Han Yan Email & Phone Number

• Extract, clean, and organize data of job descriptions and resumes from Indeed.com for training and analysis
• Develop a GUI tool for NER (Name Entity Recognition) tagging tasks, which greatly improves annotation efficiency
• Customize pipelines in spaCy and train word embedding NER models for parsed job descriptions and resumes
• Optimize word2vec model in genism package by performing directional skip-gram for word embeddings, which explicitly distinguishes directions of contextual words and improves word prediction for corpora of resumes
• Implement a tokenization tool that integrates functionalities such as TF-IDF (term frequency-inverse document frequency) retrieval, concordance, and term-index lookups, phrase merges and removals, and document.

Wuhan, Hubei, China

• Used LSTM network to develop energy consumption prediction model for the entire city based on data provided by China Resources Gas Group Limited, which can predict gas usage for the next day based on that of the.
• Added multi-variable support and optimize the usage forecast model with LSTNet framework in Tensorflow/Keras
• Applied K-Means clustering with DTW metric to divide the energy consumption of the city into different regions
• Reduced the prediction errors caused by holidays and events by using Prophet by Facebook

• Utilized OpenCV and checkerboards to calibrate camera distortion for self-taken videos on autonomous vehicles
• Annotated numbers of images with multiple classes labels and organized labeled data into training formats
• Deployed state-of-the-art Multi-Object Tracking model (FairMOT) on customized training video dataset with PyTorch
• Improved the classification results for with YOLOv3 backbone and B-CNN model and tuning parameters
• Estimated the position and trajectories of autonomous vehicles by analyzing sequences of camera images