Abstract

This talk will mostly focus on our line of work around iterative learning and planning. We will start with a 4-year collaboration with a crowdsourcing food rescue platform, where we combined offline ML model with online optimization to improve volunteer engagement. We will discuss our randomized controlled trial, and our experience rolling it out to over 25 cities across North America. Lifting ourselves beyond this particular application domain, we propose bandit data-driven optimization, a theoretical paradigm for principled iterative prediction-prescription to address the unique challenges that arise in low-resource sustainability settings. We will also briefly discuss our other projects, including one with the World Wildlife Fund which won a 2023 IAAI Deployed Application Award. We will conclude the talk with a discussion of how recent advances like large language models can be leveraged by public sector organizations.

Time

2023-10-10  10:00 - 11:00   

Speaker

Zheyuan Shi, University of Pittsburgh

Room

Room 308