A5069418884- Machine Learning in Healthcare
- Time Series Analysis and Forecasting
- Anomaly Detection Techniques and Applications
- Machine Learning and Data Classification
- Neural Networks and Reservoir Computing
- Imbalanced Data Classification Techniques
- Explainable Artificial Intelligence (XAI)
- Text and Document Classification Technologies
- Privacy-Preserving Technologies in Data
- Neural Networks and Applications
- COVID-19 diagnosis using AI
- Complex Systems and Time Series Analysis
- Advanced Memory and Neural Computing
- Infrastructure Maintenance and Monitoring
- Traditional Chinese Medicine Studies
- Optical Network Technologies
- Graph Theory and Algorithms
- Industrial Vision Systems and Defect Detection
- Rough Sets and Fuzzy Logic
- Biomedical Text Mining and Ontologies
- Face and Expression Recognition
- Multimodal Machine Learning Applications
- Healthcare Technology and Patient Monitoring
- FinTech, Crowdfunding, Digital Finance
- Chronic Disease Management Strategies
Xi'an Jiaotong University
2024
Peking University
2021-2023
San Francisco State University
2019-2021
Hypergraphs are natural and expressive modeling tools to encode high-order relationships among entities. Several variations of Hypergraph Neural Networks (HGNNs) proposed learn the node representations complex in hypergraphs. Most current approaches assume that input hypergraph structure accurately depicts relations However, inevitably contains noise, task-irrelevant information, or false-negative connections. Treating as ground-truth information unavoidably leads sub-optimal performance. In...
Prediction based on Irregularly Sampled Time Series (ISTS) is of wide concern in real-world applications. For more accurate prediction, methods had better grasp data characteristics. Different from ordinary time series, ISTS characterized by irregular intervals intra-series and different sampling rates inter-series. However, existing have suboptimal predictions due to artificially introducing new dependencies a series biasedly learning relations among when modeling these two In this work, we...
Continuous diagnosis and prognosis are essential for critical patients. They can provide more opportunities timely treatment rational allocation. Although deep-learning techniques have demonstrated superiority in many medical tasks, they frequently forget, overfit, produce results too late when performing continuous prognosis. In this work, we summarize the four requirements; propose a concept, classification of time series (CCTS); design training method deep learning, restricted update...
Abstract Monitoring aircraft structural health with changing loads is critical in aviation and aerospace engineering. However, the load equation needs to be calibrated by ground testing which costly, inefficient. Here, we report a general deep learning-based model for strain prediction calibration through two-phase process. First, identified causality between key flight parameters strains. The was then integrated into monitoring process build more coefficients calibration. This achieves...
In the real world, class of a time series is usually labeled at final time, but many applications require to classify every point. e.g. outcome critical patient only determined end, he should be diagnosed all times for timely treatment. Thus, we propose new concept: Continuous Classification Time Series (CCTS). It requires model learn data in different stages. But evolves dynamically, leading distributions. When learns multi-distribution, it always forgets or overfits. We suggest that...
Continuous diagnosis and prognosis are essential for intensive care patients. It can provide more opportunities timely treatment rational resource allocation, especially sepsis, a main cause of death in ICU, COVID-19, new worldwide epidemic. Although deep learning methods have shown their great superiority many medical tasks, they tend to catastrophically forget, over fit, get results too late when performing the continuous mode. In this work, we summarized three requirements task, proposed...
Irregularly Sampled Time Series (ISTS) include partially observed feature vectors caused by the lack of temporal alignment across dimensions and presence variable time intervals. Especially in medical applications, because patients' examinations depend on their health status, observations this event-based series are nonuniformly distributed. When using deep learning models to classify ISTS, most work defines problem that needs be solved as alignment-caused data missing or...
Time series widely exists in real-world applications and many deep learning models have performed well on it. Current research has shown the importance of strategy for models, suggesting that benefit is order size samples. However, no effective been proposed time due to its abstract dynamic construction. Meanwhile, existing one-shot tasks continuous necessitate distinct processes mechanisms. No all-purpose approach suggested. In this work, we propose a novel Curricular CyclicaL loss...
Machine Learning (ML) is becoming an increasingly critical technology in many areas. However, its complexity and frequent non-transparency create significant challenges, especially the biomedical health One of components addressing above challenges explainability or transparency ML systems, which refers to model (related whole data) sample specific samples). Our research focuses on both Random Forest (RF) classifiers. RF ex plainer, RFEX, designed from ground up with non-ML experts mind,...
Length of Stay (LoS) estimation is important for efficient healthcare resource management. Since the distribution LoS highly skewed, some previous works frame as a multi-class classification problem by dividing range into buckets. However, they ignore ordinal relationship between labels. The bucketed LoS, with heavy head and tail, still imbalanced since long tail grouped last bucket. This paper proposes Deep Ordinal neural network stay Estimation in intensive care units (DOSE). DOSE can...
Prediction based on Irregularly Sampled Time Series (ISTS) is of wide concern in the real-world applications. For more accurate prediction, methods had better grasp data characteristics. Different from ordinary time series, ISTS characterised with irregular intervals intra-series and different sampling rates inter-series. However, existing have suboptimal predictions due to artificially introducing new dependencies a series biasedly learning relations among when modeling these two In this...
Since data is presented long-tailed in reality, it challenging for Federated Learning (FL) to train across decentralized clients as practical applications. We present Global-Regularized Personalization (GRP-FED) tackle the imbalanced issue by considering a single global model and multiple local models each client. With adaptive aggregation, treats fairly mitigates issue. Each learned from aligns with its distribution customization. To prevent just overfitting, GRP-FED applies an adversarial...