A5101925135- Stochastic Gradient Optimization Techniques
- Advanced Neural Network Applications
- Privacy-Preserving Technologies in Data
- Bioinformatics and Genomic Networks
- Sentiment Analysis and Opinion Mining
- Advanced Text Analysis Techniques
- AI in cancer detection
- Digital Imaging for Blood Diseases
- Text and Document Classification Technologies
- Target Tracking and Data Fusion in Sensor Networks
- Complex Network Analysis Techniques
- Radiomics and Machine Learning in Medical Imaging
- Gene expression and cancer classification
- Advanced Clustering Algorithms Research
- Visual Attention and Saliency Detection
- Ferroelectric and Negative Capacitance Devices
- Lung Cancer Diagnosis and Treatment
- Emotion and Mood Recognition
- Metabolomics and Mass Spectrometry Studies
- Sparse and Compressive Sensing Techniques
- Optical Imaging and Spectroscopy Techniques
- Network Packet Processing and Optimization
- Meat and Animal Product Quality
- Animal Behavior and Welfare Studies
- Smart Agriculture and AI
Xi’an University of Posts and Telecommunications
2022-2023
South China Agricultural University
2023
Peking Union Medical College Hospital
2022
Chinese Academy of Medical Sciences & Peking Union Medical College
2022
Alibaba Group (United States)
2019-2021
Tongji University
2021
Alibaba Group (China)
2021
University of South China
2021
Rice University
2021
Carnegie Mellon University
2020
In federated optimization, heterogeneity in the clients' local datasets and computation speeds results large variations number of updates performed by each client communication round. Naive weighted aggregation such models causes objective inconsistency, that is, global model converges to a stationary point mismatched function which can be arbitrarily different from true objective. This paper provides general framework analyze convergence heterogeneous optimization algorithms. It subsumes...
In federated learning, heterogeneity in the clients' local datasets and computation speeds results large variations number of updates performed by each client communication round. Naive weighted aggregation such models causes objective inconsistency, that is, global model converges to a stationary point mismatched function which can be arbitrarily different from true objective. This paper provides general framework analyze convergence optimization algorithms with heterogeneous training...
The live weight of pigs has always been an important reference index for growth monitoring and the health status breeding pigs. An accurate acquisition is key to guide scientific feeding improve economic benefits. Compared with traditional contact measurement method, non-contact weighing method can greatly reduce human–pig errors. In this paper, a deep neural network constructed which automatically accurately predict by measuring multiple body parameters. Because good generalization ability...
Distributed stochastic gradient descent (SGD) is essential for scaling the machine learning algorithms to a large number of computing nodes. However, infrastructures variability such as high communication delay or random node slowdown greatly impedes performance distributed SGD algorithm, especially in wireless system sensor networks. In this paper, we propose an algorithmic approach named Overlap Local-SGD (and its momentum variant) overlap and computation so speedup training procedure. The...
Graph neural networks (GNNs) have gained increasing popularity among researchers recently and been employed in many applications. Training GNNs introduce a crucial stage called graph sampling. One of the most important sampling algorithms is Random Walk. However, Walk its variants share suffer from same performance problem caused by random fragmented memory access patterns, leading to significant system degradation. In this work, we present an efficient engine on modern FPGAs integrated with...
Abstract Background The lack of standardized delineation lymph node station in lung cancer radiotherapy leads to nonstandard clinical target volume (CTV) contouring, especially patients with bulky lump gross nodes (GTVnd). This study defines region boundaries for and automatically contours stations based on the International Association Study Lung Cancer (IASLC) map. Methods Computed tomography (CT) scans 200 small cell were collected. zone defined IASLC map, adjustments meet requirements....
Pathology image classification plays an important role in cancer diagnosis and precision treatment. Convolutional neural network has been widely employed pathology classification. Due to its convolution pooling operation, it a great advantage extracting local features of small objects images. However, lacks the ability extract global contextual information contained long-rang tissue structures Transformer, which adopts innate self-attention mechanisms, obtained remarkable performance on...
Convolutional Neural Network (CNN) models are becoming complex with advanced OPs and structures, which introduces design challenges for FPGA-based system. In this paper, we present the of an CNN inference system, PAI-FCNN, to support modern models. PAI-FCNN consists scalable hardware a model reconstruction flow in software compiler. way, like Deconv, Conv upsampling, Dilated Conv, Concatenation can be processed by high performance efficiency. also incorporates reduced precision boost...
The detection of protein complexes from protein-protein interaction network is a fundamental issue in bioinformatics and systems biology. To solve this problem, numerous methods have been proposed different angles the past decades. However, study on detecting statistically significant still has not received much attention. Although there are few available literature for identifying complexes, none these can provide more strict control error rate complex terms family-wise (FWER). In paper, we...
The reasons for fuzzy data association in a densely cluttered environment are analyzed this paper, and possibilistic based algorithm multi-target tracking is proposed. This paper fully analyses the shortcomings of conventional approaches proposes which can improve performance multiple targets tracking. It reduce errors caused by clutters greatly. simulation results show that has superiority over approaches, track real time.
In recent years, with the rapid advancements in large language models (LLMs), achieving excellent empathetic response capability has become a crucial prerequisite. Consequently, managing and understanding large-scale video datasets gained increasing importance. However, data are typically trained without any quality selection, leading to inefficient usage wasted computational resources. Additionally, using raw can result low performance dialogues. this work, we present Efficient-Empathy,...
In recent years, with the rapid advancements in large language models (LLMs), achieving excellent empathetic response capabilities has become a crucial prerequisite. Consequently, managing and understanding datasets have gained increasing significance. However, data are typically human-labeled, leading to insufficient wasted human labor. this work, we present Synth-Empathy, an LLM-based generation quality diversity selection pipeline that automatically generates high-quality while discarding...
Distributed stochastic gradient descent (SGD) is essential for scaling the machine learning algorithms to a large number of computing nodes. However, infrastructures variability such as high communication delay or random node slowdown greatly impedes performance distributed SGD algorithm, especially in wireless system sensor networks. In this paper, we propose an algorithmic approach named Overlap-Local-SGD (and its momentum variant) overlap and computation so speedup training procedure. The...
The main objective of this study was to use a short time series image data extract the paddy field spatial information effectively. For purpose, paper put forward method extracting by combining multi-temporal remote sensing images. In paper, Multi-temporal GF-1 Jiang Ning District in Nanjing where planting large number rice used as data. Then datasets for extraction were constructed with two parts covering growth cycle rice, one is multiband reflectance(MBR) their derivative features (WDRVI...
Pedestrian detection is one of hot topics in the field computer vision and pattern recognition, which great value to video surveillance, intelligent traffic human-computer interaction, etc. how improve rate speed key. Most traditional pedestrian methods are based on pyramid sliding window scanning method, for images majority region does not contain a body, inefficient. This study presents body sampling algorithm binarised normed gradients, can quickly effectively extract image that most...
Classification of nuclei in histopathology images is an important step pathology workflow. Although deep learning methods have been extensively employed this task, automatic classification still a challenging task because the large inter-and intra-class variability as well serious clustered together. To address challenge, we propose modified network for images. The proposed method adopts encoding-decoding structure and leverages segmented instance to guide with same feature maps from...
Aiming at the problem of Direction-of-arrival (DOA) time-varying, a covariance matrix method with noise removal DOA tracking particle filter algorithm based on generalized label multi-Bernoulli (GLMB) is proposed in this paper. Since measurement uniform linear array (ULA) superimposed, will cause association map (MAP) GLMB update step to mismatch. In order make match successful, we propose novel Measurement Association Mapping (NMAP) strategy. And then, using MUSIC spatial spectrum function...