A5070982537- Generative Adversarial Networks and Image Synthesis
- Domain Adaptation and Few-Shot Learning
- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Neural Networks and Applications
- Advanced Image and Video Retrieval Techniques
- Face and Expression Recognition
- Machine Learning and ELM
- Human Pose and Action Recognition
- Anomaly Detection Techniques and Applications
- Advanced Image Processing Techniques
- Multimodal Machine Learning Applications
- Advanced Vision and Imaging
- Neural dynamics and brain function
- Cancer-related gene regulation
- Face recognition and analysis
- Helicobacter pylori-related gastroenterology studies
- Video Analysis and Summarization
- Machine Learning and Data Classification
- Gene expression and cancer classification
- Adversarial Robustness in Machine Learning
- Handwritten Text Recognition Techniques
- Advanced Clustering Algorithms Research
- Blind Source Separation Techniques
- Digital Media Forensic Detection
South China University of Technology
2016-2025
City University of Hong Kong
2004-2024
Wuhan University
2022-2024
Inner Mongolia Medical University
2023-2024
Hunan Normal University
2024
Jilin University
2024
Union Hospital
2024
Sinopec (China)
2024
Peking University
2020-2023
Guangzhou Chemistry (China)
2022-2023
Incomplete multi-view clustering (IMVC) is challenging, as it requires adequately exploring complementary and consistency information under the incompleteness of data. Most existing approaches attempt to overcome at instance-level. In this work, we develop a new approach facilitate IMVC from perspective. Specifically, transfer issue missing instances similarity graph completion problem for incomplete views, propose self-supervised algorithm infer associated entries. Further, by incorporating...
Traditional cluster ensemble approaches have three limitations: (1) They do not make use of prior knowledge the datasets given by experts. (2) Most conventional methods cannot obtain satisfactory results when handling high dimensional data. (3) All members are considered, even ones without positive contributions. In order to address limitations approaches, we first propose an incremental semi-supervised clustering framework (ISSCE) which makes advantage random subspace technique, constraint...
In this work, we study a more realistic challenging scenario in multiview clustering (MVC), referred to as incomplete MVC (IMVC) where some instances certain views are missing. The key IMVC is how adequately exploit complementary and consistency information under the incompleteness of data. However, most existing methods address problem at instance level they require sufficient perform data recovery. develop new approach facilitate based on graph propagation perspective. Specifically,...
Call admission control is one of the key elements in ensuring quality service mobile wireless networks. The traditional trunk reservation policy and its numerous variants give preferential treatment to handoff calls over new arrivals by reserving a number radio channels exclusively for handoffs. Such schemes, however, cannot adapt changes traffic pattern due static nature. This paper introduces novel stable dynamic call mechanism (SDCA), which can maximize channel utilization subject...
People naturally escape from a place when unexpected events happen. Based on this observation, efficient detection of crowd behavior in surveillance videos is promising way to perform timely anomalous situations. In paper, we propose Bayesian framework for by directly modeling motion both the presence and absence events. Specifically, introduce concepts potential destinations divergent centers characterize above two cases respectively, construct corresponding class-conditional probability...
Cluster analysis of gene expression data from a cDNA microarray is useful for identifying biologically relevant groups genes. However, finding the natural clusters in and estimating correct number are still two largely unsolved problems. In this paper, we propose new clustering framework that able to address both these By using one-prototype-take-one-cluster (OPTOC) competitive learning paradigm, proposed algorithm can find input data, solution not sensitive initialization. order estimate...
Oriented object detection, which aims at detecting objects with orientation property, shows great potential for visual analysis in complex scenarios, such as aerial images. However, the powerful detection performance relies on abundant and accurate annotations, deteriorates once annotations become insufficient. Semi-supervised learning, utilizes unannotated data to improve target model, is a promising method address problem of annotation deficiency. In this work, we propose Pseudo-Siamese...
Nowadays, massive amounts of data are available for analysis in natural and social systems the tasks to depict system structures from data, i.e., inverse problems, become one central issues wide interdisciplinary fields. In this paper, we study problem dynamic complex networks driven by white noise. A simple universal inference formula double correlation matrices noise-decorrelation (DCMND) method is derived analytically, numerical simulations confirm that DCMND can accurately both network...
Deep mutual learning jointly trains multiple essential networks having similar properties to improve semi-supervised classification. However, the commonly used consistency regularization between outputs of may not fully leverage difference them. In this paper, we explore how capture complementary information enhance learning. For purpose, propose a correction network (CCN), built on top networks, learn mapping from output one ground truth label, conditioned features learnt by another. To...
Face retouching aims to remove facial imperfections from image and videos while at the same time preserving face attributes. The existing methods are designed perform non-interactive end-to-end retouching, ability interact with users is highly demanded in downstream applications. In this paper, we propose RetouchGPT, a novel framework that leverages Large Language Models (LLMs) guide interactive process. Towards end, design an instruction-driven imperfection prediction module accurately...
The rapid advancement of 3D Generative Adversarial Networks (GANs) has significantly enhanced the diversity and quality generated images. Despite these breakthroughs, manipulation capabilities GANs remain unexplored, presenting substantial challenges for practical applications where user interaction modification are essential. Current methods often lack precision needed fine-grained attribute manipulation, struggle to maintain multi-view consistency during editing process. To address...
In this paper, we propose a crowd motion partitioning approach based on local-translational approximation in scattered field. To represent an accurate and parsimonious way, compute optical flow at the salient locations instead of all pixel locations. We then transform problem into field segmentation. Based our assumption that local can be approximated by translational field, develop local-translation domain segmentation (LTDS) model which evolution boundaries is derived from Gâteaux...
Classification of high-dimensional data with very limited labels is a challenging task in the field mining and machine learning. In this paper, we propose multiobjective semisupervised classifier ensemble (MOSSCE) approach to address challenge. Specifically, subspace selection process (MOSSP) MOSSCE first designed generate optimal combination feature subspaces. Three objective functions are then proposed for MOSSP, which include relevance features, redundancy between reconstruction error....
Learning class-conditional data distributions is crucial for Generative Adversarial Networks (GAN) in semi-supervised learning. To improve both instance synthesis and classification this setting, we propose an enhanced TripleGAN (EnhancedTGAN) model work. We follow the adversarial training scheme of original TripleGAN, but completely re-design targets generator classifier. Specifically, adopt feature-semantics matching to enhance learning from aspects statistics latent space semantics...
Using an ensemble of neural networks with consistency regularization is effective for improving performance and stability deep learning, compared to the case a single network. In this paper, we present semi-supervised Deep Coupled Ensemble (DCE) model, which contributes learning classification landmark exploration better locating final decision boundaries in learnt latent space. First, multiple complementary regularizations are integrated into our DCE model enable members learn from each...