Understanding and modeling users’ interests from their online activity trails collected from wide specter of data sources for prospective and retargeting conversion prediction tasks.
Modeling mobile users demographics and interests from their installed mobile apps.
Bid shading (diminishing bid price) for first-price auctions by modeling bidding landscape.
Developing major extensions aimed to: 1) facilitating ensemble-based users embedding where constituents can be replaced by embedding models of any type; 2) learning distributed representations for the target events-time purchase pairs.
Fusing large scale hospital discharge records databases with various domain knowledge sources to obtain novel and meaningful insights and improved predictions in healthcare.