Work Experience

University of Michigan Library

Data Engineer

Sep 2023 - Present

Collect relevant publications information using RestAPI and web scraping.
Perform data cleaning and store data using both RDBMS and MongoDB based on different needs on query efficiency and data scalability.
Use python pandas and sklearn to analyze publications that involved self-plagiarism from different perspective like reasons and categories. Data visualization with seaborn and matplotlib.
Cultivate project management skills. Design and schedule the project from the beginning. Automate codes for later use.

Ernst & Young

Data Analytics Intern

Dec 2022 - Mar 2023

In-depth participation in data analysis projects with four client companies. Research on related projects, financial data and revenue situations.
Participate in IPO project of publicly listed company (TUHU Car Inc.), conduct user purchase analysis, identify abnormal store operations, and compare third-party transactions based on structural data analysis and machine learning skills like SVM and Regression Models.
Use data analysis tools like SQL and Python to do database normalization and perform basic ETL operations.
Utilize big data tools like Spark SQL and Hive to analyze data of millions of entries.
Improve database query efficiency by 50% in PostfreSQL by adding B-Tree type indexes and introducing inverted indexes like GIN.

Shanghai Jiaotong University, Joint Institute

Teaching Assistant

Sep 2022 - Dec 2022

Course VE215 Circuit Analysis Teaching Assistant.
Conduct tutorials and review sessions to reinforce understanding of circuit analysis techniques.
Assist students during laboratory sessions to ensure smooth execution of experiments.
Offer regular office hours to address individual student concerns and provide additional assistance.

Project Experience

Information Retrieval System Implementation

Implement a Google-like search engine based on inverted index and text analysis.
Use different scorer like BM25 and TFIDF to rank results returned by search algorithm.
Use neural network called learning to rank to examine and modify search results and improve accuracy by 25%.
Add deep learning features by using BiEncoders and CrossEncoders to take queries into consideration and further increase model robustness and completeness.

Data Augmentation in Deep Reinforcement Learning

Adapt the baseline from Mujoco platform and DrQ-v2 in Pytorch, which is an open-source advanced deep reinforment learning method.
Implement basic methods of data augmentation on the original inputs as well as the feature maps, including RandomRotation, RandomAffine, etc.
Use saliency map from deep learning models to extract important features from images and augment the inputs. The accuracy gets improved by over 20%.

Alzheimer MRI Diagnosis and Reproduction with CNN and GAN

Use classification methods like Naive Bayes, Logistic Regression, SVM to classify Alzheimer illness degrees.
Increase the accuracy to up to 98.30% by using deep learning tools like CNN.
Data augmentation skills like rotation, squeeze, clipping to produce more training samples.
Use GAN to reproduce similar MRI images of different degrees of illness with tuned GLoss 5.1 and DLoss 0.4.

Visual Analysis on Exploring Cyber Asset Graphs of Cybercrime Gangs

Web crawl to gather data indicating edges and vertices of the cyber asset graph.
Utilize search algorithms including BFS, DFS and A* to detect potential relationships between nodes.
Create an interactive web page with Flask and JavaScript D3 library to show relations between different cyber assets.

A Risk-Resistant Trading Strategy Modeled by VaR-LP based on SW-ANN Prediction

Make forecasting on stock price based on various regression models and compare their performance.
Develop a strategy combining ideas of sliding-window and neural networks to make proper portfolio investments which can at most increase original assets by 40 times based on the given dataset.