- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Syntax, Semantics, Linguistic Variation
- Advanced Adaptive Filtering Techniques
- Image Processing Techniques and Applications
- Advanced Algorithms and Applications
- Language, Discourse, Communication Strategies
- Natural Language Processing Techniques
- Advanced Sensor and Control Systems
- Blind Source Separation Techniques
- Neural Networks and Applications
- Advanced Image and Video Retrieval Techniques
- Image and Signal Denoising Methods
- Advanced Image Fusion Techniques
- Photonic and Optical Devices
- Adversarial Robustness in Machine Learning
- Advanced Wireless Communication Technologies
- Industrial Vision Systems and Defect Detection
- Advanced Image Processing Techniques
- Image and Object Detection Techniques
- Optical Wireless Communication Technologies
- Topic Modeling
- Linguistic Variation and Morphology
- Optical Coatings and Gratings
- Photonic Crystals and Applications
Harbin University of Science and Technology
2011-2025
Shanxi Agricultural University
2025
Hefei University of Technology
2022-2023
China Southern Power Grid (China)
2023
Peking University
2013-2021
Optica
2013-2017
Princeton University
1991-2011
Boeing (Australia)
2002-2010
The Ohio State University
1998
University of Victoria
1990-1992
Stochastic gradient adaptive filtering algorithms using variable step sizes are investigated. The variable-step-size algorithm improves the convergence rate while sacrificing little in steady-state error. Expressions describing of mean and mean-squared values coefficients developed used to calculate mean-square-error evolution. initial error also performance is studied when a power-of-two quantizer used, finite-word-length effects considered. analytical results verified with simulations...
With the rapid development of semiconductor industry, demand for high-speed testing in large-scale production devices and integrated circuit lines continues to grow. As one key tools device performance testing, source measure unit (SMU) plays a crucial role high-precision transient response scenarios. In measurement scenarios, multiple measurements are often required averaged improve accuracy, but this can slow down speed. This article proposes acceleration algorithm based on BA-Informer...
Aiming at the problems of rapid-expanding random trees (RRT) in path planning, such as strong search blindness, high randomness, slow convergence, and non-smooth generated paths, this paper proposes a Multi-Strategy Fusion RRT (MSF-RRT) algorithm to improve RRT. Firstly, target bias strategy introduces higher probability that region samples points; secondly, expansion expands sampling points an orderly manner; then, adaptive step size adjusts according map complexity. Finally, preliminary...
The stochastic gradient algorithm using a simplified arithmetic is analyzed in this paper. A power-of-two quantizer used for the input of multiplier to reduce multiplication at most simple shift. In spite its implementation, performance shown be comparable classical LMS algorithm. linearized approximation first derived, followed by analysis an exact nonlinear model. derivation based on Gaussian assumption, and effects removing assumption are later considered. roundoff error due finite-bit...
In this letter, the outage performance of a satellite-assisted cooperative non-orthogonal multiple access (NOMA) system is studied, where terrestrial source communicates with one destination directly while other assisted by regenerative satellite. We consider satellite channels and undergo shadowed-Rician fading Nakagami-m fading, respectively. Based on this, we derive exact closed-form expressions for probabilities. addition, asymptotic at high signal-to-noise ratio are provided to gain...
This study investigates a problem of resource allocation for an uplink multi‐carrier non‐orthogonal multiple access (NOMA) system. To provide general problem, the order successive interference cancellation (SIC), subcarrier assignment and power are all considered as optimisation variables interest, with objective maximising weighted sum rate under transmit constraint each user equipment (UE). Firstly, novel optimal SIC is proposed based on inherent structure NOMA Then, order, Lagrange dual...
This paper according to the traditional PID control system of nonlinear time-varying, is not easy realize on-line adjustment flaw and proposed one kind based on FPGA (Field Programmable Gate Array), implementation fuzzy algorithm its application brushless DC motor speed system. Using VHDL (Very High Speed Integrated Circuit HardwareDescription Language) language Quartus II built-in macro module modules are constructed. In construction theory join. The Matlab simulation tools waveform...
In order to achieve the purpose of data collection, save and analyze. this paper, through LabVIEW-based platform, we designed a set signal acquisition storage playback in one virtual system, under premise microcontroller serial port send signals, waveforms which has been collected, Both with waveform spectrum analysis function functions traditional acquisition, The user can operation speed, gives design diagram simulation results.
Aviation bearing assembled detection is the final barrier to quality and safety. Therefore, an accurate method of aviation that based on local characteristics designed solve problem mis-assembly miss-assembly balls in assembled. When considering spatial limitation image acquisition, dynamic distribution interference lubricating grease surface, a ball segmentation model U-Net network with symmetrical structure achieve region bearing. Subsequently, incomplete circle fitting algorithm segmented...
Rain has an undesirable negative effect on the clarity of collected images. In situation where images are captured in rain, it can lead to a loss information and disability reflecting real situation. Consequently, rain become obstacle outdoor scientific research studies. The reason why difficult process is due indistinguishable interactions between background features textures. Since current image data only processed with CNN (convolutional neural network) model, trained network remove...
Network binarization (i.e., binary neural networks, BNNs) can efficiently compress deep networks and accelerate model inference but cause severe accuracy degradation. Existing BNNs are mainly implemented based on the commonly used full-precision network backbones, then is improved with various techniques. However, there a question of whether backbone well adapted to BNNs. We start from factors performance degradation analyze problems directly using backbones for BNNs: given computational...
With the rapid development of power industry, safety and reliability transmission lines have become a focal point attention. In this study, new method for front-end hazard identification line images using convolutional neural networks (CNN) based on deep learning technology is investigated. Transmission are collected model established image analysis defect recognition. Preprocessing techniques such as enhancement, segmentation, denoising applied to improve quality effectiveness images. Field...
We examine a class of English reflexive pronouns that we call middle-distance reflexives. show while not occurring in direct argument positions, reflexives can either be syntactically bound or interpreted according to pragmatic and discourse conditions, suggesting syntactic American extend beyond positions. will also discuss uses British Chinese comparison with those English. While these languages demonstrate variations the distribution reflexives, relevant facts indicate binding natural may...
As the convolutional neural network (CNN) gets deeper and wider in recent years, requirements for amount of data hardware resources have gradually increased. Meanwhile, CNN also reveals salient redundancy several tasks. The existing magnitude-based pruning methods are efficient, but performance compressed is unpredictable. While accuracy loss after based on structure sensitivity relatively slight, process time-consuming algorithm complexity notable. In this article, we propose a novel...
To apply deep CNNs to mobile terminals and portable devices, many scholars have recently worked on the compressing accelerating convolutional neural networks. Based this, we propose a novel uniform channel pruning (UCP) method prune CNN, modified squeeze-and-excitation blocks (MSEB) is used measure importance of channels in layers. The unimportant channels, including kernels related them, are pruned directly, which greatly reduces storage cost number calculations. There two types residual...