Stanford University

UCLA

Applied Researcher at eBay

Netflix

Research Engineer

eBay

Applied Researcher 2

- Train models to get embedding of user query and item description. Finetuned Splade model to get sparse representation of query and item title to improve search retrieval experience through embedding based retrieval.
- Build XML model to predict multiple category intent from user search queries

Applied Researcher 1

- Work on end-to-end NLP projects to identify buyer intent, including gender intent, item category intent.
- Deploy DL/ML models for query understanding to improve search recall size and provide ranking signals.
- Design model automation and iteration pipelines.

Applied Researcher

Teaching assistant for CS 224N Natural Language Processing with Deep Learning

Graduate Teaching Assistant

- Implemented in Python a UniLM-based language model to rewrite queries from search engine
- Deployed diverse beam search, top-k sampling, nucleus sampling and repetition penalties in the design of prediction decoding process, improved the diversity of generated queries by 18%
- Incorporated adversarial training with gradient ascent in UniLM to improve robustness of the model
- Designed a multi-task BERT model to rank rewritten queries by their diversity and correlation

Alibaba Group

Machine Learning Engineer Intern

Research assistant in Department of Sociology: 
- Perform social network analysis and network graph modeling on personnel flow in Chinese bureaucracy
- Created network topology inference and models, predicted network flows with network graph models

Research Assistant

- Designed, constructed and trained gradient boosting models for risk management of China Merchant Bank Flash Loan, and improved ks score of the model by 20%  
- Developed and implemented in Python an automated feature extraction tool to group data by users, extract relevant information and preprocess features from high dimensional data (over 30,000 dimensions)  
- Collected and analyzed data with Hive, constructed tools for automated online and offline test service

CreditX 氪信

Data Scientist

Research Assistant at Joseph Teran's Lab for analysis of FEM: 
- Apply FEM(finite element method) to simulate the motion of particles.
- Accelerate the calculation and simulation process using machine learning
- Analyze the restriction of SVD(singular value decomposition) and find the closest semi-definite hessian matrix for indefinite hessian matrix to improve the efficiency of FEM computation

Yuyan Wang