A5068514844- Domain Adaptation and Few-Shot Learning
- Multimodal Machine Learning Applications
- Embedded Systems and FPGA Design
- AI and Big Data Applications
- Stroke Rehabilitation and Recovery
- Advanced Computing and Algorithms
- Embedded Systems and FPGA Applications
- Anomaly Detection Techniques and Applications
- Embedded Systems Design Techniques
- Smart Grid and Power Systems
- Advanced Computational Techniques and Applications
- Data Stream Mining Techniques
- Real-time simulation and control systems
- Technology and Human Factors in Education and Health
- Fire Detection and Safety Systems
- Time Series Analysis and Forecasting
- Text and Document Classification Technologies
- Distributed and Parallel Computing Systems
- Constraint Satisfaction and Optimization
- Cancer-related molecular mechanisms research
- Advanced Graph Theory Research
- Advanced Image and Video Retrieval Techniques
- Formal Methods in Verification
- Advanced Sensor and Control Systems
Wuhan University
2023-2024
Chang'an University
2024
Henan Polytechnic University
2010-2023
Harbin University of Science and Technology
2009
Abstract The data stream is a dynamic collection of that changes over time, and predicting the class can be challenging due to sparse samples, complex interdependent characteristics between data, random fluctuations. Accurately in create challenges. Due its incremental learning nature, neural networks suitable approach for streaming visualization. However, high computational cost limits their applicability high-speed streams, which has not yet been fully explored existing approaches. To...
Machine learning applied to fire alarm systems is an increasingly common optimization problem. In this paper, a method based on comprehensive evaluation and machine proposed. Firstly, relying the literature data sets, 10 indicators such as number of fires, failure rate, accuracy etc. are extracted comprehensively evaluate performance each type component, entropy weight method, information TOPSIS used derive level rating intelligent research judgment system constructed. Then, BP neural...
Adapter-based parameter-efficient transfer learning has achieved exciting results in vision-language models. Traditional adapter methods often require training or fine-tuning, facing challenges such as insufficient samples resource limitations. While some overcome the need for by leveraging image modality cache and retrieval, they overlook text modality's importance cross-modal cues efficient adaptation of parameters visual-language This work introduces a approach named XMAdapter. XMAdapter...
Domain generalization faces challenges due to the distribution shift between training and testing sets, presence of unseen target domains. Common solutions include domain alignment, meta-learning, data augmentation, or ensemble learning, all which rely on labels adversarial techniques. In this paper, we propose a Dual-Stream Separation Reconstruction Network, dubbed DSDRNet. It is disentanglement-reconstruction approach that integrates features both inter-instance intra-instance through...
Data augmentation has undoubtedly enabled a significant leap forward in training high-accuracy deep network. Besides the commonly used to target data, e.g., random cropping, flipping, and rotation, recent works have been dedicated mining generalized knowledge by using multiple sources. However, along with plentiful data comes huge distribution gap between different sources (hybrid shift). To mitigate this problem, existing methods tend manually annotate more data. Unlike previous methods,...
Abstract Domain Generalization (DG) aims to transfer knowledge learned from multiple source domains unseen domains. One of the primary challenges hinders DG is insufficient diversity domains, which hampers model's ability learn generalize. Traditional data augmentation methods, fuse content, style, labels, etc., unable effectively global features In this paper, we present an innovative approach domain generalization learning technique, called PatchMix, by stitching patches different together...
Backdoor sets of SAT problem can quickly decide the satisfiability real-world instances, and QBF is generalization problem, so backdoor are crucial to its solution. We propose a new algorithm computing QHorn deletion in this paper, which contains two stages. Firstly, we compute renamed formula according largest renamable R <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">max</sub> matrix formula, here only existent variables renamed. Then RQHorn...
With the scale of equipment needed to be managed in heat metering monitoring system becoming larger, how realize simple management and ensure scalability is a problem that needs studied. Based on comprehensive application information communication technology, this paper studies expounds intelligent based cloud platform, analyses plans platform metering. It convenient for heating companies monitor status site, find problems time deal with them. realizes user addition facilitates business...