A5057867117- Adversarial Robustness in Machine Learning
- Anomaly Detection Techniques and Applications
- Advanced Neural Network Applications
- Advanced Malware Detection Techniques
- Bacillus and Francisella bacterial research
- Digital Media Forensic Detection
- Domain Adaptation and Few-Shot Learning
- Advanced Image and Video Retrieval Techniques
- Video Surveillance and Tracking Methods
- Privacy-Preserving Technologies in Data
- Speech and Audio Processing
- Speech Recognition and Synthesis
- Advanced Image Fusion Techniques
- Higher Education and Teaching Methods
- Topic Modeling
- Software-Defined Networks and 5G
- Advanced Computational Techniques and Applications
- Security and Verification in Computing
- Complex Network Analysis Techniques
- Nuclear Materials and Properties
- Mobile Crowdsensing and Crowdsourcing
- Network Traffic and Congestion Control
- Model Reduction and Neural Networks
- Visual Attention and Saliency Detection
- Wireless Signal Modulation Classification
Zhejiang University of Science and Technology
2010-2024
Zhejiang University of Technology
2022
Zhejiang University
2014
Zhejiang University of Water Resource and Electric Power
2013
Zhejiang Water Conservancy and Hydropower Survey and Design Institute
2010
RGB–thermal scene parsing has recently attracted increasing research interest in the field of computer vision. However, most existing methods fail to perform good boundary extraction for prediction maps and cannot fully use high-level features. In addition, these simply fuse features from RGB thermal modalities but are unable obtain comprehensive fused To address problems, we propose an edge-aware guidance fusion network (EGFNet) parsing. First, introduce a prior edge map generated using...
Abstract Deep Neural Networks (DNNs) have demonstrated outstanding performance in various medical image processing tasks. However, recent studies revealed a heightened vulnerability of DNNs to adversarial attacks compared their natural counterparts. In this work, we present novel perspective by analyzing the disparities between datasets and datasets, specifically focusing on dataset collection process. Our analysis uncovers unique differences data distribution across different classes...
Dynamic link prediction aims to predict future connections among unconnected nodes in a network. It can be applied for friend recommendations, completion, and other tasks. Network representation learning algorithms have demonstrated considerable effectiveness various However, most network are based on homogeneous networks static that do not consider rich semantic dynamic information. Additionally, existing methods neglect the neighborhood interaction structure of node. In this work, we...
Edge intelligence has played an important role in constructing smart cities, but the vulnerability of edge nodes to adversarial attacks becomes urgent problem. A so-called example can fool a deep learning model on node for misclassification. Due transferability property examples, adversary easily black-box by local substitute model. general have limited resources, which cannot afford complicated defense mechanism like that cloud data center. To address challenge, we propose dynamic...
In this paper, the influence of data imbalance on neural networks is discussed, and an improved learning algorithm to solve problem proposed. The experimental results show that in case imbalanced data, training error network converges slowly generalization ability poor. Our theoretical analysis shows process training, gradient descent direction weights dominated by major-classes, which accounts for slow convergence error. Based these results, we propose Equilibration Mini-batch Stochastic...
Convolutional neural networks (CNNs) have been successfully applied to various fields. However, CNNs' overparameterization requires more memory and training time, making it unsuitable for some resource-constrained devices. To address this issue, filter pruning as one of the most efficient ways was proposed. In article, we propose a feature-discrimination-based importance criterion, uniform response criterion (URC), key component pruning. It converts maximum activation responses into...
Nowadays, deep learning models play an important role in a variety of scenarios, such as image classification, natural language processing, and speech recognition. However, are shown to be vulnerable; small change the original data may affect output model, which incur severe consequences misrecognition privacy leakage. The intentionally modified is referred adversarial examples. In this paper, we explore security vulnerabilities designed for textual analysis. Specifically, propose visual...
Recent work shows that well-designed adversarial examples can fool deep neural networks (DNNs). Due to their transferability, also attack target models without extra information, called black-box attacks. However, most existing ensemble attacks depend on numerous substitute cover the vulnerable subspace of a model. In this work, we find three types with non-overlapping frequency regions, which large enough subspace. Based finding, propose lightweight named LEA <sup...
RGB thermal scene parsing has recently attracted increasing research interest in the field of computer vision. However, most existing methods fail to perform good boundary extraction for prediction maps and cannot fully use high level features. In addition, these simply fuse features from modalities but are unable obtain comprehensive fused To address problems, we propose an edge-aware guidance fusion network (EGFNet) parsing. First, introduce a prior edge map generated using images capture...
Java programming is an important and basic specialized course for students of computer specialty in higher vocational colleges. In addition, object orientation currently widespread method. this paper, basing on the characteristic our practical teaching experiences, we proposed many measures how to improve quality.
With the boom of edge intelligence, its vulnerability to adversarial attacks becomes an urgent problem. The so-called example can fool a deep learning model on node misclassify. Due property transferability, adversary easily make black-box attack using local substitute model. Nevertheless, limitation resource nodes cannot afford complicated defense mechanism as doing cloud data center. To overcome challenge, we propose dynamic mechanism, namely EI-MTD. It first obtains robust member models...