A5035943349- Advanced Neural Network Applications
- Domain Adaptation and Few-Shot Learning
- Radiomics and Machine Learning in Medical Imaging
- COVID-19 Digital Contact Tracing
- Machine Learning and Data Classification
- COVID-19 epidemiological studies
- Quantum Chromodynamics and Particle Interactions
- Adversarial Robustness in Machine Learning
- Video Surveillance and Tracking Methods
- COVID-19 diagnosis using AI
- Advanced Vision and Imaging
- Renal cell carcinoma treatment
- Anomaly Detection Techniques and Applications
- Generative Adversarial Networks and Image Synthesis
- Natural Language Processing Techniques
- Data-Driven Disease Surveillance
- Advanced Image and Video Retrieval Techniques
- Advanced Image Processing Techniques
- Topic Modeling
- High-Energy Particle Collisions Research
- Black Holes and Theoretical Physics
- Particle physics theoretical and experimental studies
- IoT and Edge/Fog Computing
- COVID-19 and Mental Health
- Multimodal Machine Learning Applications
Quzhou University
2021-2024
University of Electronic Science and Technology of China
2021-2024
East China Normal University
2023
Chengdu Medical College
2021-2022
China Three Gorges University
2022
Chongqing Jiaotong University
2021
South China University of Technology
2016
Diagnosing liver lesions is crucial for treatment choices and patient outcomes. This study develops an automatic diagnosis system using multiphase enhanced computed tomography (CT). A total of 4039 patients from six data centers are enrolled to develop Liver Lesion Network (LiLNet). LiLNet identifies focal lesions, including hepatocellular carcinoma (HCC), intrahepatic cholangiocarcinoma (ICC), metastatic tumors (MET), nodular hyperplasia (FNH), hemangioma (HEM), cysts (CYST). Validated in...
This retrospective study aimed to develop and validate deep-learning-based models for grading clear cell renal carcinoma (ccRCC) patients. A cohort enrolling 706 patients (n = 706) with pathologically verified ccRCC was used in this study. temporal split applied verify our models: the first 83.9% of cases (years 2010-2017) development last 16.1% (year 2018-2019) validation (development cohort: n 592; 114). Here, we demonstrated a deep learning(DL) framework initialized by self-supervised...
We employ the VQCD model, a holographic approach that dynamically simulates essential QCD characteristics, including linear mass spectra, confinement, asymptotic freedom, and magnetic charge screening, while incorporating quark flavor effects. Using this we first calculate proton spectrum wave function, anomalous dimensions to refine our results. Next, compute structure functions across range of Bjorken $x$ values using consistent parameters. Furthermore, derive electromagnetic form factor...
In this paper, we propose a novel data augmentation strategy named Cut-Thumbnail, that aims to improve the shape bias of network. We reduce an image certain size and replace random region original with reduced image. The generated not only retains most information but also has global in call as thumbnail. Furthermore, find idea thumbnail can be perfectly integrated Mixed Sample Data Augmentation, so put one image's on another while ground truth labels are mixed, making great achievements...
A novel coronavirus disease (COVID-19) is a pandemic has caused 4 million deaths and more than 200 infections worldwide (as of August 4, 2021). Rapid accurate diagnosis COVID-19 infection critical to controlling the spread epidemic. In order quickly efficiently detect reduce threat human survival, we have firstly proposed detection framework based on reinforcement learning for diagnosis, which constructs mixed loss function that can integrate advantages multiple functions. This paper uses...
In a convolutional neural network (CNN), dropout cannot work well because dropped information is not entirely obscured in layers where features are correlated spatially. Except for randomly discarding regions or channels, many approaches try to overcome this defect by dropping influential units. paper, we propose non-random method named FocusedDropout, aiming make the focus more on target. use simple but effective search target-related features, retain these and discard others, which...
Generative Adversarial Networks(GANs) are powerful generative models on numerous tasks and datasets but also known for their training instability mode collapse. The latter is because the optimal transportation map discontinuous, DNNs can only approximate continuous ones. One way to solve problem introduce multiple discriminators or generators. However, impacts limited cost function of each component same. That is, they homogeneous. In contrast, with different functions yield various...
Background Clear cell Renal Cell Carcinoma (ccRCC) is the most common malignant tumor in urinary system and predominant subtype of renal tumors with high mortality. Biopsy main examination to determine ccRCC grade, but it can lead unavoidable complications sampling bias. Therefore, non-invasive technology (e.g., CT examination) for grading attracting more attention. However, noise labels on images containing multiple grades only one label make prediction difficult. exist images, which...
We assume that the initial hydrodynamic environment is a quark gluon plasma (QGP) phase where merely $u$ and $d$ quarks are considered when adding magnetic field. When considering chiral effect in relativistic heavy ion collisions, an anomalous current will be formed QGP environment. The by these has impact on quarkonium. By using fluid/gravity duality, metric with flow established so as to introduce field into corresponding metric. And then we use quarkonium probe study transition, utilize...
Abstract Over-fitting is a significant threat to the integrity and reliability of deep neural networks with generous parameters. One problem that model learned more specific features than general in training process. To solve problem, we propose an adversarial method assist strengthening representation learning. In this method, make classification as generator G introduce unsupervised discriminator D distinguish hidden feature from real images limit their spatial distance. Notably, will fall...
Introduction: Close contacts have become a potential threat to the spread of coronavirus disease 2019 (COVID-19). The purpose this study was understand epidemiological characteristics close confirmed or suspected cases COVID-19 in surrounding cities Chengdu, China, so as provide basis for management strategy contacts. Methods: were determined through investigation indicated cases, and relevant information entered “Close Contact Information Management System.” Retrospective data from January...
With the development of novel coronavirus disease 2019 (COVID-19) epidemic and increase in cases, as a potential source infection, risk close contact has gradually increased. However, few studies have analyzed tracking management cross-regional personnel. In this study, we hope to understand effectiveness feasibility existing measures Chengdu, so provide reference for further prevention control epidemic. The mode epidemiological characteristics 40,425 contacts from January 22, 2020, March 1,...
Cross-lingual representation learning transfers knowledge from resource-rich data to resource-scarce ones improve the semantic understanding abilities of different languages. However, previous works rely on shallow unsupervised generated by token surface matching, regardless global context-aware semantics surrounding text tokens. In this paper, we propose an Unsupervised Pseudo Semantic Data Augmentation (UniPSDA) mechanism for cross-lingual natural language enrich training without human...
Motion blur is pervasive in object tracking due to the camera and target movement. Most approaches are prone drift when blurred. Sparse representation of both normal templates able improve robustness appearance model, but how update template set remains a difficult problem. To solve this problem, we propose two-step observation correction strategy for updating by: 1) treating motion as trivial information with Laplace distribution using different ways correct target; 2) utilizing incremental...
Generative Adversarial Networks (GANs) are powerful generative models for numerous tasks and datasets. However, most of the existing suffer from mode collapse. The recent research indicates that reason it is optimal transportation map random noise to data distribution discontinuous, but deep neural networks (DNNs) can only approximate continuous ones. Instead, latent representation a better raw material used construct point than noise. Because low-dimensional mapping related distribution,...
Image resolution that has close relations with accuracy and computational cost plays a pivotal role in network training. In this paper, we observe the reduced image retains relatively complete shape semantics but loses extensive texture information. Inspired by consistency of as well fragility information, propose novel training strategy named Temporally Resolution Decrement. Wherein, randomly reduce images to smaller time domain. During alternate original images, unstable information...
In recent years, various loss functions have been proposed to boost the performance of deep neural networks. Every function has its own specific theoretical motivation, and can easily learn preference features training data compared with other functions. Thus, combining multiple capture more becomes an attractive idea for model improvement. this paper, instead using a single or linear weighted sum functions, we present method named Multiple Independent Losses Scheduling (MILS), which allows...
Purpose: By leveraging deep learning (DL) techniques, our objective was to explore and uncover the latent potential within image diversity by improving performance of a prognostic model for advanced nasopharyngeal carcinoma (NPC, stage III-IVA ). This particular relies on pre-treatment magnetic resonance imaging (MRI) data.Methods: A retrospective study conducted at West China Hospital Sichuan University, where MRI data 312 patients with NPC were collected. To unveil diversity, different...
Abstract As a potential source of infection, the risk close contacts has gradually increased. Close contact management mode and epidemiological characteristics for 20,254 from January 22, 2020 to April 1, 2021 in Chengdu, China were analyzed. The relationship with index cases was mainly co-passengers (73.52%) relatives (13.64%), frequency occasional (68.31%). 277 (1.37%) who converted into found first second nucleic acid tests (58.27%), by sharing transportation (37.55%). In terms time, both...
In this paper, we propose Selective Output Smoothing Regularization, a novel regularization method for training the Convolutional Neural Networks (CNNs). Inspired by diverse effects on from different samples, Regularization improves performance encouraging model to produce equal logits incorrect classes when dealing with samples that classifies correctly and over-confidently. This plug-and-play can be conveniently incorporated into almost any CNN-based project without extra hassle. Extensive...
Regularization and data augmentation methods have been widely used become increasingly indispensable in deep learning training. Researchers who devote themselves to this considered various possibilities. But so far, there has little discussion about regularizing outputs of the model. This paper begins with empirical observations that better performances are significantly associated output distributions, smaller average values variances. By audaciously assuming is causality involved, we...