A5101646256- Face and Expression Recognition
- Stochastic Gradient Optimization Techniques
- Recommender Systems and Techniques
- Domain Adaptation and Few-Shot Learning
- Electrospun Nanofibers in Biomedical Applications
- Sparse and Compressive Sensing Techniques
- Welding Techniques and Residual Stresses
- ZnO doping and properties
- Imbalanced Data Classification Techniques
- Machine Learning and Data Classification
- Adversarial Robustness in Machine Learning
- Advanced Optimization Algorithms Research
- Iron and Steelmaking Processes
- Risk and Safety Analysis
- Fault Detection and Control Systems
- Text and Document Classification Technologies
- Machine Learning and ELM
- Ga2O3 and related materials
- Advanced MIMO Systems Optimization
- Machine Learning and Algorithms
- Advanced Measurement and Metrology Techniques
- Risk and Portfolio Optimization
- Human Pose and Action Recognition
- Advanced Neural Network Applications
- Advanced Welding Techniques Analysis
Purdue University West Lafayette
2024
University of Chinese Academy of Sciences
2018-2024
Beihang University
2022-2024
Institute of Computing Technology
2024
Chinese Academy of Sciences
2018-2024
Tencent (China)
2024
Zhejiang University
2016-2024
First Affiliated Hospital Zhejiang University
2024
China Medical University
2024
Central South University
2008-2024
Native arteries contain a distinctive intima-media composed of organized elastin and an adventitia containing mature collagen fibrils. In contrast, implanted biodegradable small-diameter vascular grafts do not present spatially regenerated, elastin. The elastin-containing structures within the region encompass elastic lamellae (EL) internal lamina (IEL) are crucial for normal arterial function. Here, development novel electrospun graft that facilitates de novo formation structurally...
Data-free model stealing involves replicating the functionality of a target into substitute without accessing model's structure, parameters, or training data. Instead, adversary can only access predictions for generated samples. Once closely approximates behavior model, attackers exploit its white-box characteristics subsequent malicious activities, such as adversarial attacks. Existing methods within cooperative game frameworks often produce samples with high confidence prediction which...
Learning representation from relative similarity comparisons, often called ordinal embedding, gains rising attention in recent years. Most of the existing methods are batch designed mainly based on convex optimization, say, projected gradient descent method. However, they generally time-consuming due to that singular value decomposition (SVD) is commonly adopted during update, especially when data size very large. To overcome this challenge, we propose a stochastic algorithm SVRG-SBB, which...
As pairwise ranking becomes broadly employed for elections, sports competitions, recommendations, and so on, attackers have strong motivation incentives to manipulate the list. They could inject malicious comparisons into training data fool victim. Such a technique is called poisoning attack in regression classification tasks. In this paper, best of our knowledge, we initiate first systematic investigation attacks on algorithms, which can be formalized as dynamic static games between ranker...
Rank aggregation with pairwise comparisons is widely encountered in sociology, politics, economics, psychology, sports, etc. Given the enormous social impact and consequent incentives, potential adversary has a strong motivation to manipulate ranking list. However, ideal attack opportunity excessive adversarial capability cause existing methods be impractical. To fully explore risks, we leverage an online on vulnerable data collection process. Since it independent of rank lacks effective...
Rank aggregation with pairwise comparisons has shown promising results in elections, sports competitions, recommendations, and information retrieval. However, little attention been paid to the security issue of such algorithms, contrast numerous research work on computational statistical characteristics. Driven by huge profit, potential adversary strong motivation incentives manipulate ranking list. Meanwhile, intrinsic vulnerability rank methods is not well studied literature. To fully...
When facing rich multimedia content and making a decision, users tend to be overwhelmed with redundant options. Recommendation system can improve the users' experience by predicting possible preference of given user. The vast majority literature adopts collaborative framework, which relies on static fixed formulation rating score prediction function (in most cases an inner product function). However, such learning paradigm is not consistent dynamic feature human intelligence. Motivated this,...
A comparative study is given for the sensitivity of several typical Si nanophotonic waveguides, including SOI (silicon-on-insulator) nanowires, nanoslot suspended and nanofibers. The cases gas sensing (ncl ~ 1.0) liquid 1.33) are considered. When using nanowires (with a SiO2 buffer layer), (S 0.55) higher than that 0.35) due to lower asymmetry in vertical direction. By nanofibers, one could achieve compared with free-space beam = 1.0). index-contrast. waveguide an optimized nanowire as high...
The poor interfacial adhesion between silk fiber and polyester species remains a critical problem for the optimal mechanical performance of silk-reinforced composites. Here, we investigated in quantitative terms properties natural fibers polycaprolactone (PCL) at nano-, micro-, macroscales fabricated continuous silk-PCL composite filaments by melt extrusion drawing processing PCL 100, 120, 140 °C. Bombyx mori (Bm) silk, Antheraea pernyi (Ap) polyamide6 (PA6) were compared to with PCL. Ap...
Knowledge Distillation (KD) for object detection aims to train a compact detector by transferring knowledge from teacher model. Since the model perceives data in way different humans, existing KD methods only distill that is consistent with labels annotated human expert while neglecting not perception, which results insufficient distillation and sub-optimal performance. In this paper, we propose inconsistent (IKD), inherent model's counter-intuitive perceptions. We start considering...
Penetration estimation is a prerequisite of the automation backing welding based on vision sensing technology. However, arc interference in process leads to difficulties extracting weld pool characteristic information, which brings great challenges penetration estimation. At present, most researches focus extraction geometry parameters, and visual systems are complex structure complicated image processing algorithms. The research parameters still exploratory stage. purpose this paper...
Living in the era of internet, we are now facing with a big bang online information. As consequence, often find ourselves troubling hundreds and thousands options before making decision. way to improve quality users' experience, Recommendation System aims facilitate personalized decision processes via predicting responses toward different options. However, vast majority literature field merely focus on datasets sufficient amount samples. Different from traditional methods, propose novel...