Steven Xu Email & Phone Number

Bellevue, Wa, Us

Seattle, Washington, United States

Building DoorDash's Product Knowledge Graph with Multi-modal LLMs.

Seattle, Washington, United States

Personalization; Information Extraction; Product Knowledge Graph; Generative AI.

Cary, North Carolina

I research on Machine Learning algorithms and apply them to analyze high-frequency data collected by intelligent sensors. Examples include robust principal component analysis (RPCA) and KTChart (a SVDD based statistical monitoring chart) for anomaly detection, and independent component analysis (ICA) for blind source separation. Some of the projects I have worked on are:- Online unsupervised wildfire detection using streaming satellite imagery, where I proposed a RPCA based multispectral outlier detection algorithm to monitor California wildfires using GOES-R images.- SAS bee project, where I applied RPCA to monitor and track bee movements using real-time video data. - PROC KTCHART, where I was involved in the DevOps of a new SAS procedure for statistical monitoring. I wrote scripts for programming testing, simulation study, and application.- Online blind source separation using ICA, where I researched on various ICA algorithms (e.g., Infomax, FastICA, Picard, Orica), wrote scripts for simulation study, and help incorporate Orica into SAS modules.

Cary, North Carolina, United States

- Worked with AI Model Development CoE on cybersecurity applications.- Rebuilt False Positive Agent (FPA) that identifies true customer login attempt from Blacklisted IPs. - Researched on machine learning methods for highly imbalanced classification (2% event rate).- Created pipelines for feature engineering, data visualization, and model selection.- Improved the out-of-time performance metric of production model by more than 20%.

Raleigh-Durham, North Carolina Area

- Supervise graders to ensure they work effectively- Manage course website- Teach 3 sessions (35 students each) of undergraduate statistics course.- Educate entry-level concepts to students from various backgrounds with minimum exposure to statistics.- Evaluated by students as an approachable, caring teacher always willing to provide clarification and effective feedback.

Chapel Hill, North Carolina

Under the supervision of Katie Mollan, MS and Dr. Angie Bengston, I worked on the project "Transporting outcomes from randomized trials to a population of interest under bootstrap multiple imputation" where I applied bootstrap multiple imputation (BootMI) to analyze the uncertainty of causal effects of efavirenz intake (an antiviral treatments for HIV) on developing suicidal thoughts, when that effect was transported from ACTG trials to CNICS cohort suffering from missing covariates. I wrote R scripts for the BootMI algorithm and the main analysis. I also co-authored a manuscript that was accepted by the American Journal of Epidemiology (https://doi.org/10.1093/aje/kwab136).

Greater Seattle Area

Implement some of the newest probabilistic forecasting techniques that had been proposed in the literature and test them using data that has been made available as part of forecasting competitions in the area of renewable energy. Time series regression with models such as gradient boosting machine and neural net. R packages used: GBM, SVM, XGBoost, MXNet, quantregForest, ggplot2.

Doctor Of Philosophy - Phd, Statistics

Bachelor Of Science (B.S.), Statistics

Activities and Societies: Probability Research Organization