- Advanced Neural Network Applications
- Advanced Memory and Neural Computing
- Advanced Vision and Imaging
- COVID-19 diagnosis using AI
- Visual perception and processing mechanisms
- Medical Image Segmentation Techniques
- Image Enhancement Techniques
- Indoor and Outdoor Localization Technologies
- Multilevel Inverters and Converters
- Advanced DC-DC Converters
- Domain Adaptation and Few-Shot Learning
- Supercapacitor Materials and Fabrication
- Adversarial Robustness in Machine Learning
- Anomaly Detection Techniques and Applications
- Power System Optimization and Stability
- Ferroelectric and Negative Capacitance Devices
- CCD and CMOS Imaging Sensors
- Lung Cancer Diagnosis and Treatment
- Machine Learning and ELM
- Neural Networks and Reservoir Computing
- Neuroscience and Neural Engineering
- Advanced Image Processing Techniques
- Power Systems Fault Detection
- Photonic and Optical Devices
- Cerebrovascular and Carotid Artery Diseases
Huazhong Agricultural University
2025
Qiqihar Medical University
2024
Suzhou Research Institute
2024
Shanghai Jiao Tong University
2015-2023
Ruijin Hospital
2023
China University of Geosciences
2020-2023
Shandong Institute of Automation
2023
Ministry of Education of the People's Republic of China
2023
Sichuan University
2019-2022
Guangdong University of Technology
2021-2022
Researches on deep neural networks with discrete parameters and their deployment in embedded systems have been active promising topics. Although previous works successfully reduced precision inference, transferring both training inference processes to low-bitwidth integers has not demonstrated simultaneously. In this work, we develop a new method termed as "WAGE" discretize where weights (W), activations (A), gradients (G) errors (E) among layers are shifted linearly constrained integers. To...
Batch Normalization (BN) has been proven to be quite effective at accelerating and improving the training of deep neural networks (DNNs). However, BN brings additional computation, consumes more memory generally slows down process by a large margin, which aggravates effort. Furthermore, nonlinear square root operations in also impede low bit-width quantization techniques, draws much attention learning hardware community. In this work, we propose an L1-norm (L1BN) with only linear both...
Due to the strict requirements of extremely high accuracy and fast computational speed, real-time transient stability assessment (TSA) has always been a tough problem in power system analysis. Fortunately, development artificial intelligence big data technologies provide new prospective methods this issue, there have some successful trials on using intelligent method, such as support vector machine (SVM) method. However, traditional SVM method cannot avoid false classification,...
The real-time transient stability assessment (TSA) and emergency control are effective measures to suppress accident expansion, prevent system instability, avoid large-scale power outages in the event of failure. However, is extremely demanding on computing speed, traditional method not competent. In this paper, an improved deep belief network (DBN) proposed for fast stability, which considers structural characteristics construction loss function. Deep learning has been many fields, but...
Stable and safe operation of power grids is an important guarantee for economy development. Support Vector Machine (SVM) based stability analysis method a significant started in the last century. However, SVM has several drawbacks, e.g. low accuracy around hyperplane heavy computational burden when dealing with large amount data. To tackle above problems model, algorithm proposed this paper optimized from three aspects. Firstly, gray area model judged by probability output corresponding...
Neural networks trained by backpropagation have achieved tremendous successes on numerous intelligent tasks. However, naïve gradient-based training and updating methods memristors impede applications due to intrinsic material properties. Here, we built a 39 nm 1 Gb phase change memory (PCM) memristor array quantified the unique resistance drift effect. On this basis, spontaneous sparse learning (SSL) scheme that leverages improve PCM-based network is developed. During training, SSL regards...
This study aimed to develop a deep learning (DL) model improve the diagnostic performance of EIC and ASPECTS in acute ischemic stroke (AIS).Acute patients were retrospectively enrolled from 5 hospitals. We proposed simultaneously segment infarct estimate automatically using baseline CT. The segmentation scoring was evaluated dice similarity coefficient (DSC) ROC, respectively. Four raters participated multi-reader multicenter (MRMC) experiment fulfill region-based reading under assistance or...
Compact convolutional neural networks gain efficiency mainly through depthwise convolutions, expanded channels and complex topologies, which contrarily aggravate the training process. Besides, 3x3 kernels dominate spatial representation in these models, whereas even-sized (2x2, 4x4) are rarely adopted. In this work, we quantify shift problem occurs kernel convolutions by an information erosion hypothesis, eliminate it proposing symmetric padding on four sides of feature maps (C2sp, C4sp)....
Using a prior data to solve for the characteristic parameters of moving magnetic target is crucial in its detection. For this purpose, we developed real-time detection method using distributed scalar sensor networks based on hybrid algorithm combining particle swarm optimization and Gauss-Newton method. The anomaly fitting model established magnetic-dipole-moment principle. PSO’s insensitivity initial solution used obtain rough coefficients fitting, then more accurate method, exploiting...
Effectively leveraging multimodal data such as various images, laboratory tests and clinical information is becoming increasingly attractive in a variety of AI-based medical diagnosis prognosis tasks. Most existing multi-modal techniques only focus on enhancing their performance by the differences or shared features from modalities fusing feature across different modalities. These approaches are generally not optimal for settings, which pose additional challenges limited training data, well...
Abstract Recent years have witnessed tremendous progress of intelligent robots brought about by mimicking human intelligence. However, current are still far from being able to handle multiple tasks in a dynamic environment as efficiently humans. To cope with complexity and variability, further toward scalability adaptability essential for robots. Here, we report brain-inspired robotic platform implemented an unmanned bicycle that exhibits network scale, quantity diversity the changing needs...
Magnetic anomaly detection technologies have been widely used for tracking moving targets. In this paper, we present a fast-tracking method magnetic abnormalities using distributed Overhauser magnetometer system based on the genetic algorithm. Our proposed framework of employs multiple sensors to eliminate background interference, and algorithm efficiently solves data without requiring derivation objective function. Test platforms were built evaluate Results from natural outdoor magnetism...
Mining data streams has attracted much attention recently. Labeled samples needed by most current stream classification methods are more difficult and expensive to obtain than unlabeled ones. This paper proposed a semi-supervised learning algorithm - clustering-training utilize the samples. It uses clustering select confidently samples, them re-train classifier incrementally. Experiments on synthetic real set showed effectiveness of