Leading multiple data science teams on various projects related to Connected TV buying strategies.

Leading a data science team that focuses on Connected TV buying strategy. Building big data & ML-powered solutions (Scala/Python/Spark/Databricks) leveraging a variety of datasets for both linear TV and connected TV.

Help optimize corporate clients' marketing and advertising efforts with bespoke data science solutions leveraging advanced statistics, machine learning, deep learning, and industry knowledge.

Researched into look-alike models for audience expansion using semi-supervised and supervised machine learning approaches using Spark ML. Prototyped data processing pipeline in BigQuery and PySpark.Led several analytics projects on offline data hygiene/quality assessment and productization supporting multiple products for LiveRamp's major customers.Led the research in modeling and analytics for our new matching product: statistical matching for Connected TV.Designed and implemented mobile publisher quality metrics.Designed and implemented identity connectedness score for fraud detection.

Developed an algorithm to identify cookie churn with > 99% precision contributing to a ~30% cut in edge store size.Analyzed data quality and demographic bias of publisher websites then made suggestions on billing.Built a dashboard with anomaly detection and forecasting for daily traffic of pixel servers.Using Python, R, Redshift, MySQL and Apache Spark, Looker

• Obtained data-related job postings from ~1,300 companies across 29 US states using the Indeed.com API.• Processed job description texts with natural language processing techniques (e.g. bag of words, Tf-Idf) using Pandas, NLTK, Beautiful Soup and Scikit-learn.• Trained a random forest classifier with 200 trees to predict the job title given the job description text and achieved 0.65 average accuracy (compared to 0.17 by the baseline model) in 5-fold cross-validations.• Built a web application (JoBreaker, deployed with Flask and AWS) that takes text description as input and returns a list of data-focused job titles ranked by their matching probabilities.

Developed dynamic models predicting species interaction and community structure applying probability theory, combinatorics and information theory. Designed and implemented simulations in python and R to study food web network structure.

The name of the class is ER 102: Quantitative aspects of global environmental problems.For one semester, I gave five lectures (one hour each) per week. In each lecture, I went through class materials and sample problems to start with the problem set of the week.I also held two office hours (three hours each) per week, during which I went through key points of the problem sets, and answer any questions that the student may have.Besides, I prepared midterm and final review sessions (two hours each) and review materials.I helped grade midterm and final exams, put all scores into spread sheets and conducted analyses.

• Applied cluster analysis to identify user groups using >200,000 user posts with 22 demographic and activity features queried from a MySQL database.• Identified keywords for each user group and association among keywords using the point-wise mutual information (PMI) metric and visualized the result with correlation networks (R igraph).• Designed and implemented an interactive dashboard (R shiny) to show the results, which was used by the website administrator to improve bot detection and group recommendation.

For two summers conducted field work and theoretical research in RMBL. Designed and set up sample transects, collected, compiled and compared data with multiple statistical models.

As a sophomore I joined the Natural Science Foundation of China project: Climate change and land surface processes and was in charge of data mining, compilation and statistical analyses.