State College, Pennsylvania, United States
I have 1.5 years of working experience in deep learning and computer vision before I started to study for a master's degree in 2019. From 2019 to 2021, I worked with Dr. James Z. Wang and Dr. Jia Li for explainable image feature representation learning and applications of DL/CV in real-world problems. In the year 2020, I am lucky to publish three papers in top-tier conferences including ACM and ECCV. I am experienced in image/video-based human emotion analysis. Moreover, I have two papers on image segmentation and graph neural networks that are under review. Specialties: Deep Learning, Computer Vision, Data Mining and Visualization, Machine Learning.
1. Semantic segmentation in stereo images. 2. Model optimization.
Teaching assistant for two courses: Data Science DS 340W.003 with Prof. Kaamran Raahemifar Data Science DS 340W.004 with Prof. James Z. Wang
Currently working on two research projects: 1. Optimal transport with Wasserstein distance. 2. Extent of corrosion image analysis.
Teaching assistant. Course: Applied Data Science Instructor: Dr. James Wang
1. TCT (thinprep cytologic test) Project: cancer diagnosis from biomedical images • Designed and implemented an end-to-end deep neural network model. Incorporated object detection, classification and segmentation within the model and achieved above 85% accuracy in 13 disease categories. • Modified and optimized existing object detection models to accommodate the medical images dataset. Extended detection branch and added external classification branch to YoloV3 model to capture small objects and make use of unlabeled data, which achieved 10% more recall over the baseline model. • Deployed a high-performance computing platform with stable 70% utilization rate over distributed, multiple GPUs at prediction, and introduced by Southern Hospital of China. 2. TMB (tumor mutation burden) Project: mutation count prediction from biomedical images • Updated the data preprocessing module of Keras and greatly improved the training efficiency by 30%. Developed a tool for manipulating the lmdb files used for Caffe that enabled append/insert/delete/replace operations. • Jointly trained three CNN models to extract features of individual images sliced out of whole slide images. • Designed and developed two levels of feature clustering and PCA strategies to generate final features for the big images and based on which achieved about 80% classification accuracy. 3. OCR (Optical Character Recognition) Project • Fabricated millions of synthetic text data in natural images using open-source tool SynthText. • Worked in a team to construct a detection & word recognition model that supported 0/90/180/270 degree orientations in both English and Chinese languages. 4. Malware Detection Competition • Won 1st prize on the preliminary of malware detection competition hosted by Qihoo 360. • Ensembled both classifiers like SVM or XGBoost on manually engineered features from malware generated XML files, and CNN classifiers like Xception on grayscale images mapped from bytecodes of XML files.