San Francisco Bay Area
Published 6 research papers (4 first-authored) and 2 open-source softwares. • BotTriNet: built a social fraud detection system, accomplished a user profile pipeline via NLP, improved the performance based on representation learning, and published on IEEE ISDFS 2023. • MedLens: developed a mortality prediction system with a features selection pipeline, proposed a regression-based time series interpolation approach to tackle data-missing problems, accepted by IEEE CCAI 2023.
•Lead the recommendation manipulation project, detect risk live-streaming where anchors massively manipulate audiences behaviors, govern abused traffic (occupying x% of the platform), and recover biased feed/ranking.Design an active-learning architecture to gradually boost from cold-start, alleviation to comprehensive governance. •E-commerce Retention Rate Optimization: Key contributor for the e-commerce retention rate improvement project, forecasting refund orders/users in order creating time, contributing to redistributing natural traffic to high retention rate shops. •Bots Detection and Banning: Contribute to bots and fraud banning project across multiple social applications with 100 M+ DAU, build detection model at registration/login time with > 99.9% precision and < 0.1% complaint rate .
Designed and implemented a machine learning-based disk failure prediction system using sequential features mined from system logs and disk SMART logs. Achieved a recall of 94%, with a false alarm rate of 0.04% in a very large cloud service provider owning million-level disks.
Designed a unified and robust performance anomaly detection framework that predicts the expected quantities of KPI time series. Cooperated with front-end engineers to deploy a data visualization platform for network anomaly monitoring.