A5100709711- Anomaly Detection Techniques and Applications
- Human Pose and Action Recognition
- Video Surveillance and Tracking Methods
- Topic Modeling
- Radio Frequency Integrated Circuit Design
- Optical Network Technologies
- Advanced Image and Video Retrieval Techniques
- Video Analysis and Summarization
- Network Security and Intrusion Detection
- Natural Language Processing Techniques
- Risk and Safety Analysis
- Speech and dialogue systems
- Occupational Health and Safety Research
- Advanced Neural Network Applications
- Multimodal Machine Learning Applications
- Nuclear and radioactivity studies
- Advanced Optical Network Technologies
- Advanced Malware Detection Techniques
- Hand Gesture Recognition Systems
- Photonic and Optical Devices
- Photonic Crystal and Fiber Optics
- Privacy-Preserving Technologies in Data
- Infrastructure Maintenance and Monitoring
- Advancements in Semiconductor Devices and Circuit Design
- Advanced Photonic Communication Systems
Shanghai Electric (China)
2025
The University of Tokyo
2018-2024
Chinese Academy of Medical Sciences & Peking Union Medical College
2018-2024
Peking Union Medical College Hospital
2018-2024
Yunnan University
2024
National University of Defense Technology
2016-2024
Shanghai Polytechnic University
2023-2024
Jiangsu Provincial Posts & Telecommunications Planning & Design Institute (China)
2023
Zhejiang University
2005-2023
Xinjiang University
2023
Mode-division multiplexing over fibers has attracted increasing attention the last few years as a potential solution to further increase fiber transmission capacity. In this paper, we demonstrate viability of orbital angular momentum (OAM) modes 50-km few-mode (FMF). By analyzing mode properties eigen in an FMF, study inner group differential modal delay (DMD) which may influence capacity long-distance OAM and multiplexing. To mitigate impact large DMD fiber-based transmission, use...
By mapping traditional amplitude modulation to spatial and employing adaptive optics compensation technique, we propose experimentally demonstrate a high-speed Bessel beam encoding/decoding free-space optical link through atmospheric turbulence. The speed is not limited by the conventional slow switching response of light modulator (SLM) but fully determined rate an intensity modulator, which easily supports tens gigabits per second resultant encoding/decoding. We use SLM loaded with...
Over the past decade, incorporation of spatial information has drawn increasing attention in multispectral and hyperspectral data analysis. In particular, property autocorrelation among pixels shown great potential for improving understanding remotely sensed imagery. this paper, we provide a comprehensive review state-of-the-art techniques incorporating image classification spectral unmixing. For classification, is accounted stages pre-classification, sample selection, classifiers,...
In this paper, with the help of knowledge base, we build and formulate a semantic space to connect source target languages, apply it sequence-to-sequence framework propose Knowledge-Based Semantic Embedding (KBSE) method.In our KB-SE method, sentence is firstly mapped into based space, generated using recurrent neural network internal meaning preserved.Experiments are conducted on two translation tasks, electric business data movie data, results show that proposed method can achieve...
As the services in elastic optical networks (EONs) are bandwidth-intensive, unpredictable and dynamic, increasing network factors emerging to affect performance of survivable planning operation from capacity efficiency. Most traditional protection restoration approaches may become before long inefficient due improvement a particular metric always at expense others. We argue that it would be more beneficial for comprehensive optimization consider main metrics jointly. Moreover, highly dynamic...
Decommissioning of the Fukushima Daiichi nuclear power station (NPS) is challenging due to industrial and chemical hazards as well radiological ones. The decommissioning workers in these sites are instructed wear proper Personal Protective Equipment (PPE) for radiation protection. However, may not be able accurately comply with safety regulations at sites, even prior education training. In response difficulties on-site PPE management, this paper presents a vision-based automated monitoring...
Task-oriented dialogue (ToD) benchmarks provide an important avenue to measure progress and develop better conversational agents. However, existing datasets for end-to-end ToD modeling are limited a single language, hindering the development of robust systems multilingual countries regions. Here we introduce BiToD, first bilingual multi-domain dataset task-oriented modeling. BiToD contains over 7k dialogues (144k utterances) with large realistic knowledge base. It serves as effective...
Graph neural networks (GNNs) have achieved remarkable empirical success in processing and representing graph-structured data across various domains. However, a significant challenge known as "oversmoothing" persists, where vertex features become nearly indistinguishable deep GNNs, severely restricting their expressive power practical utility. In this work, we analyze the asymptotic oversmoothing rates of GNNs with without residual connections by deriving explicit convergence for normalized...
ABSTRACT With the rise of social media and spread a large number pictures on Internet, protecting data privacy verifying copyright has become hot research. A common method is to use digital watermarking. However, existing blind watermarking methods only consider embedding watermark in image itself ignore fact that attacker can remove through other shooting devices. Therefore, solve this problem, we propose an based few‐shot learning. We autoencoder learn extraction watermarks. Then,...
The lack of labeled data is one the main challenges when building a task-oriented dialogue system. Existing datasets usually rely on human labeling, which expensive, limited in size, and low coverage. In this paper, we instead propose our framework auto-dialabel to automatically cluster intents slots. framework, collect set context features, leverage an autoencoder for feature assembly, adapt dynamic hierarchical clustering method intent slot labeling. Experimental results show that can...
Named entity recognition (NER) models generally perform poorly when large training datasets are unavailable for low-resource domains. Recently, pre-training a large-scale language model has become promising direction coping with the data scarcity issue. However, underlying discrepancies between modeling and NER task could limit models' performance, rarely been studied since collected small or but low quality. In this paper, we construct massive corpus relatively high quality, pre-train...
Relayed transmission is an efficient way to combat wireless channel fading and increase system throughput. By allocating the overall power transmitter relay optimally, performance can be further improved. This paper discusses optimal allocation in frequency-selective non-regenerative relaying channels, where OFDM employed. Firstly, between a cooperative subcarrier pair analyzed equivalent gain model for pairs developed. Based on this, sub-optimal selection method given by applying Hungarian...