Past Events

Qualifying Exam Cross-Domain User Embedding - Solving Cold-Start Recommendation with Edge Computing
× Export Event Save iCal Send to Google Calendar Send to Yahoo Calendar Send to Outlook Live Send to Microsoft Outlook Save iCal Preserve formatting in description (only supported in some calendar applications) Close Tuesday, November 24, 2020, 10:00am - 12:00pm

Speaker: Shuchang Liu

Location : Remote via Zoom

Committee: 

Prof. Amelie Marian (Advisor)

Prof. Dmitris Metaxas

Prof. Yongfeng Zhang

Prof. Yipeng Huang

Event Type: Qualifying Exam

Abstract: When a user starts exploring within a recommendation system looking for items from new areas, cross-domain recommendation techniques come into help by transferring abundant knowledge from the user's ``acquainted'' domains.While this is usually achieved by sharing information between service providers on the clouds, we argue that it can be approached more naturally, efficiently, and effectively through learning on edge devices such as smartphones and laptops.In this work, we formalize the cross-domain recommendation problem under the mobile computing environment as compared to that in a centralized environment and identify two challenges to the community: unavailable direct transfer and heterogeneous domain-specific user representation.We then propose to learn and maintain cross-domain embedding on each user's personal device.The optimization follows a variational inference framework that maximizes the mutual information between cross-domain embeddings and domain-specific user information, and empirical study on real-world datasets exhibits its effectiveness on recommendation tasks and its superiority over domain-pairwise transfer models.We demonstrate that the resulting system has good scalability and allows flexible plugin of domain-specific encoder and decoders.

: 

Zoom link: https://rutgers.zoom.us/j/93328280185?pwd=KzZPaTdTL2s5alIvbmN0M3A3d3JxUT09

If join by ID: Meeting ID: 933 2828 0185, Password: 248138