A5041918034- Privacy-Preserving Technologies in Data
- Adversarial Robustness in Machine Learning
- Video Coding and Compression Technologies
- Tensor decomposition and applications
- Advanced Data Compression Techniques
- Advanced Neural Network Applications
- Stochastic Gradient Optimization Techniques
- Cryptography and Data Security
- Topic Modeling
- Image and Video Quality Assessment
- Sparse and Compressive Sensing Techniques
- Anomaly Detection Techniques and Applications
- Mobile Crowdsensing and Crowdsourcing
- Advanced Vision and Imaging
- Internet Traffic Analysis and Secure E-voting
- Game Theory and Applications
- Auction Theory and Applications
- Game Theory and Voting Systems
- Advanced Malware Detection Techniques
- Machine Learning in Healthcare
- Face and Expression Recognition
- Domain Adaptation and Few-Shot Learning
- Natural Language Processing Techniques
- Explainable Artificial Intelligence (XAI)
- Privacy, Security, and Data Protection
Sun Yat-sen University
2025
Shanghai Jiao Tong University
2023-2024
Zhejiang University
2002-2024
Xidian University
2021-2024
Zhejiang University of Science and Technology
2023-2024
Nanjing University of Aeronautics and Astronautics
2022-2023
Northeastern University
2009-2023
Guangzhou University
2023
Wenzhou Medical University
2022
China Mobile (China)
2022
Prompts have significantly improved the performance of pre-trained Large Language Models (LLMs) on various downstream tasks recently, making them increasingly indispensable for a diverse range LLM application scenarios. However, backdoor vulnerability, serious security threat that can maliciously alter victim model's normal predictions, has not been sufficiently explored prompt-based LLMs. In this paper, we present PoisonPrompt, novel attack capable successfully compromising both hard and...
Data-driven machine learning has become ubiquitous. A marketplace for models connects data owners and model buyers, can dramatically facilitate data-driven applications. In this paper, we take a formal perspective propose the first en<u> D </u>-to-end mod <u>e</u> l m <u>a</u> rketp <u>l</u> ace with diff rential p <u>r</u> ivacy ( Dealer ) towards answering following questions: How to formulate owners' compensation functions...
Tensor factorization has been demonstrated as an efficient approach for computational phenotyping, where massive electronic health records (EHRs) are converted to concise and meaningful clinical concepts. While distributing the tensor tasks local sites can avoid direct data sharing, it still requires exchange of intermediary results which could reveal sensitive patient information. Therefore, challenge is how jointly decompose under rigorous principled privacy constraints, while support...
Federated Learning (FL) is a promising framework for multiple clients to learn joint model without directly sharing the data. In addition high utility of model, rigorous privacy protection data and communication efficiency are important design goals. Many existing efforts achieve by ensuring differential intermediate parameters, however, they assume uniform parameter all clients. practice, different may have requirements due varying policies or preferences. this paper, we focus on explicitly...
Large Language Models (LLMs) have showcased remarkable capabilities across various domains. Accompanying the evolving and expanding deployment scenarios of LLMs, their challenges escalate due to sheer scale advanced yet complex activation designs prevalent in notable model series, such as Llama, Gemma, Mistral. These become particularly pronounced resource-constrained scenarios, where mitigating inference efficiency bottlenecks is imperative. Among recent efforts, approximation has emerged a...
Safety alignment is critical in pre-training large language models (LLMs) to generate responses aligned with human values and refuse harmful queries. Unlike LLM, the current safety of VLMs often achieved post-hoc fine-tuning. However, these methods are less effective white-box attacks. To address this, we propose $\textit{Adversary-aware DPO (ADPO)}$, a novel training framework that explicitly considers adversarial. (ADPO)}$ integrates adversarial into enhance under worst-case perturbations....
Current Stackelberg security game models primarily focus on isolated systems in which only one defender is present, despite being part of a more complex system with multiple players. However, many real such as transportation networks and the power grid exhibit interdependencies among targets and, consequently, between decision makers jointly charged protecting them. To understand multidefender strategic interactions present scenarios, authors investigate games defenders. Unlike most prior...
Modern healthcare systems knitted by a web of entities (e.g., hospitals, clinics, pharmacy companies) are collecting huge volume data from large number individuals with various medical procedures, medications, diagnosis, and lab tests. To extract meaningful concepts (i.e., phenotypes) such higher-arity relational data, tensor factorization has been proven to be an effective approach received increasing research attention, due their intrinsic capability represent the high-dimensional data....
Tensor singular value decomposition (t-SVD) has recently become increasingly popular for tensor recovery under partial and/or corrupted observations. However, the existing t -SVD-based methods neither make use of a rank prior nor provide an accurate estimation (RE), which would limit their performance. From practical perspective, RE problem is nontrivial and difficult to solve. In this article, we, therefore, aim determine correct intrinsic low-rank from observations based on t-SVD further...
The robustness and security of natural language processing (NLP) models are significantly important in real-world applications. In the context text classification tasks, adversarial examples can be designed by substituting words with synonyms under certain semantic syntactic constraints, such that a well-trained model will give wrong prediction. Therefore, it is crucial to develop techniques provide rigorous provable guarantee against attacks. this paper, we propose WordDP achieve certified...
Machine unlearning is an emerging task of removing the influence selected training datapoints from a trained model upon data deletion requests, which echoes widely enforced regulations mandating Right to be Forgotten. Many methods have been proposed recently, achieving significant efficiency gains over naive baseline retraining scratch. However, existing focus exclusively on standard models and do not apply adversarial (ATMs) despite their popularity as effective defenses against examples....
Top-k frequent items detection is a fundamental task in data stream mining. Many promising solutions are proposed to improve memory efficiency while still maintaining high accuracy for detecting the items. Despite concern, users could suffer from privacy loss if participating without proper protection, since their contributed local streams may continually leak sensitive individual information. However, most existing works solely focus on addressing either memory-efficiency problem or...
Over the past years, Machine Learning-as-a-Service (MLaaS) has received a surging demand for supporting Learning-driven services to offer revolutionized user experience across diverse application areas. MLaaS provides inference service with low latency based on an ML model trained using dataset collected from numerous individual data owners. Recently, sake of owners' privacy and comply "right be forgotten (RTBF)" as enacted by protection legislation, many machine unlearning methods have been...
Spear-phishing attacks pose a serious threat to sensitive computer systems, since they sidestep technical security mechanisms by exploiting the carelessness of authorized users. A common way mitigate such is use e-mail filters which block e-mails with maliciousness score above chosen threshold. Optimal choice threshold involves tradeoff between risk from delivered malicious emails and cost blocking benign traffic. further complicating factor strategic nature an attacker, who may selectively...
Single sample per person face recognition (SSPP FR), i.e., identifying a (i.e., data subject) with single image only for training, has several attractive potential applications, but it is still challenging problem. Existing generic learning methods usually leverage prototype plus variation (P+V) model SSPP FR provided that samples in the biometric enrolment database are variation-free and thus can be treated as prototypes of subjects. However, this condition not satisfied when these...
ABSTRACT BACKGROUND The COVID-19 epidemic, first emerged in Wuhan during December 2019, has spread globally. While the mass population movement for Chinese New Year significantly influenced spreading disease, little direct evidence exists about relevance to epidemic and its control of from Wuhan, local emergency response, medical resources China. METHODS Spearman’s correlation analysis was performed between official data confirmed cases Jan 20 th Feb 19 , 2020 real-time travel health data....
As an important perceptual characteristic of the Human Visual System (HVS), Just Noticeable Difference (JND) has been studied for decades with image and video processing (e.g., visual signal compression). However, there is little exploration on existence JND Deep Machine Vision (DMV), although DMV made great strides in many machine vision tasks. In this paper, we take initial attempt, demonstrate that JND, termed as DMV-JND. We then propose a model classification task DMV. It discovered can...
Federated learning is a prominent framework that enables clients (e.g., mobile devices or organizations) to collaboratively train global model under central server's orchestration while keeping local data private. However, the aggregation step in federated vulnerable adversarial attacks as server cannot enforce clients' behavior. As result, performance of and convergence training process can be affected such attacks. To mitigate this vulnerability, existing works have proposed robust methods...
Recently, based on a new tensor algebraic framework for third-order tensors, the singular value decomposition (t-SVD) and its associated tubal rank definition have shed light low-rank modeling. Its applications to robust image/video recovery background modeling show promising performance due superior capability in cross-channel/frame information. Under t-SVD framework, we propose norm called spectral k-support (TSP-k) by an alternative convex relaxation. As interpolation between existing...
Rationale: Coronavirus disease 2019 (COVID-19) has caused a global pandemic.A classifier combining chest X-ray (CXR) with clinical features may serve as rapid screening approach.Methods: The study included 512 patients COVID-19 and 106 influenza A/B pneumonia.A deep neural network (DNN) was applied, derived from CXR findings formed fused for diagnosis prediction. Results:The of showed different patterns.Patients experienced less fever, more diarrhea, salient hypercoagulability.Classifiers...