A5103224748- Matrix Theory and Algorithms
- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Advanced Numerical Methods in Computational Mathematics
- Autophagy in Disease and Therapy
- Electromagnetic Scattering and Analysis
- Molecular Sensors and Ion Detection
- Machine Learning and ELM
- Time Series Analysis and Forecasting
- Power Systems and Renewable Energy
- Tensor decomposition and applications
- Iron and Steelmaking Processes
- Neural Networks and Applications
- Metallurgical Processes and Thermodynamics
- Smart Grid and Power Systems
- Speech and Audio Processing
- Advanced Graph Neural Networks
- Photosynthetic Processes and Mechanisms
- Brain Tumor Detection and Classification
- Sulfur Compounds in Biology
- Advanced Image Fusion Techniques
- Blind Source Separation Techniques
- Nanoplatforms for cancer theranostics
- Retinal Imaging and Analysis
- Advanced Clustering Algorithms Research
Chinese Academy of Medical Sciences & Peking Union Medical College
2024-2025
South China Agricultural University
2025
Zhejiang University
2024
Harbin Institute of Petroleum
2024
Beijing University of Technology
2023
Soochow University
2023
University of Hong Kong
2019-2022
Northeastern University
2020-2021
Huawei Technologies (Sweden)
2021
Southwest Jiaotong University
2021
The influence of earthquake disasters on human social life is positively related to the magnitude and intensity earthquake, effectively avoiding casualties property losses can be attributed accurate prediction earthquakes. In this study, an electromagnetic sensor investigated assess earthquakes in advance by collecting signals. At present, mainstream comprises two methods. On one hand, most geophysicists or data analysis experts extract a series basic features from precursor signals for...
Object detection in remote sensing images has important applications various aspects. algorithms with deep convolutional neural networks (DCNNs) have made remarkable progress. However, when processing objects on vastly multiple scales high-resolution optical images, there is a high computational cost. Therefore, to simplify network multiscale training and inference, an automatic inference framework proposed balance the speed accuracy of object detection. We use attention mechanism that uses...
The LSTM network was proposed to overcome the difficulty in learning long-term dependence, and has made significant advancements applications. With its success drawbacks mind, this paper raises question - do RNN have long memory? We answer it partially by proving that not memory from a statistical perspective. A new definition for networks is further introduced, requires model weights decay at polynomial rate. To verify our theory, we convert into making minimal modification, their...
The cross temperature measuring device is applied to monitor the top and gas flow distribution in blast furnace (BF) during ironmaking process. However, of BF roof relatively high, which causes damage device. Therefore, this paper proposes an efficient estimation method. First all, data filtering method are used solve problems noise interference. Moreover, as fact that measurement affected by interference factors, it easy cause information redundancy. Aiming at this, maximum coefficient...
This paper introduces a concept of layer aggregation to describe how information from previous layers can be reused better extract features at the current layer. While DenseNet is typical example mechanism, its redundancy has been commonly criticized in literature. motivates us propose very light-weighted module, called recurrent (RLA), by making use sequential structure deep CNN. Our RLA module compatible with many mainstream CNNs, including ResNets, Xception and MobileNetV2, effectiveness...
Accurate detection for pulverized coal injection (PCI) blockage and coke size distribution are effective blast furnace (BF) operation. Nowadays, tuyere cameras have been applied in BF. However, the appearance of captured raceway images changes, which requests method to be adaptive. Therefore, an intelligent PCI is proposed. An adaptive pre-processing algorithm firstly developed improve image quality. Secondly, fitting circle used locate region improved U-net network investigated lance...
Deep anomaly detection has become popular for its capability of handling complex data. However, training a deep detector is fragile to data contamination due overfitting. In this work, we study the performance detectors under and construct data-efficient countermeasure against contamination. We show that induces an implicit kernel machine. then derive information-theoretic bound degradation with respect ratio. To mitigate degradation, propose contradicting approach. Apart from learning...
Many problems in science and engineering can be reduced to the recovery of an unknown large matrix from a small number random linear measurements. Matrix factorization arguably is most popular approach for low-rank recovery. methods have been proposed using different loss functions, example widely used L <sub xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sub> loss, more robust choices such as xmlns:xlink="http://www.w3.org/1999/xlink">1</sub>...
Autophagy plays a crucial role in maintaining the balance of normal physiological status living cells. Since viscosity lysosomes fluctuates during membrane fusion process autophagy, monitoring is an important means for real-time visualization autophagy process. Two electroneutral fluorescent probes 1a-1b exhibiting good response and aggregation-induced emission (AIE) properties were designed synthesized this paper. Using D-π-A strategy, acceptor quinoline/naphthyridine unit donor...
The neutral fluorescent probes 1a-1c based on D-π-A-π-D structure, with benzopyranoquinoline as the fluorophore and electron-donating groups at both ends, were developed, which emitted dual-channel fluorescence in different pH environments. Among them, probe 1a N,N-diethyl group morpholine ring exhibited best performance, showing a red shift of 85 nm dual emission 505 590 nm, Stokes 95 channels. displayed green specifically targeting mitochondria, while protonated form 1a+H+ resulting...
Autoregressive networks can achieve promising performance in many sequence modeling tasks with short-range dependence. However, when handling high-dimensional inputs and outputs, the massive amount of parameters network leads to expensive computational cost low learning efficiency. The problem be alleviated slightly by introducing one more narrow hidden layer network, but sample size required a certain training error is still substantial. To address this challenge, we rearrange weight...