A5006477225- Video Surveillance and Tracking Methods
- Domain Adaptation and Few-Shot Learning
- Advanced Image and Video Retrieval Techniques
- Human Pose and Action Recognition
- Robotics and Sensor-Based Localization
- Multimodal Machine Learning Applications
- Advanced Neural Network Applications
- Advanced Vision and Imaging
- Image Retrieval and Classification Techniques
- 3D Shape Modeling and Analysis
- Anomaly Detection Techniques and Applications
- Face and Expression Recognition
- Visual Attention and Saliency Detection
- Image Enhancement Techniques
- Tactile and Sensory Interactions
- COVID-19 diagnosis using AI
- Colorectal Cancer Screening and Detection
- Sparse and Compressive Sensing Techniques
- 3D Surveying and Cultural Heritage
- Robot Manipulation and Learning
- Generative Adversarial Networks and Image Synthesis
- Video Analysis and Summarization
- Advanced Image Processing Techniques
- Computer Graphics and Visualization Techniques
- Optical measurement and interference techniques
South China University of Technology
2023-2025
Yangzhou University
2023-2025
Chinese Academy of Sciences
2015-2024
Dalian Polytechnic University
2024
Soochow University
2024
University of Chinese Academy of Sciences
2017-2024
Shandong University
2024
Institute of Soil Science
2024
Shenyang Institute of Automation
2014-2023
State Key Laboratory of Robotics
2017-2022
Previous studies have suggested that breast cancer stem cells (BCSCs) mediate metastasis, are resistant to radiation and chemotherapy, contribute relapse. Although several BCSC markers been described, it is unclear whether these identify the same or independent BCSCs. Here, we show BCSCs exist in distinct mesenchymal-like (epithelial-mesenchymal transition [EMT]) epithelial-like (mesenchymal-epithelial [MET]) states. Mesenchymal-like characterized as CD24−CD44+ primarily quiescent localized...
We propose to detect abnormal events via a sparse reconstruction over the normal bases. Given an over-complete basis set (e.g., image sequence or collection of local spatio-temporal patches), we introduce cost (SRC) dictionary measure normalness testing sample. To condense size dictionary, novel selection method is designed with sparsity consistency constraint. By introducing prior weight each during reconstruction, proposed SRC more robust compared other outlier detection criteria. Our...
The rapid growth of consumer videos requires an effective and efficient content summarization method to provide a user-friendly way manage browse the huge amount video data. Compared with most previous methods that focus on sports news videos, personal is more challenging because its unconstrained lack any pre-imposed structures. We formulate as novel dictionary selection problem using sparsity consistency, where key frames selected such original can be best reconstructed from this...
The appearance of generative adversarial networks (GAN) provides a new approach and framework for computer vision. Compared with traditional machine learning algorithms, GAN works via training concept is more powerful in both feature representation. also exhibits some problems, such as non-convergence, model collapse, uncontrollability due to high degree freedom. How improve the theory apply it computer-vision-related tasks have now attracted much research efforts. In this paper, recently...
Federated machine learning which enables resource-constrained node devices (e.g., Internet of Things (IoT) and smartphones) to establish a knowledge-shared model while keeping the raw data local, could provide privacy preservation, economic benefit by designing an effective communication protocol. However, this protocol can be adopted attackers launch poisoning attacks for different nodes, has been shown as big threat most models. Therefore, we in article intend study vulnerability federated...
Unsupervised domain adaptation without accessing expensive annotation processes of target data has achieved remarkable successes in semantic segmentation. However, most existing state-of-the-art methods cannot explore whether representations across domains are transferable or not, which may result the negative transfer brought by irrelevant knowledge. To tackle this challenge, paper, we develop a novel <u>K</u> nowledge <u>A</u> ggregation-induced <u>T</u> ransferability <u>P</u> erception...
Robots are often required to generalize the skills learned from human demonstrations fulfil new task requirements. However, skill generalization will be difficult realize when facing with following situations: for a complex multistep includes number of features; some special constraints imposed on robots during process reproduction; and completely situation quite different one in which given robot. This work proposes framework facilitate robot generalization. The basic idea lies that first...
Unsupervised domain adaptation has attracted growing research attention on semantic segmentation. However, 1) most existing models cannot be directly applied into lesions transfer of medical images, due to the diverse appearances same lesion among different datasets; 2) equal been paid all representations instead neglecting irrelevant knowledge, which leads negative untransferable knowledge. To address these challenges, we develop a new unsupervised model including two complementary modules...
Video anomaly detection plays a critical role for intelligent video surveillance. We present an abnormal event system that considers both spatial and temporal contexts. To characterize the video, we first perform spatio-temporal segmentation then propose new region-based descriptor called "Motion Context," to describe motion appearance information of segment. For measurements, formulate as matching problem, which is more robust than statistic model-based methods, especially when training...
<h3>Importance</h3> Despite data aggregation and removal of protected health information, there is concern that deidentified physical activity (PA) collected from wearable devices can be reidentified. Organizations collecting or distributing such suggest the aforementioned measures are sufficient to ensure privacy. However, no studies, our knowledge, have been published demonstrate possibility impossibility reidentifying data. <h3>Objective</h3> To evaluate feasibility accelerometer-measured...
Underwater robot technologies are crucial for marine resource exploration and autonomous manipulation, many breakthroughs have been achieved with key indicators (e.g., dive depth navigation range). However, due to the complicated underwater environment, state-of-the-art sensing cannot handle all needs of observations. To improve operating capacity robots, there is an urgent need develop technology. Therefore, in this paper, we first introduce development platforms. We then review some such...
Accurate lesion segmentation based on endoscopy images is a fundamental task for the automated diagnosis of gastrointestinal tract (GI Tract) diseases. Previous studies usually use hand-crafted features representing images, while feature definition and are treated as two standalone tasks. Due to possible heterogeneity between models, these methods often result in sub-optimal performance. Several fully convolutional networks have been recently developed jointly perform learning model training...
Spectral clustering (SC) has become one of the most widely-adopted algorithms, and been successfully applied into various applications. We in this work explore problem spectral a lifelong learning framework termed as <u>G</u> eneralized <u>L</u> ife <u>l</u> ong <u>S</u> pectral <u>C</u> lustering (GL <inline-formula><tex-math notation="LaTeX">$^2$</tex-math></inline-formula> SC). Different from current studies, which concentrate on fixed task set cannot efficiently incorporate new task,...
Text-to-image generative models can produce diverse high-quality images of concepts with a text prompt, which have demonstrated excellent ability in image generation, translation, etc. We this work study the problem synthesizing instantiations user's own never-ending manner, <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">i.e.,</i> create your world, where new from user are quickly learned few examples. To achieve goal, we propose <underline...
Consider the lifelong machine learning paradigm whose objective is to learn a sequence of tasks depending on previous experiences, e.g., knowledge library or deep network weights. However, libraries networks for most recent models are prescribed size and can degenerate performance both learned coming ones when facing with new task environment (cluster). To address this challenge, we propose novel incremental clustered framework two libraries: feature model library, called <u>F</u>lexible...
Federated learning (FL) is a hot collaborative training framework via aggregating model parameters of decentralized local clients. However, most FL methods unreasonably assume data categories are known and fixed in advance. Moreover, some new clients that collect novel unseen by other may be introduced to irregularly. These issues render global undergo catastrophic forgetting on old categories, when receive consecutively under limited memory storing categories. To tackle the above issues, we...
Initializing an effective dictionary is indispensable step for sparse representation. In this paper, we focus on the selection problem with objective to select a compact subset of basis from original training data instead learning new matrix as models do. We first design model via l2,0 norm. For optimization, propose two methods: one standard forward-backward greedy algorithm, which not suitable large-scale problems; other based gradient cues at each forward iteration and speeds up process...
Weakly-supervised learning under image-level labels supervision has been widely applied to semantic segmentation of medical lesions regions. However, 1) most existing models rely on effective constraints explore the internal representation lesions, which only produces inaccurate and coarse regions; 2) they ignore strong probabilistic dependencies between target dataset (e.g., enteroscopy images) well-to-annotated source diseases gastroscope images). To better utilize these dependencies, we...
Here, we show that HEMATOLOGICAL AND NEUROLOGICAL EXPRESSED 1-LIKE (HN1L) is a targetable breast cancer stem cell (BCSC) gene altered in 25% of whole and significantly correlated with shorter overall or relapse-free survival triple-negative (TNBC) patients. HN1L silencing reduced the population BCSCs, inhibited tumor initiation, resensitized chemoresistant tumors to docetaxel, hindered progression multiple TNBC line-derived xenografts. Additionally, signatures associated disease-free We...
The state-of-the-art multitask multiview (MTMV) learning tackles a scenario where multiple tasks are related to each other via shared feature views. However, in many real-world scenarios sequence of the task comes, higher storage requirement and computational cost retraining previous with MTMV models have presented formidable challenge for this lifelong scenario. To address challenge, article, we propose new continual model that integrates deep matrix factorization sparse subspace unified...
Weakly-supervised learning has attracted growing research attention on medical lesions segmentation due to significant saving in pixel-level annotation cost. However, 1) most existing methods require effective prior and constraints explore the intrinsic characterization, which only generates incorrect rough prediction; 2) they neglect underlying semantic dependencies among weakly-labeled target enteroscopy diseases fully-annotated source gastroscope lesions, while forcefully utilizing...