I support the Friending product team, the Growth Machine Learning team, the Growth Foundation & Data team, and the Graph Understanding Platform at Snap. We develop features to help users make virtuous friends on Snapchat, deliver ML solutions and infrastructure to power growth products (e.g., friend recommendations, push notification, search, in App send-to, and off-platform sharing), and build the Graph Understanding Platform to mine signals from complex entity relationships on Snapchat, unlocking new business opportunities. My team is looking for kind, smart and creative engineers (client, cackend, infra, and machine learning). Please drop me a note if you are interested.

I have a passion in teaching and helping more people into the field of machine learning and data science. Currently I teach several machine learning courses in the Foster School of Business at University of Washington, including advanced ML (MSIS 522), Deep Learning and Big Data (MSIS 549), and NLP (MSIS 541).

I led several machine learning driven projects including: 1) seller personalization and recommendation on seller central homepage, 2) seller email governance and suppression, 3) seller downstream impact analytics, and 4) sourcing best quality products for Amazon promotions.

Our team uses machine learning techniques to power eBay Paid Internet Marketing, including search (Google, Bing), social (Facebook), and display. - Optimize the bids of keywords (Text Ads) and eBay listings (Product Listing Ads) sent to Google and Bing so that we improve the ROI and brings more new buyers to eBay. - Design bidding strategies to retarget eBay users via social and display.- Content recommendation in retargeting.Tech stack: Scalding, Spark, Scala, Python and R.Award: eBay Critical Talent Award and eBay Seattle Technical Achievement Award.

In eBay Trust Science team, I built large-scale machine learning models to manage seller risk and reduce defects (bad buyer experience) on eBay via Search Ranking, Merchandizing, and Marketing. Our production job processes Terabytes of data daily on Hadoop to predict the risk of 30+ million active sellers and has banked over 100+ million dollars GMB lift over the next 12 months. I have been recognized for this work and won the eBay Spot Award.

PhD Dissertation - Species Distribution Modeling using Citizen Science DataSpecies distribution modeling (SDM) is to predict the species range spatially. We develop a probabilistic graphical model that incorporates the expertise of observers to build SDMs using data from a large-scale citizen science project.Data quality control in citizen science projects: Data quality is essential for SDM in citizen science projects. We build up an automated data verification process for eBird project and show that it identifies 50\% more invalid observations and reduces the human screening efforts by about 40\% compared to the expert-defined data filters.Discovering species confusions: One way to improve the skills of novice birders is to teach them how to distinguish confusing species. We propose to discover observer confusion between species by modeling multiple species simultaneously and show that this information helps novice birders contribute better quality data.Multi-species distribution modeling: Single-species model can not capture species interactions is SDM. To address this problem, we apply a multi-label learning algorithm on this problem and show that multi-species modeling improves the predictions, especially on the rare species.

Inexperienced users with different purchase intents use ambiguous queries. We propose a Latent Dirichlet Allocation-based approach to retrieve diverse items so that the risk of users with diverse intents not seeing any relevant items is minimized. Our LDA-based approach improves the user satisfaction at eBay by more than 6\% compared to the eBay production ranker.