Zepu Wang

Zepu Wang

M.S. in University of Pennsylvania

Zepu Wang 王泽普

CV

My Curriculum Vitae, updated in 11/13/2023

Education

  • Master of Science in Engineering in Data Science, University of Pennsylvania
    08/2022 ~ 08/2024
    GPA: 3.9/4.0
    Relevant Courses: Machine Learning, Deep Learning, Graph Neural Networks, Natural Language Processing, Time Series for Business, Statistics for Data Science

  • Bachelor of Science in Data Science, Duke University/Duke Kunshan University 08/2018 ~ 05/2022
    GPA: 3.95/4.0
    Relevant Courses: Machine Learning, Deep Learning, Stochastic Process, Mathematical Modeling
    Awards: Merit-based Undergraduate Official Scholarship, Summa Cum Laude (Undergraduate highest honor), Dean’s List (2018, 2019, 2020, 2021, 2022)

Research Experience

  • Online Research Intern, Massachusetts Institute of Technology (MIT)
    02/2023 ~ 08/2023
    Supervisor: Prof. Jinhua Zhao, Prof. Shenhao Wang
    Project 1: Uncertainty Quantification in Traffic Data Imputation
    • Utilized graph attention layers and bidirectional recurrent units to capture spatio-temporal traffic data patterns and predicted uncertainty of traffic data.
    Project 2:Mixture Models for Uncertainty Quantification in Sparse Travel Demand Prediction
    • Used mixture models (Laplace, Poisson, Gaussian) to approximate complex travel demand distribution, addressing high zero occurrences. link

  • Research Intern, Duke Kunshan University
    07/2021 ~ present
    Supervisor: Prof. Peng Sun, Ph.D.
    Project 1: A Velocity-based Model in Traffic Flow Prediction
    • Conducted evaluations and comparisons among popular traffic flow prediction models (Linear Regression, SVR, Decision Tree, Random Forest, LSTMs, and GRUs) in a single intersection based on accuracy (RMSE,MAE,R^2).
    • Increased 2% accuracy (RMSE) on average by considering the vehicles’ speed in surrounding intersections to adjust the original results.
    Project 2: A Hybrid Model in Traffic Flow Prediction!
    • Combined Long Short-Term Memory (LSTM) neural networks, Wavelet Analysis and Spectral Analysis to design an accurate traffic flow forecasting algorithm.
    Project 3: Traffic Flow Prediction using Auto-encoder
    • Applied Auto-encoder as a dimension reduction technique for large road networks and increased the time efficiency by 27.4% with sacrifice of only 5% accuracy (MSE).
    • Provided a comprehensive analysis of trade-off between original data information loss and noises’ filtration from the original data while applying dimension reduction technique.

  • Research Intern, Westlake University
    07/2021 ~ 08/2021
    Supervisor: Prof. Stan Z. Li
    Project: Protein-Protein Interactions (PPIs)
    • Predicted PPIs based on primary protein structures, using traditional natural language processing (NLP) methods (RNNs) and advanced NLP methods (Transformers).
    • Increased accuracy by 2% (MSE) in RNNs methods by using the trick of pad sequences.