A5100624436- Video Surveillance and Tracking Methods
- AI in cancer detection
- Advanced Neural Network Applications
- Face and Expression Recognition
- Face recognition and analysis
- Image Retrieval and Classification Techniques
- Advanced Data Compression Techniques
- Long-Term Effects of COVID-19
- Machine Learning and Data Classification
- SARS-CoV-2 and COVID-19 Research
- COVID-19 Clinical Research Studies
- Image and Signal Denoising Methods
- Radiomics and Machine Learning in Medical Imaging
- Hand Gesture Recognition Systems
- Neural Networks and Applications
- Advanced Image and Video Retrieval Techniques
- Digital Media Forensic Detection
- Biometric Identification and Security
- Caching and Content Delivery
- Advanced Steganography and Watermarking Techniques
- Hepatitis B Virus Studies
- Human Pose and Action Recognition
- Higher Education and Teaching Methods
- Robotics and Sensor-Based Localization
- Fluorine in Organic Chemistry
Taipei Veterans General Hospital
2025
National Yang Ming Chiao Tung University
2024-2025
National Central University
2024
Guizhou Provincial People's Hospital
2024
Zunyi Medical University
2024
Massachusetts Institute of Technology
2022-2023
Xiamen University
2020-2022
Gannan Medical University
2022
First Affiliated Hospital of Fujian Medical University
2022
Fujian Medical University
2022
Machine learning on tiny IoT devices based microcontroller units (MCU) is appealing but challenging: the memory of microcontrollers 2-3 orders magnitude smaller even than mobile phones. We propose MCUNet, a framework that jointly designs efficient neural architecture (TinyNAS) and lightweight inference engine (TinyEngine), enabling ImageNet-scale microcontrollers. TinyNAS adopts two-stage search approach first optimizes space to fit resource constraints, then specializes network in optimized...
Pose estimation plays a critical role in human-centered vision applications. However, it is difficult to deploy state-of-the-art HRNet-based pose models on resource-constrained edge devices due the high computational cost (more than 150 GMACs per frame). In this paper, we study efficient architecture design for real-time multi-person edge. We reveal that HRNet's high-resolution branches are redundant at low-computation region via our gradual shrinking experiments. Removing them improves both...
•Association of socioeconomic development status with burden AVH was identified.•Lowest incidence and DALY rate noted for high- high-middle SDI location.•Highest low-middle- low locations.•Our findings may benefit policymakers in allocating resources. Background & AimsAcute viral hepatitis (AVH) represents an important global health problem; however, the progress understanding is limited because priority combating persistent HBV HCV infections. Therefore, improved required to help design...
On-device training enables the model to adapt new data collected from sensors by fine-tuning a pre-trained model. Users can benefit customized AI models without having transfer cloud, protecting privacy. However, memory consumption is prohibitive for IoT devices that have tiny resources. We propose an algorithm-system co-design framework make on-device possible with only 256KB of memory. faces two unique challenges: (1) quantized graphs neural networks are hard optimize due low bit-precision...
Tiny Machine Learning (TinyML) is a new frontier of machine learning. By squeezing deep learning models into billions IoT devices and microcontrollers (MCUs), we expand the scope AI applications enable ubiquitous intelligence. However, TinyML challenging due to hardware constraints: tiny memory resource makes it difficult hold designed for cloud mobile platforms. There also limited compiler inference engine support bare-metal devices. Therefore, need co-design algorithm system stack TinyML....
Tiny deep learning on microcontroller units (MCUs) is challenging due to the limited memory size. We find that bottleneck imbalanced distribution in convolutional neural network (CNN) designs: first several blocks have an order of magnitude larger usage than rest network. To alleviate this issue, we propose a generic patch-by-patch inference scheduling, which operates only small spatial region feature map and significantly cuts down peak memory. However, naive implementation brings...
A practical and facile difluorocarbene-triggered cycloaddition reaction of enaminones was developed, which delivered 2,2-difluoro-2,3-dihydrofurans without any extra additives.
Vector quantization (VQ) is an effective image coding technique at low bit rate. The side-match finite-state vector quantizer (SMVQ) exploits the correlations between neighboring blocks (vectors) to avoid large gray level transition across block boundaries. A new adaptive edge-based classified (classified FSVQ) with a quadtree map has been proposed. In FSVQ, are arranged into two main classes, edge and nonedge blocks, selecting wrong state codebook for input block. order improve quality,...
We aim to develop a diagnostic tool for pathological-image classification using transfer learning that can be applied diverse tumor types.Microscopic images of liver tissue with and without hepatocellular carcinoma (HCC) were used train validate the framework based on convolutional neural network. To evaluate universal performance artificial intelligence (AI) framework, histological from colorectal breast collected. Images training validation sets obtained Xiamen Hospital Traditional Chinese...
Serum prostate-specific antigen (PSA) is a primary metric for diagnosis and prognosis of prostate cancer (PCa). Exposure to heavy metals, such as lead, cadmium, mercury, zinc can impact PSA levels in PCa patients. However, it unclear whether this effect also occurs men without PCa, which may lead the overdiagnosis PCa.
Although abnormal liver chemistries are linked to a higher risk of coronavirus disease 2019 (COVID-19)–related death, manifestations may be diverse and even confusing. Thus, we performed meta-analysis published described the damage in patients with COVID-19 who died or discharged alive. We searched PubMed, Google Scholar, medRxiv, bioRxiv, Cochrane Library, Embase, three Chinese electronic databases through April 22, 2020. analyzed pooled data on stratified by main clinical outcome COVID-19,...
With the rapid development of sensor technologies, in nowadays various types sensors are deployed around us as a basis to build smart environment. It is without doubt that how deal with huge amount data collected by an efficient way and transform these useful information for people make use has become important research topic. The vast produced such temperature, road conditions air quality can be numerically analyzed utilizing cloud computing technology. In this research, we implemented...
This paper proposes a Chinese medical question answering system based on CNN s with self-attention embedded model. The key idea of the proposed framework is first understanding whole sentences via LSTM, obtaining feature map by convolution layer CNN, then further-comprehending and answer finally pass through to polling enhance accuracy. In addition, due word segmentation issue in especially field, character embedding applied experimental results show that improves accuracy around 80%.
With the development of times, people's lifestyle and eating habits are changed a lot. Worldwide, breast cancer is second most common type after lung fifth cause death. Diagnostic ultrasound (US) currently major clinical detection method. However, imaging usually contains large number noises speckles. That will impact greatly on diagnosis by physicians. Therefore, we proposed method to enhance computer-aided (CAD) tumors reduce time error rate. Experimental investigations demonstrated that...
Quantum target detection (QTD) utilizes nonclassical resources to enable radar-like for identifying reflecting objects in lossy and noisy environments, surpassing the performance achieved by classical methods. To fully exploit quantum advantage QTD, determining optimal probe states (OPSs) across various parameters gaining a deeper understanding of their characteristics are crucial. In this study, we employ optimization algorithms identify single-mode continuous-variable OPSs entire range...
In this paper, we propose an efficient method for image tamper detection and recovery.We separate the into several blocks, share a block information to two other blocks.It means, there have copies of one each non-overlapping block.In implementing, improve Lee's algorithm watermark embedding, especially when tampered area is rounded region or text field, our scheme has higher PSNR value than method.During recovery, use inpainting approach fill whole image.The experimental result shows that...
800x600 Normal 0 2 false EN-US ZH-TW X-NONE MicrosoftInternetExplorer4 With the development of times, people's lifestyle and eating habits are changed a lot. Worldwide, breast cancer is second most common type after lung fifth cause death. Diagnostic ultrasound (US) currently major clinical detection method. It recognized as safe, effective, highly flexible imaging modality capable providing clinically relevant information about parts body in rapid cost-effective fashion. However, usually...
This paper will discuss how to track the faces of patients who only lie in bed real-time. We propose a realtime mechanism extract and patients' facial features effectively. When camera facing patient's face, we use our DRK (Dynamic Radial Kernel) record match features. Then collect information face. It could be provided doctors for diagnostic aiding. For patients, this can bring as well warning function. hope that provide good feature health care.
The practical and facile difluorocarbene-triggered cycloaddition reaction of enaminones was developed, that delivered 2,2-difluoro-2,3-dihydrofurans with broad substrates scope. Notably, the proceeded smoothly without any extra additive. readily available sodium chlorodifluoroacetate (ClCF2CO2Na, SCDA) played as difluorocarbene precursor in this transformation through DMF-promotion. Moreover, it is proved 2,2-difluoro- 2,3-dihydrofuran derivatives exposed potential antiproliferative activity...