Shuyue Jia 贾舒越 Blog

Bio

I am a first-year M.Phil. student in Computer Science at City University of Hong Kong, where I am fortunate to be advised by Dr. Shiqi Wang. In summer 2017, I attended a summer school at University of California, Irvine in CA, USA. Prior to joining CityU, I received my bachelor's degree at Northeast Electric Power University in Jilin, China, supervised by Prof. Yimin Hou.

My research focuses on Computer Vision and Machine Learning, and my current concentration is Image and Video Quality Assessment and Perceptual Optimization.

Some of my paper survey and presentations can be found

here

Graph Neural Network (GNN)

  • Dynamic Graph Convolutional Neural Networks Survey
  • Graph Convolutional Neural Networks (Chebyshev Approximation)

    • Natural Language Processing (NLP)

  • Graph Matching
  • Sub-word BPE Algorithm for NMT
  • Concept Matching for Medical Terms

    • Computer Vision (CV)

  • 3D Human Pose Estimation and Human Body Reconstruction
  • YOLO Object Detection Algorithm

    • Other Tutorials

  • Usage of Cloud Server and Setting-up
  • Python Environment Setting-up
  • TensorFlow for Deep Learning

    • Email: shuyuej@ieee.org or shuyue.jia@my.cityu.edu.hk
    Scholar   Résumé   GitHub  

    News

    Publications

    1. A Novel Approach of Decoding EEG Four-Class Motor Imagery Tasks via Scout ESI and CNN
      Codes Paper

      Yimin Hou, Lu Zhou, Shuyue Jia, and Xiangmin Lun
      Journal of Neural Engineering      , 2020; 17(1):016048.
      See More
        We presented a novel approach that could potentially be used to improve the current stroke rehabilitation strategies by implementing a deep learning approach for an Electroencephalogram (EEG) based on MI Brain-Computer Interface System.
        • Constructed 6 convolutional layers, 2 max-pooling layers, and 3 FC layers CNNs for four-class motor imagery classification through TensorFlow, with 50% dropout (spatial dropout after every Conv layer and regular dropout for FC layers) – 11.44% accuracy improvement, batch normalization (BN) – 10.15% improvement, and Short-cut Connection – 1.76% improvement to prevent overfitting, and achieved SOTA results: 94.50% accuracy on scout R5, 94.54% at subject level, and 96% for left fist prediction.
        • Took charge of DNNs design, including methods comparisons, such as MLPs, CNNs, RNNs, LSTMs, and Autoencoders, classification results calculations, and programming. 10 and 14 subjects’ data were utilized (19,320 and 27,048 samples in the experiments)
        • Benchmark Dataset: EEG Motor Movement/Imagery Dataset.
        EEG-CNN
    2. GCNs-Net: A Graph Convolutional Neural Network Approach for Decoding Time-resolved EEG Motor Imagery Signals

      Xiangmin Lun, Shuyue Jia *, Yimin Hou, Yan Shi, and Yang Li
      Presentation Survey

      arXiv preprint arXiv:2006.08924, 2020.
      Codes Paper


      See More
        Based on the Graph Convolutional Neural Network (Graph CNN / GCN), the GCNs-Net was introduced, which filtered the EEG Motor Imagery (MI) signals considering the functional topological relationship of EEG electrodes.
        • At the subject and group level (subject-specific adaptation), 98.72% and 89.387% accuracy were achieved, respectively.
        • Pearson’s Matrix was applied to measure the correlations among channels, and represented the graph structure, i.e., graph weights and degrees.
        • Benchmark Datasets: EEG Motor Movement/Imagery Dataset, in which 20 ( a million samples), 50, 100 participants’ data were used, and the High-Gamma Dataset, in which 14 participates' data were used.
        Project2
    3. Deep Feature Mining via Attention-based BiLSTM-GCN for Human Motor Imagery Recognition

      Yimin Hou, Shuyue Jia *, Xiangmin Lun, Yan Shi, and Yang Li
      Codes Paper

      arXiv preprint arXiv:2005.00777      , 2020.
      See More
        We introduced a novel approach which combined Attention-based BiLSTM with the Graph Convolutional Neural Network (Graph CNN / GCN). The method has achieved SOTA results.
        • Open Source EEG-DL, a Deep Learning (DL) library written by TensorFlow for EEG Tasks (Signals) Classification.
        • Attention-based BiLSTM was firstly used to extract features from raw EEG signals. The followed GCN model classified the features regarding to four EEG Motor Imagery (MI) tasks, imagining left fist, right fist, both fists, and both feet.
        • 98.81% and 94.64% accuracies have been achieved for the individual subject and a group of 20 subjects.
        • Benchmark Dataset: EEG Motor Movement/Imagery Dataset.
        Project3.1 Project3.2
    4. Attention-based Graph ResNet for Motor Intent Detection from Raw EEG signals

      Shuyue Jia, Yimin Hou, Yan Shi, and Yang Li
      Codes Paper

      arXiv preprint arXiv:2007.13484      , 2020.
    5. Origin Traceability of Rice based on an Electronic Nose coupled with a Feature Reduction Strategy

      Yan Shi, Xiaofei Jia, Hangcheng Yuan, Shuyue Jia, Jingjing Liu, and Hong Men
      Paper

      Measurement Science and Technology      , 2020; 32(2):025107.
    6. Improving Performance: a Collaborative Strategy for the Multi-data Fusion of Electronic Nose and Hyperspectral to Track the Quality Difference of Rice

      Yan Shi, Hangcheng Yuan, Chenao Xiong, Shuyue Jia, Jingjing Liu, and Hong Men
      Paper

      Sensors & Actuators: B. Chemical      , 2021; 129546.

    Academic Services

    1. Reviewer for IEEE Journal of Biomedical and Health Informatics
    2. Student Member of IEEE, ACM, and CCF

    Selected Awards

    1. 2019 Interdisciplinary Contest In Modeling
      Thesis

      Honorable Mention
      ICM-2019
    2. 2018 Mathematical Contest In Modeling (Jilin, China)
      Thesis

      First Prize
      MCM
    3. 2018 Interdisciplinary Contest In Modeling
      Thesis

      Successful Participant
      ICM-2018