School of Computer and Artificial Intelligence Zhengzhou450002 China Zhengzhou University School of Cyber Science and Engineering Zhengzhou450002 China Central South University School of Computer Science and Engineering Changsha410083 China Zhongyuan University of Technology School of Cyber Space Security Zhengzhou450007 China Zhengzhou University School of Cyber Science and Engineering Zhengzhou University The Zhengzhou Key Laboratory of Blockchain and Data Intelligence Zhengzhou450002 China

Vehicle trajectory prediction plays a crucial role in IoT-based intelligent transportation systems, which can effectively address key issues such as driving safety and multi-vehicle collaboration. However, the sensitivity of trajectory data and the reluctance of data holders to share it constrain the prediction model's ability to capture vehicle behavior patterns in different scenarios. To address the above problems, we propose a blockchain-enabled asynchronous federated proximal policy optimization framework (BE-AFPPO) for the trajectory prediction of self-driving vehicles. Firstly, we propose a Curiosity Proximal Policy Optimisation (C-PPO) algorithm. The method utilizes a driven exploration strategy to actively motivate the intelligent agent to explore the unknown state space. The avoidance policy model reaches a local optimum when processing trajectory data. In addition, we design historical GRU and future GRU as input layers. The target's historical motion features and future trajectory features are extracted, respectively. Then, various data is received through asynchronous federated learning. This model can fully learn the vehicle's behavior patterns in different scenarios, which improves prediction accuracy. Based on this, we develop a blockchain-based Dynamic Group Practical Byzantine Fault Tolerance (DG-PBFT) consensus algorithm. This enhances the credibility and integrity of the data while enriching the sources of trajectory data. Finally, we perform the experiments on the publicly available dataset nuScenes. The results demonstrate that the proposed method improves the robustness and accuracy of trajectory prediction. © 2014 IEEE.

关键词： Federated learning

Rankformer: A Graph Transformer for Recommendation based on Ranking Objective 34

学校读者我要写书评

暂无评论

Rankformer: A Graph Transformer for Recommendation based on ...

34th ACM Web Conference, WWW 2025

作者： Chen, Sirui Han, Shen Chen, Jiawei Hu, Binbin Zhou, Sheng Wang, Gang Feng, Yan Chen, Chun Wang, Can Zhejiang University Hangzhou China Ant Group Hangzhou China Bangsun Technology Hangzhou China State Key Laboratory of Blockchain and Data Security Zhejiang University China College of Computer Science and Technology Zhejiang University China Hangzhou High-Tech Zone (Binjiang) Institute of Blockchain and Data Security China

ISBN: (纸本)9798400712746

关键词： Personalized Ranking Recommendation Transformer

VibSpeech: exploring practical wideband eavesdropping via bandlimited signal of vibration-based side channel 24

学校读者我要写书评

暂无评论

VibSpeech: exploring practical wideband eavesdropping via ba...

Proceedings of the 33rd USENIX Conference on security Symposium

作者： Chao Wang Feng Lin Hao Yan Tong Wu Wenyao Xu Kui Ren State Key Laboratory of Blockchain and Data Security Zhejiang University and School of Cyber Science and Technology Zhejiang University University at Buffalo the State University of New York

ISBN: (纸本)9781939133441

Vibration-based side channel is an ever-present threat to speech privacy. However, due to the target's frequency response with a rapid decay or limited sampling rate of malicious sensors, the acquired vibration signals are often distorted and narrowband, which fails an intelligible speech recovery. This paper tries to answer that when the side-channel data has only a very limited bandwidth (<500Hz), is it feasible to achieve a wideband eavesdropping based on a practical assumption? Our answer is YES based on the assumption that a short utterance (2s-4s) of the victim is exposed to the attacker. What is most surprising is that the attack can recover speech with a bandwidth of up to 8kHz. This covers almost all phonemes (voiced and unvoiced) in human speech and causes practical threat. The core idea of the attack is using vocal-tract features extracted from the victim's utterance to compensate for the side-channel data. To demonstrate the threat, we proposed a vocal-guided attack scheme called VibSpeech and built a prototype based on a mmWave sensor to penetrate soundproof walls for vibration sensing. We solved challenges of vibration artifact suppression and a generalized scheme free of any target's training data. We evaluated VibSpeech with extensive experiments and validated it on the IMU-based method. The results indicated that VibSpeech can recover intelligible speech with an average MCD/SNR of 3.9/5.4dB.

关键词：

Exploiting Robust Model Watermarking Against the Model Fine-Tuning Attack via Flat Minima Aware Optimizers

学校读者我要写书评

暂无评论

Exploiting Robust Model Watermarking Against the Model Fine-...

International Conference on Acoustics, Speech, and Signal Processing (ICASSP)

作者： Dongdong Lin Yue Li Bin Li Jiwu Huang Guangdong Provincial Key Laboratory of Intelligent Information Processing Shenzhen University Shenzhen China Shenzhen Key Laboratory of Media Security Shenzhen University Shenzhen China SZU-AFS Joint Innovation Center for AI Technology Shenzhen University Shenzhen China Xiamen Key Laboratory of Data Security and Blockchain Technology Huaqiao University Xiamen China Guangdong Laboratory of Machine Perception and Intelligent Computing Shenzhen MSU-BIT University Shenzhen China

ISBN: (数字)9798350368741

ISBN: (纸本)9798350368758

With the rapid advancement of deep neural networks (DNNs), model watermarking has emerged as a widely adopted technique for safeguarding model copyrights. A prevalent method involves utilizing a watermark decoder to retrieve watermark bits from generated outputs, but such methods are often vulnerable to model fine-tuning attacks. Traditionally, this challenge is mitigated through adversarial training or data augmentation, both of which significantly increase the computational burden. In this paper, we present a solution employing Flat Minima Aware (FMA) optimizers to bolster the robustness of model watermarking without requiring additional training data. By optimizing the watermark loss with flat minima awareness, our approaches significantly enhance the robustness of watermarks against the model fine-tuning attack. Comprehensive experiments have demonstrated our method’s superior ability to preserve watermark integrity. These findings suggest that this innovative optimization strategy offers a robust and efficient pathway for protecting models, thereby contributing to more secure and reliable model copyright protection mechanisms.

关键词： Training Resistance Computational modeling Training data Watermarking Signal processing Diffusion models Robustness Speech processing Optimization

Optimizing Human Pose Estimation Through Focused Human and Joint Regions

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Jiao, Yingying Wang, Zhigang Liu, Zhenguang Fan, Shaojing Wu, Sifan Wu, Zheqi Xu, Zhuoyue College of Computer Science and Technology Jilin University China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University China College of Computer Science and Technology Zhejiang Gongshang University China The State Key Laboratory of Blockchain and Data Security Zhejiang University China Institute of Blockchain and Data Security China School of Computing National University of Singapore Singapore

Human pose estimation has given rise to a broad spectrum of novel and compelling applications, including action recognition, sports analysis, as well as surveillance. However, accurate video pose estimation remains an open challenge. One aspect that has been overlooked so far is that existing methods learn motion clues from all pixels rather than focusing on the target human body, making them easily misled and disrupted by unimportant information such as background changes or movements of other people. Additionally, while the current Transformer-based pose estimation methods has demonstrated impressive performance with global modeling, they struggle with local context perception and precise positional identification. In this paper, we try to tackle these challenges from three aspects: (1) We propose a bilayer Human-keypoint Mask module that performs coarse-to-fine visual token refinement, which gradually zooms in on the target human body and keypoints while masking out unimportant figure regions. (2) We further introduce a novel deformable cross attention mechanism and a bidirectional separation strategy to adaptively aggregate spatial and temporal motion clues from constrained surrounding contexts. (3) We mathematically formulate the deformable cross attention, constraining that the model focuses solely on the regions centered at the target person body. Empirically, our method achieves state-of-the-art performance on three large-scale benchmark datasets. A remarkable highlight is that our method achieves an 84.8 mean Average Precision (mAP) on the challenging wrist joint, which significantly outperforms the 81.5 mAP achieved by the current state-of-the-art method on the PoseTrack2017 dataset. Copyright © 2025, The Authors. All rights reserved.

关键词： Benchmarking

Safeguarding LLM Embeddings in End-Cloud Collaboration via Entropy-Driven Perturbation

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Jin, Shuaifan Pang, Xiaoyi Wang, Zhibo Wang, He Du, Jiacheng Hu, Jiahui Ren, Kui The State Key Laboratory of Blockchain and Data Security Zhejiang University China School of Cyber Science and Technology Zhejiang University China Hong Kong University of Science and Technology Hong Kong

Recent studies improve on-device language model (LM) inference through end-cloud collaboration, where the end device retrieves useful information from cloud databases to enhance local processing, known as Retrieval-Augmented Generation (RAG). Typically, to retrieve information from the cloud while safeguarding privacy, the end device transforms original data into embeddings with a local embedding model. However, the recently emerging Embedding Inversion Attacks (EIAs) can still recover the original data from text embeddings (e.g., training a recovery model to map embeddings back to original texts), posing a significant threat to user privacy. To address this risk, we propose EntroGuard, an entropy-driven perturbation-based embedding privacy protection method, which can protect the privacy of text embeddings while maintaining retrieval accuracy during the end-cloud collaboration. Specifically, to defeat various EIAs, we perturb the embeddings to increase the entropy of the recovered text in the common structure of recovery models, thus steering the embeddings toward meaningless texts rather than original sensitive texts during the recovery process. To maintain retrieval performance in the cloud, we constrain the perturbations within a bound, applying the strategy of reducing them where redundant and increasing them where sparse. Moreover, EntroGuard can be directly integrated into end devices without requiring any modifications to the embedding model. Extensive experimental results demonstrate that EntroGuard can reduce the risk of privacy leakage by up to 8 times at most with negligible loss of retrieval performance compared to existing privacy-preserving methods. Copyright © 2025, The Authors. All rights reserved.

关键词： Differential privacy

SpatioTemporal Learning for Human Pose Estimation in Sparsely-Labeled Videos

学校读者我要写书评

暂无评论

arXiv 2025年

作者： Jiao, Yingying Wang, Zhigang Wu, Sifan Fan, Shaojing Liu, Zhenguang Xu, Zhuoyue Wu, Zheqi College of Computer Science and Technology Jilin University China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University China College of Computer Science and Technology Zhejiang Gongshang University China School of Computing National University of Singapore Singapore The State Key Laboratory of Blockchain and Data Security Zhejiang University China Institute of Blockchain and Data Security China

Human pose estimation in videos remains a challenge, largely due to the reliance on extensive manual annotation of large datasets, which is expensive and labor-intensive. Furthermore, existing approaches often struggle to capture long-range temporal dependencies and overlook the complementary relationship between temporal pose heatmaps and visual features. To address these limitations, we introduce STDPose, a novel framework that enhances human pose estimation by learning spatiotemporal dynamics in sparsely-labeled videos. STDPose incorporates two key innovations: 1) A novel Dynamic-Aware Mask to capture long-range motion context, allowing for a nuanced understanding of pose changes. 2) A system for encoding and aggregating spatiotemporal representations and motion dynamics to effectively model spatiotemporal relationships, improving the accuracy and robustness of pose estimation. STDPose establishes a new performance benchmark for both video pose propagation (i.e., propagating pose annotations from labeled frames to unlabeled frames) and pose estimation tasks, across three large-scale evaluation datasets. Additionally, utilizing pseudo-labels generated by pose propagation, STDPose achieves competitive performance with only 26.7% labeled data. Copyright © 2025, The Authors. All rights reserved.

关键词： Large datasets

Towards Fair Federated Learning via Unbiased Feature Aggregation

学校读者我要写书评

暂无评论

IEEE Transactions on Dependable and Secure Computing 2025年

作者： He, Zeqing Wang, Zhibo Dong, Xiaowei Sun, Peng Ren, Ju Ren, Kui Zhejiang University State Key Laboratory of Blockchain and Data Security China Institute of Blockchain and Data Security Hangzhou310027 China Wuhan University School of Cyber Science and Engineering Wuhan430072 China Hunan University College of Computer Science and Electronic Engineering Changsha410082 China Tsinghua University Department of Computer Science and Technology Beijing100084 China Zhongguancun Laboratory Beijing100094 China

Federated learning (FL) is a distributed machine learning framework that enables multiple clients to collaboratively train models without raw data exchange. Prior studies on FL mainly focus on optimizing learning performance, enhancing privacy preservation, and improving attack resilience. However, little work studies how to mitigate the unfairness of federated trained models while unfair models would make discriminatory decisions toward certain groups or populations (e.g., favoring males over females), leading to serious ethical concerns. Thus, it is crucial to mitigate model unfairness in FL, yet challenging as this requires centralized access to each data point's fairness-sensitive information (e.g., race, gender), which is prohibited in FL. In this work, we propose a novel fair FL framework FedUFA, where the server can aggregate clients' learned knowledge in an unbiased manner, to obtain fair and high-usability federated trained models. Specifically, to unearth the bias in clients' local data and account for potentially heterogeneous local models, we propose a knowledge distillation-based FL scheme, where clients' knowledge of learned features on a public dataset is amalgamated to the server for aggregation. We train an unbiased feature mapper at the server to remove fairness-sensitive latent features and extract fair representations from clients' submitted raw features. In particular, we design an adversarial training method to train the mapper, which involves a predictor aiming to maximize the prediction accuracy on the FL task and a discriminator intending to help identify fairness-sensitive features. Extensive experiments on real-world datasets demonstrate the effectiveness of FedUFA. © 2025 IEEE. All rights reserved,

关键词： Federated learning