A5101618713- Remote Sensing and Land Use
- Remote-Sensing Image Classification
- Statistical Methods and Inference
- Land Use and Ecosystem Services
- Advanced Statistical Methods and Models
- Face and Expression Recognition
- Control Systems and Identification
- Advanced Image Fusion Techniques
- Advanced Algorithms and Applications
- Advanced Computational Techniques and Applications
- Advanced Image and Video Retrieval Techniques
- Neural Networks and Applications
- Remote Sensing in Agriculture
- 3D Surveying and Cultural Heritage
- Advanced Vision and Imaging
- Geochemistry and Geologic Mapping
- Water Quality Monitoring and Analysis
- Image Retrieval and Classification Techniques
- Simulation and Modeling Applications
- Aquatic Ecosystems and Phytoplankton Dynamics
- Bayesian Methods and Mixture Models
- Marine and coastal ecosystems
- Imbalanced Data Classification Techniques
- Bayesian Modeling and Causal Inference
- Anomaly Detection Techniques and Applications
Tongji University
2015-2025
Chongqing University
2024-2025
Chongqing Cancer Hospital
2024-2025
Harbin Medical University
2024
Wuhan University
2013-2024
Hangzhou Dianzi University
2024
State Key Laboratory of Remote Sensing Science
2018-2023
Sichuan University
2023
University of Hong Kong
2018-2020
Peking University
2018
Summary We consider the problem of selecting grouped variables (factors) for accurate prediction in regression. Such a arises naturally many practical situations with multifactor analysis-of-variance as most important and well-known example. Instead factors by stepwise backward elimination, we focus on accuracy estimation extensions lasso, LARS algorithm non-negative garrotte factor selection. The are recently proposed regression methods that can be used to select individual variables. study...
We propose penalized likelihood methods for estimating the concentration matrix in Gaussian graphical model. The lead to a sparse and shrinkage estimator of that is positive definite, thus conduct model selection estimation simultaneously. implementation nontrivial because definite constraint on matrix, but we show computation can be done effectively by taking advantage efficient maxdet algorithm developed convex optimization. BIC-type criterion tuning parameter methods. connection between...
Two-category support vector machines (SVM) have been very popular in the machine learning community for classification problems. Solving multicategory problems by a series of binary classifiers is quite common SVM paradigm; however, this approach may fail under various circumstances. We propose (MSVM), which extends to case and has good theoretical properties. The proposed method provides unifying framework when there are either equal or unequal misclassification costs. As tuning criterion...
We propose a new method for model selection and fitting in multivariate nonparametric regression models, the framework of smoothing spline ANOVA. The “COSSO” is regularization with penalty functional being sum component norms, instead squared norm employed traditional method. COSSO provides unified several recent proposals linear models ANOVA models. Theoretical properties, such as existence rate convergence estimator, are studied. In special case tensor product design periodic functions,...
AbstractIn this article we study random forests through their connection with a new framework of adaptive nearest-neighbor methods. We introduce concept potential nearest neighbors (k-PNNs) and show that can be viewed as adaptively weighted k-PNN Various aspects studied from perspective. the effect terminal node sizes on prediction accuracy forests. further splitting schemes assign weights to k-PNNs in desirable way: for estimation at given target point, these voting point according local...
Summary We study the non-negative garrotte estimator from three different aspects: consistency, computation and flexibility. argue that is a general procedure can be used in combination with estimators other than original least squares as its form. In particular, we consider using lasso, elastic net ridge regression along ordinary initial estimate garrotte. prove has nice property that, probability tending to 1, solution path contains an correctly identifies set of important variables...
We propose an empirical Bayes method for variable selection and coefficient estimation in linear regression models. The is based on a particular hierarchical formulation, the estimator shown to be closely related LASSO estimator. Such connection allows us take advantage of recently developed quick algorithm compute estimate, provides new way select tuning parameter method. Unlike previous methods, which most practical situations can implemented only through greedy stepwise algorithm, our...
Abstract The inference of cell–cell communication (CCC) is crucial for a better understanding complex cellular dynamics and regulatory mechanisms in biological systems. However, accurately inferring spatial CCCs at single-cell resolution remains significant challenge. To address this issue, we present versatile method, called DeepTalk, to infer CCC by integrating RNA sequencing (scRNA-seq) data transcriptomics (ST) data. DeepTalk utilizes graph attention network (GAT) integrate scRNA-seq ST...
To deal with the curse of dimensionality in high-dimensional nonparametric problems, we consider using tensor product space ANOVA models, which extend popular additive models and are able to capture interactions any order. The multivariate function is given an decomposition, that is, it expressed as a constant plus sum functions one variable (main effects), two variables (two-factor interactions)and so on. We assume be spaces.We show both regression white noise settings, optimal rate...
The analysis of experiments in which numerous potential variables are examined is driven by the principles effect sparsity, hierarchy, and heredity. We propose an efficient variable selection strategy to specifically address unique challenges faced such analysis. proposed methods natural extensions LARS general-purpose algorithm. They can be computed very rapidly find sparse models that better satisfy goals experiments. Simulations real examples used illustrate wide applicability methods.
With the continuous increase in highway traffic volume, abnormal events occur frequently, seriously affecting fluidity and safety. This study aims to develop an efficient event detection method. Based on simulation data generated by VISSIM software, K-means clustering algorithm is applied, with speed, density, occupancy rate as key feature parameters, aiming effectively identify highways. The research results show that, after comparing performance of models using different combinations model...
AbstractThis article presents a nonparametric penalized likelihood approach for variable selection and model building, called basis pursuit (LBP). In the setting of tensor product reproducing kernel Hilbert space, we decompose log-likelihood into sum different functional components such as main effects interactions, with each component represented by appropriate functions. Basis functions are chosen to be compatible building in context smoothing spline ANOVA model. is applied obtain optimal...