A5086112796- Domain Adaptation and Few-Shot Learning
- Anomaly Detection Techniques and Applications
- Multimodal Machine Learning Applications
- Machine Learning and Data Classification
- COVID-19 diagnosis using AI
- Medical Imaging and Analysis
- Machine Learning in Healthcare
- AI in cancer detection
- Advanced Neural Network Applications
- Topic Modeling
- Adversarial Robustness in Machine Learning
- Radiomics and Machine Learning in Medical Imaging
- Spine and Intervertebral Disc Pathology
- Network Security and Intrusion Detection
- Advanced Chemical Sensor Technologies
- Brain Tumor Detection and Classification
- Machine Learning and ELM
- Digital Imaging for Blood Diseases
- Respiratory viral infections research
- Spinal Fractures and Fixation Techniques
- Explainable Artificial Intelligence (XAI)
- Data-Driven Disease Surveillance
- Digital Media Forensic Detection
- Machine Learning and Algorithms
- Imbalanced Data Classification Techniques
King Abdullah University of Science and Technology
2024-2025
Carnegie Mellon University
2024
Mohamed bin Zayed University of Artificial Intelligence
2023-2024
Shandong University
2019-2023
Capital University
2023
Capital Medical University
2023
Beijing Anzhen Hospital
2023
Shandong University of Traditional Chinese Medicine
2017-2020
Western University
2018-2019
Nanjing Institute of Astronomical Optics & Technology
2012
Automated breast cancer multi-classification from histopathological images plays a key role in computer-aided diagnosis or prognosis. Breast is to identify subordinate classes of (Ductal carcinoma, Fibroadenoma, Lobular etc.). However, faces two main challenges from: (1) the great difficulties methods contrasting with classification binary (benign and malignant), (2) subtle differences multiple due broad variability high-resolution image appearances, high coherency cancerous cells, extensive...
Automated Screening of COVID-19 from chest CT is emergency and importance during the outbreak SARS-CoV-2 worldwide in 2020. However, accurate screening still a massive challenge due to spatial complexity 3D volumes, labeling difficulty infection areas, slight discrepancy between other viral pneumonia CT. While few pioneering works have made significant progress, they are either demanding manual annotations areas or lack interpretability. In this paper, we report our attempt towards achieving...
The automatic and precision classification for breast cancer histopathological image has a great significance in clinical application. However, the existing analysis approaches are difficult to addressing problem because feature subtle differences of inter-class accuracy still hard meet Recent advancements data-driven sharing processing multi-level hierarchical learning have made available considerable chance dope out solution this problem. To address challenging problem, we propose novel...
Source-free domain adaptation (SFDA) newly emerges to transfer the relevant knowledge of a well-trained source model an unlabeled target domain, which is critical in various privacy-preserving scenarios. Most existing methods focus on learning domain-invariant representations depending solely data, leading obtained are target-specific. In this way, they cannot fully address distribution shift problem across domains. contrast, we provide fascinating insight: rather than attempting learn...
Deep semi-supervised learning (SSL) methods aim to take advantage of abundant unlabeled data improve the algorithm performance. In this paper, we consider problem safe SSL scenario where unseen-class instances appear in data. This setting is essential and commonly appears a variety real applications. One intuitive solution removing these after detecting them during process. Nevertheless, performance identification limited by small number labeled ignoring availability To ensure performance,...
Unsupervised domain adaptation (UDA) enables a learning machine to adapt from labeled source an unlabeled target under the distribution shift. Thanks strong representation ability of deep neural networks, recent remarkable achievements in UDA resort domain-invariant features. Intuitively, goal is that good feature and hypothesis learned can generalize well domain. However, processes features hypotheses inevitably involve domain-specific information would degrade generalizability models on...
This paper addresses the new problem of automated screening coronavirus disease 2019 (COVID-19) based on chest X-rays, which is urgently demanded toward fast stopping pandemic. However, robust and accurate COVID-19 from X-rays still a globally recognized challenge because two bottlenecks: 1) imaging features share some similarities with other pneumonia 2) misdiagnosis rate very high, cost expensive. While few pioneering works have made much progress, they underestimate both crucial...
In many non-stationary environments, machine learning algorithms usually confront the distribution shift scenarios. Previous domain adaptation methods have achieved great success. However, they would lose algorithm robustness in multiple noisy environments where examples of source become corrupted by label noise, feature or open-set noise. this paper, we report our attempt toward achieving noise-robust adaptation. We first give a theoretical analysis and find that different noises disparate...
Source free domain adaptation (SFDA) aims to transfer a trained source model the unlabeled target without accessing data. However, SFDA setting faces performance bottleneck due absence of data and supervised information, as evidenced by limited gains newest methods. Active (ASFDA) can break through problem exploring exploiting small set informative samples via active learning. In this paper, we first find that those satisfying proper-ties neighbor-chaotic, individual-different,...
Deep semi-supervised learning (SSL) aims to utilize a sizeable unlabeled set train deep networks, thereby reducing the dependence on labeled instances. However, often carries unseen classes that cause SSL algorithm lose generalization. Previous works focus data level they attempt remove class or assign lower weight them but could not eliminate their adverse effects algorithm. Rather than focusing level, this paper turns attention model parameter level. We find only partial parameters are...
Global population aging presents increasing challenges to healthcare systems, with coronary artery disease (CAD) responsible for approximately 17.8 million deaths annually, making it a leading cause of global mortality. As CAD is largely preventable, early detection and proactive management are essential. In this work, we introduce DigitalShadow, an advanced warning system CAD, powered by fine-tuned facial foundation model. The pre-trained on 21 images subsequently into LiveCAD, specialized...
Source free domain adaptation (SFDA) transfers a single-source model to the unlabeled target without accessing source data. With intelligence development of various fields, zoo models is more commonly available, arising in new setting called multi-source-free (MSFDA). We find that critical inborn challenge MSFDA how estimate importance (contribution) each model. In this paper, we shed Bayesian light on fact posterior probability connects discriminability and transferability. propose...
Foundation models have attracted significant attention for their impressive generalizability across diverse downstream tasks. However, they are demonstrated to exhibit great limitations in representing high-frequency components and fine-grained details. In many medical imaging tasks, precise representation of such information is crucial due the inherently intricate anatomical structures, sub-visual features, complex boundaries involved. Consequently, limited prevalent foundation can result...
In non-stationary environments, learning machines usually confront the domain adaptation scenario where data distribution does change over time. Previous works have achieved great success in theory and practice. However, they always lose robustness noisy environments labels features of examples from source become corrupted. this paper, we report our attempt towards achieving accurate noise-robust adaptation. We first give a theoretical analysis that reveals how harmful noises influence...
Automated pancreas segmentation in abdominal computed tomography (CT) scans is of high clinical relevance (i.e. cancer diagnosis and prognosis), but extremely difficult because the a soft, small, flexible organ with anatomical variability, which causes previous methods to result low precision. In this study, authors present new deep recurrent adversarial network (DRAN) tackle challenge. DRAN contains three steps: (i) preserving global resolution CT modifying receptive field kernel adaptively...