A5101631903- Advanced Neural Network Applications
- Advanced Image and Video Retrieval Techniques
- Advanced Image Processing Techniques
- Image and Signal Denoising Methods
- Neural Networks and Applications
- Advanced Data Compression Techniques
- Domain Adaptation and Few-Shot Learning
- Advanced Image Fusion Techniques
- Robotics and Sensor-Based Localization
- Process Optimization and Integration
- Educational Technology and Assessment
- Multimodal Machine Learning Applications
- Face and Expression Recognition
- Evaluation Methods in Various Fields
- Advanced Combustion Engine Technologies
- Advanced Vision and Imaging
- Combustion and flame dynamics
- Allergic Rhinitis and Sensitization
- High voltage insulation and dielectric phenomena
- Ferroelectric and Negative Capacitance Devices
- Coding theory and cryptography
- Electrical Fault Detection and Protection
- Machine Learning and ELM
- Educational Technology and Pedagogy
- Graph Labeling and Dimension Problems
Shanghai Jiao Tong University
2020-2025
Sichuan University of Science and Engineering
2024
Northeastern University
2017
South China Agricultural University
2014
Tsinghua University
2012
Shanghai Electric Cable Research Institute
2012
Access
2012
Shenyang Institute of Engineering
2010-2011
Lanzhou Jiaotong University
2011
Shaanxi Normal University
2003
Image-text retrieval is a fundamental cross-modal task whose main idea to learn image-text matching. Generally, according whether there exist interactions during the process, existing methods can be classified into independent representation matching and cross-interaction methods. The generate embeddings of images sentences independently thus are convenient for with hand-crafted measures (e.g., cosine or Euclidean distance). As methods, they achieve improvement by introducing...
Model quantization is essential to deploy deep convolutional neural networks (DCNNs) on resource-constrained devices. In this article, we propose a general bitwidth assignment algorithm based theoretical analysis for efficient layerwise weight and activation of DCNNs. The proposed develops prediction model explicitly estimate the loss classification accuracy led by with geometrical approach. Consequently, dynamic programming adopted achieve optimal weights estimated error. Furthermore,...
Existing methods for integerized training speed up deep learning by using low-bitwidth weights, activations, gradients, and optimizer buffers. However, they overlook the issue of full-precision latent which consume excessive memory to accumulate gradient-based updates optimizing weights. In this paper, we propose first weight quantization schema general training, minimizes perturbation process via residual with optimized dual quantizer. We leverage eliminate correlation between suppressing...
Gas turbine health management is of great significant for enhancing flight safety and reducing maintenance costs. Based on the architecture gas fault identification system, a multi-arm spiral microphone array with high positioning accuracy, beam resolution, strong anti-interference capability designed in this paper simulation experiment. On basis, by utilizing multi-channel acoustic digital acquisition analysis equipment, high-precision, wide dynamic range, high-real-time audio data from...
Objective: To establish graded forecast models of pollen concentration in spring and summer-autumn northern China, based on long-term data allergic rhinitis (AR) medical visits 8 cities China. Methods: Pollen the characteristics AR patients from including Beijing, Baotou, Hohhot, Xi'an, Xining, Cangzhou, Liaocheng Zibo, were analyzed. Spearman's correlation was used to examine relationship between daily patient visits. A grading establish, a model created using eXtreme gradient boosting...
A new eigenvalue analysis-based method is presented for the construction of skeletal reduced mechanisms from complex chemical reaction mechanisms.A mechanism 21 species and 83 elementary reactions methane-air combustion was generated detailed GRI1.2.The ignition delay time, obtained different values equivalence ratio, initial temperature pressure on basis this mechanism, were compared with those based GRI1.2, another DRM19.The agreed favorably model, performed more accurately than DRM19.Two...
Scattering networks on Euclidean domains are capable of analytically realizing signal representation invariant to transformations such as translation, rotation and scaling with wavelets. However, existing scattering defined the sphere Riemannian manifolds only consider axisymmetric wavelets restricted in by isotropic filter structures. In this paper, we propose a novel anisotropic spherical network achieve multi-scale directional for signals. The transform is realized cascading spin modulus...
Multidimensional scaling (MDS) is an attractive technique for a moving source localisation from time and frequency difference of arrival (time differences (TDOA)/frequency (FDOA)) measurements. However, its optimality has not yet been proven theoretically because the difficult Moore–Penrose pesudo-inverse operation. In addition to theoretical incompleteness MDS technique, sensor uncertainties are considered in framework either. A closed-form estimator proposed TDOA/FDOA-based with senor by...
Network quantization can facilitate the practical deployment of U-Net in various medical applications, especially image segmentation. However, existing methods for do not sufficiently exploit skip connection between its encoder and decoder structures. In this paper, we propose a novel mixed-precision scheme that leverages split convolution skip-supervised aware training to address problem. Specifically, decouples feature maps obtained by up-sampling enable fine-grained bitwidth allocation...
Localization and landing are of fundamental importance for Unmanned Aerial Vehicles (UAVs), with the research on compute vision information processing, artificial landmark detection, a more efficient accurate method, becomes hot topic. In this paper, in order to provide sufficient localization, we design an annulus some circles inside specific positions colors. We firstly use color hierarchy find possible patter, then calculate its cross-ratio check pattern, localize our UAV at last. To...
The maculosus group of the genus Celaenorrhinus from China is reviewed, in which 7 species are treated. Taiwanese taiwanus reinstated to full rank. hitherto unknown female C. kuznetsovi, as well genitalia and major described. Adults illustrated, a brief description each given. A Key Chinese provided.
This work employed a hybrid approach combining computational fluid dynamics (CFD), artificial neural network (ANN), and genetic algorithm (GA) to optimize the non-dispersed solvent extraction (NDSX) process of H2O2. A CFD model with high accuracy was developed. The relative error between simulation results experimental data controlled within 2.2%. indicated velocity lumen side temperature are key factors affecting efficiency. feed-forward back-propagation ANN 20 hidden layer neurons trained...
Network quantization is promising to alleviate the storage and deployment burden for deep neural networks. However, existing methods focus on directly quantizing network parameters but ignore correlations between initialized (generated by random seeds) optimized (obtained when training converges). In this paper, we reveal redundancy propose a novel scheme with rescaled initialization. We develop delta parameters, i.e. , difference initialization, as shown in Figure 1 . The scale learned...
On-device computing, or edge is becoming increasingly important for remote sensing, particularly in applications like deep network-based perception on on-orbit satellites and unmanned aerial vehicles (UAVs). In these scenarios, two brain-like capabilities are crucial sensing models: (1) high energy efficiency, allowing the model to operate devices with limited computing resources, (2) online adaptation, enabling quickly adapt environmental variations, weather changes, sensor drift. This work...
The Mesoproterozoic Era has long been considered a relatively stable, silent and even ‘boring' period in Earth history, during which the lithosphere was tectonically inactive. However, an increasing amount of evidence suggests that metamorphism magmatism were much more widespread intense than previously thought, implying this dynamic. Here, we report at c. 1.4 Ga along southeastern margin North China Craton eastern China. We conducted analyses using TESCAN Integrated Mineral Analyser (TIMA)...