A5022027609- Adversarial Robustness in Machine Learning
- Privacy-Preserving Technologies in Data
- Advanced Malware Detection Techniques
- Face recognition and analysis
- Ethics and Social Impacts of AI
- Biometric Identification and Security
- Heavy Metal Exposure and Toxicity
- Physical Unclonable Functions (PUFs) and Hardware Security
- High Entropy Alloys Studies
- Heavy Metals in Plants
- Semiconductor materials and devices
- Digital Media Forensic Detection
- Forensic and Genetic Research
- Security and Verification in Computing
- High-Temperature Coating Behaviors
- Generative Adversarial Networks and Image Synthesis
- Heavy metals in environment
- Advanced Steganography and Watermarking Techniques
- Network Security and Intrusion Detection
University of Virginia
2023
ZheJiang Institute For Food and Drug Control
2022
Fudan University
2018-2022
Adversarial examples revealed the weakness of machine learning techniques in terms robustness, which moreover inspired adversaries to make use attack systems employing learning. Existing researches covered methodologies adversarial example generation, root reason existence examples, and some defense schemes. However practical against real world did not appear until recent, mainly because difficulty injecting a artificially generated into model behind hosting system without breaking...
PCIe (Peripheral Component Interconnect express) protocol is the de facto to bridge CPU and peripheral devices like GPU, NIC, SSD drive. There an increasing demand install more on a single machine, but interfaces offered by Intel CPUs are fixed. To resolve such contention, switch, PCH (Platform Controller Hub), or virtualization cards installed machine allow multiple share interface. Congestion happens when collective traffic from overwhelm link capacity, transmission delay then...
In recent years, the security issues of artificial intelligence have become increasingly prominent due to rapid development deep learning research and applications. Backdoor attack is an targeting vulnerability models, where hidden backdoors are activated by triggers embedded attacker, thereby outputting malicious predictions that may not align with intended output for a given input. this work, we propose novel black-box backdoor based on machine unlearning. The attacker first augments...
The security of the Autonomous Driving (AD) system has been gaining researchers' and public's attention recently. Given that AD companies have invested a huge amount resources in developing their models, e.g., localization these especially parameters, are important intellectual property deserve strong protection.
Recently, diffusion models (DMs) have become the state-of-the-art method for image synthesis. Editing based on DMs, known their high fidelity and precision, inadvertently introduced new challenges related to copyright infringement malicious editing. Our work is first formalize address this issue. After assessing attempting enhance traditional watermarking techniques, we recognize limitations in emerging context. In response, develop a novel technique, RIW (Robust Invisible Watermarking),...
Face verification system (FVS), which can automatically verify a person's identity, has been increasingly deployed in the real-world settings. Key to its success is inclusion of face embedding, technique that detect similar photos same person by deep neural networks.
Machine Learning (ML) already has been integrated into all kinds of systems, helping developers to solve problems with even higher accuracy than human beings. However, when integrating ML models a system, may accidentally take not enough care the outputs models, mainly because their unfamiliarity and AI, resulting in severe consequences like hurting data owners' privacy. In this work, we focus on understanding risks abusing embeddings an important popular way using ML. To show consequence,...
Recent advancements in deep learning have spotlighted a crucial privacy vulnerability to membership inference attack (MIA), where adversaries can determine if specific data was present training set, thus potentially revealing sensitive information. In this paper, we introduce technique, weighted smoothing (WS), mitigate MIA risks. Our approach is anchored on the observation that samples differ their MIA, primarily based distance clusters of similar samples. The intuition will make model...
Due to the unique characteristics, high-entropy alloys (HEAs) have potential applications as antioxidant alloy materials or coatings in high-temperature environments. Al, Cr, Ti, and Si had strong oxidation resistance were used matrix elements form HEAs, effects of doping rare earth (RE) lanthanum (La) yttrium (Y) explored this study. AlCrTiSi 0.2 RE 0.02 HEAs prepared by vacuum arc-melting mass gain experiments carried out at 1000 °C. The was only 1.195 mg/cm 2 after 53 h oxidation, best...