A5100698859- Adversarial Robustness in Machine Learning
- Anomaly Detection Techniques and Applications
- Machine Learning and Algorithms
- Fault Detection and Control Systems
- Video Surveillance and Tracking Methods
- Advanced Neural Network Applications
- Robotics and Sensor-Based Localization
- Innovative Teaching and Learning Methods
- Advanced Image and Video Retrieval Techniques
- Face recognition and analysis
- Advanced Vision and Imaging
- Data-Driven Disease Surveillance
- Biometric Identification and Security
- Statistical Methods and Bayesian Inference
- Subtitles and Audiovisual Media
- Mental Health via Writing
- Formal Methods in Verification
- Estrogen and related hormone effects
- Marine and fisheries research
- Cardiac Imaging and Diagnostics
- Advanced Control Systems Optimization
- Digital Media and Visual Art
- Machine Learning in Healthcare
- Online Learning and Analytics
- Ginseng Biological Effects and Applications
University of Liverpool
2014-2025
Tianjin University
2024
South China University of Technology
2023
Shanghai Medical College of Fudan University
2023
Fudan University
2023
Tianjin University of Technology
2023
Obstetrics and Gynecology Hospital of Fudan University
2023
University of Toronto
2021
University of Oklahoma
2021
Beijing Institute of Technology
2021
Facial images have extensive practical applications. Although the current large-scale text-image diffusion models exhibit strong generation capabilities, it is challenging to generate desired facial using only text prompt. Image prompts are a logical choice. However, methods of this type generally focus on general domain. In paper, we aim optimize image makeup techniques images. Specifically, (1) built dataset 4 million high-quality face image-text pairs (FaceCaptionHQ-4M) based LAION-Face...
Identifying illegal fishing activities from Automatic Identification System (AIS) data is difficult since AIS messages are broadcast cooperatively, the ship’s master controls timing, and content of transmission interest usually occur far away shore. This paper presents our work to predict ship types using satellites: in such data, there a pronounced imbalance between for different ships, refresh rate relatively low, misreporting information. To mitigate these issues, prediction algorithm...
Depression is a common mental health condition that often occurs in association with other chronic illnesses, and varies considerably severity. Electronic Health Records (EHRs) contain rich information about patient's medical history can be used to train, test maintain predictive models support improve patient care. This work evaluated the feasibility of implementing an environment for predicting crisis among people living depression based on both structured unstructured EHRs. A large EHR...
This paper proposes an approach to detect moving objects in Wide Area Motion Imagery (WAMI), which the are both small and well separated. Identifying only using foreground appearance is difficult since a 100–pixel vehicle hard distinguish from comprising background. Our based on background subtraction as efficient unsupervised method that able output shape of objects. In order reliably low contrast objects, we configure extract regions might be interest. While this dramatically increases...
Intestinal microbiota is the primary target for multifunctional nature of berberine. Berberine can reverse structure and composition gut under pathological conditions. This study aimed to investigate effects berberine on uric acid (UA) metabolism in a hyperuricemia rat model established using potassium oxonate.Sprague-Dawley (SD) male rats were divided into normal control group (n= 10), (n = 12) berberine-treated 11). The UA level serum, urine fecal, blood xanthine oxidase (XOD), urate...
This paper proposes a moving object detection algorithm which can handle videos taken by camera in the presence of pronounced parallax. The considers idea that objects image be considered to spatially distributed across multiple planes, movement each estimated using Visual Odometry (VO) algorithm. For plane, Homography matrix between consecutive frames is calculated facilitate motion compensation. approach also identifies regions where parallax and most recent data are then assumed provide...
This paper studies the reliability of a real-world learning-enabled system, which conducts dynamic vehicle tracking based on high-resolution wide-area motion imagery input. The system consists multiple neural network components - to process inputs and symbolic (Kalman filter) analyse processed information for tracking. It is known that networks suffer from adversarial examples, make them lack robustness. However, it unclear if how examples over learning can affect overall system-level...
Cardiovascular disease caused by atherosclerosis is the leading cause of death in developed world. Narrowing vessel lumen, due to atherosclerotic plaque development or rupturing established plaques, interrupts normal blood flow various morbidities such as myocardial infarction and stroke. In clinic endovascular procedures angioplasty are commonly performed reopen lumen. However, these treatments inevitably damage wall well vascular endothelium, triggering an excessive healing response a...
We study for the first time verification problem on learning-enabled state estimation systems robotics, which use Bayes filter localisation, and deep neural network to process sensory input into observations filter. Specifically, we are interested in a robustness property of systems: given certain ability an adversary it attack without being noticed, whether or not system is able function with only minor loss localisation precision? For purposes, reduce novel class labelled transition...
This paper proposes an improvement to FastSLAM. The approach is applicable when the dynamic model describing motion of camera has linear sub-structure. core novelty proposed algorithm separate consideration camera's into two sub-models without constraining have independent noise processes. In contrast commonly-used FastSLAM algorithms, which use a particle filter consider both these sub-models, used for one sub-model and Kalman other. tactic Rao-Blackwellisation same as that underpins...
We aim to examine and reestablish the correlational linear regression relationships, as well predictive value, between significant facial tongue features metabolic parameters in type 2 diabetes mellitus (T2DM).
This paper proposes a new generic method to detect anomalies (i.e., statistical outliers) which can be used with generative topic model. In this paper, we specify in the context of Mixture-of-unigrams model, is widely text mining. It possible model by applying threshold likelihood. However, it challenging choose since choice needs consider both similarities topics and length documents. describes intuitive simply manipulates output trained Such an approach anticipated have parameters that are...
Cotreatment with metformin and Diane-35 is conventionally used in the clinic to ameliorate ovulatory dysfunction insulin resistance women polycystic ovary syndrome (PCOS). We previously showed that this combination treatment could reverse endometrial hyperplasia cancer (EC) patients PCOS. Here, we aimed investigate influence of dihydrotestosterone (DHT) cotreatment on endometrium, along related mechanisms.We treated a DHT-exposed rat model or alone their investigated 3-dimensional (3D)...
This paper studies the reliability of a real-world learning-enabled system, which conducts dynamic vehicle tracking based on high-resolution wide-area motion imagery input. The system consists multiple neural network components -- to process inputs and symbolic (Kalman filter) analyse processed information for tracking. It is known that networks suffer from adversarial examples, make them lack robustness. However, it unclear if how examples over learning can affect overall system-level...
Introduction: Xiyanping injection (XYP), a type of Traditional Chinese Medicine, is widely used and often applied in combination with other medications treating bronchitis, tonsillitis, bacillary dysentery China. In recent years, an elevated risk allergic reactions has been observed following XYP, but whether concomitant medication use contributes to this still unknown. Objective: This study aims investigate the association between XYP 25 most frequently co-applied suspected for China's...
When face and facial part detectors encounter partial or complete occlusion, their output becomes increasingly noisy, making it difficult for such algorithms to tackle complex scenes: parts are only detected intermittently, false alarms become more prevalent, mis-identified, some consistently missed altogether. In this paper we propose a remedy problem: probabilistic fusion of detectors. The method jointly improves the localization tracking performance by modelling spatial temporal...
This paper presents a formal verification guided approach for principled design and implementation of robust resilient learning-enabled systems. We focus on state estimation systems (LE-SESs), which have been widely used in robotics applications to determine the current (e.g., location, speed, direction, etc.) complex system. The LE-SESs are networked composed set connected components including: Bayes filters localisation, neural networks processing sensory input. study from perspective...
This paper proposes an approach to detect moving objects in Wide Area Motion Imagery (WAMI), which the are both small and well separated. Identifying only using foreground appearance is difficult since a $100-$pixel vehicle hard distinguish from comprising background. Our based on background subtraction as efficient unsupervised method that able output shape of objects. In order reliably low contrast objects, we configure extract regions might be interest. While this dramatically increases...