A5101634327- Advanced Neural Network Applications
- Advanced Vision and Imaging
- Statistical Methods and Inference
- Mathematical and Theoretical Epidemiology and Ecology Models
- Statistical Methods and Bayesian Inference
- Nonlinear Differential Equations Analysis
- Optical measurement and interference techniques
- Advanced Causal Inference Techniques
- Reinforcement Learning in Robotics
- Data Visualization and Analytics
- Robotics and Sensor-Based Localization
- Image Retrieval and Classification Techniques
- Spectroscopy and Chemometric Analyses
- Adversarial Robustness in Machine Learning
- Time Series Analysis and Forecasting
- Artificial Intelligence in Healthcare and Education
- Brain Tumor Detection and Classification
- 3D Surveying and Cultural Heritage
- Glycosylation and Glycoproteins Research
- Satellite Communication Systems
- Domain Adaptation and Few-Shot Learning
- Antenna Design and Optimization
- Enzyme Catalysis and Immobilization
- Advanced Image and Video Retrieval Techniques
- Machine Learning and Data Classification
Alibaba Group (China)
2020-2023
Beijing Jiaotong University
2021-2023
First Affiliated Hospital of Guangdong Pharmaceutical University
2022
Guangdong Pharmaceutical University
2022
Chinese Academy of Sciences
2006-2020
Shandong Institute of Automation
2020
Huazhong Agricultural University
2014
Worcester Polytechnic Institute
2010
Institute of Oceanology
2006
Synchronized stochastic gradient descent (SGD) optimizers with data parallelism are widely used in training large-scale deep neural networks. Although using larger mini-batch sizes can improve the system scalability by reducing communication-to-computation ratio, it may hurt generalization ability of models. To this end, we build a highly scalable learning for dense GPU clusters three main contributions: (1) We propose mixed-precision method that significantly improves throughput single...
Benefiting from the powerful expressive capability of graphs, graph-based approaches have been popularly applied to handle multi-modal medical data and achieved impressive performance in various biomedical applications. For disease prediction tasks, most existing methods tend define graph manually based on specified modality (e.g., demographic information), then integrated other modalities obtain patient representation by Graph Representation Learning (GRL). However, constructing an...
Accurate LiDAR-camera online calibration is critical for modern autonomous vehicles and robot platforms. Dominant methods heavily rely on hand-crafted features, which are not scalable in practice. With the increasing popularity of deep learning (DL), a few recent efforts have demonstrated advantages DL feature extraction this task. However, their reported performances sufficiently satisfying yet. We believe one improvement can be problem formulation with proper consideration underneath...
Transformers have achieved remarkable success across diverse domains, but their monolithic architecture presents challenges in interpretability, adaptability, and scalability. This paper introduces a novel modular Transformer that explicitly decouples knowledge reasoning through generalized cross-attention mechanism to shared base, specifically designed for effective retrieval. Critically, we provide rigorous mathematical derivation demonstrating the Feed-Forward Network (FFN) standard is...
Abstract The use of testing and calibration equipment to calibrate maintain satellite ground station systems is an important step in the construction operation stations. Calibration can eliminate system errors generated during stations, but traditional fixed towers cannot meet far-field usage conditions Ka frequency bands. Therefore, using unmanned aerial vehicle platforms instead simultaneously needs S/X/Ka tri band improve manoeuvrability flexibility usage, reduce costs. Through study laws...
Distributed training techniques have been widely deployed in large-scale deep neural networks (DNNs) on dense-GPU clusters. However, public cloud clusters, due to the moderate inter-connection bandwidth between instances, traditional state-of-the-art distributed systems cannot scale well models. In this paper, we propose a new computing and communication efficient top-k sparsification library for training. To further improve system scalability, optimize I/O by proposing simple yet...
We consider image classification in a weakly supervised scenario where the training data are annotated at different levels of abstractions. A subset with coarse labels (e.g. wolf, dog), while rest fine breeds wolves and dogs). Each label corresponds to superclass several labels. Our goal is learn model that can classify new into one classes. investigate how coarsely labeled help improve classification. Since it usually much easier collect than those labels, problem setup considered this...
Esterase, as a type of powerful catabolic enzyme for the degradation pyrethroid pesticides (PYRs), appears promising in improving quality crops and environment contaminated by pesticide residues. The purpose this research is to provide detailed introduction enzymatic properties, optimal production immobilization conditions, ability Est804 PYRs. study on properties indicated that was an alkaline esterase with pH 8.0 broad temperature range 35−50°C. activity free calculated be 112.812 U,...
<title>Abstract</title> The concept of causality plays a significant role in human cognition. In the past few decades, causal effect estimation has been well developed many fields, such as computer science, medicine, economics, and other industrial applications. With advancement deep learning, it increasingly applied against counterfactual data. Typically, models map characteristics covariates to representation space then design various objective functions estimate data unbiasedly. Different...
Although various mining algorithms have been proposed in the literature to efficiently compute clusters, few strides made date helping analysts interactively explore such patterns stream context. We present a framework called CLUES both computationally and visually support process of real-time density-based clusters. is composed three major components. First, as foundation CLUES, we develop an evolution model clusters data streams that captures complete spectrum cluster types across...
In this paper, we explore the task to estimate density map of objects from single image with unknown perspective map. We follow recent progress in object counting through estimation. Object estimation is usually suffered scale variance caused by perspective. addition, background and irrelevant lead artifacts resulting maps, which build up error when heat maps are needed aggregating maps. propose a multi-CNN network gradient boosting (MCNN-boost) address these two problems at once. The...
In order to construct the high-precision and high-fidelity 3D plants model quickly economically, this study present a new approach combing color structured light with silhouette-based method repair model's depth information. First, project coded which is using spatial code on surface of plants. And photograph two non-contract cameras from different angles. Second, we need extract center stripes decode base space conversion. Then, gain feature points matching epipolar constraint. Finally,...
Chlorophyll content plays an important role in the growth of crop. crop not only shows situation but also has significance disease diagnosis. By using hyperspectral imaging technology, chlorophyll can be diagnosed non-destructively. Before correlation analysis, experiment gets SPAD values and vegetation indexes. The result that VOG1 quadratic model best performance (R <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> =0.83 RMSE=2.68)....
Scalability remains a challenge in multi-agent reinforcement learning and is currently under active research. A framework named mean-field (MFRL) could alleviate the scalability problem by employing Mean Field Theory to turn many-agent into two-agent problem. However, this lacks ability identify essential interactions nonstationary environments. Causality contains relatively invariant mechanisms behind interactions, though environments are nonstationary. Therefore, we propose an algorithm...
The display of surfaces and solids has usually been restricted to the domain scientific visualization; however, little work done on visualization dimensionality higher than three or four. Indeed, most high-dimensional focuses data points. However, ability eectively model visualize dimensional objects such as clusters patterns would be quite useful in studying their shapes, relationships, changes over time. In this paper we describe a method for description, extraction, N-dimensional solids....
Treatment effect estimation, which refers to the estimation of causal effects and aims measure strength relationship, is great importance in many fields but a challenging problem practice. As present, data-driven faces two main challenges, i.e., selection bias missing counterfactual. To address these issues, most existing approaches tend reduce by learning balanced representation, then estimate counterfactual through representation. However, they heavily rely on finely hand-crafted metric...