Event Date
Speaker: Dogyoon Song, Post-Doctoral Research Fellow, University of Michigan
Title: "Advancing prediction for informed decisions"
Abstract: Decision making is crucial in various fields, and prediction is a key component. In this data-centric era, it is vital to use data effectively for accurate predictions, and translate these reliably into actionable decisions. My research focuses on enhancing decision-making processes by integrating machine learning (ML) techniques with statistical principles, thereby reinforcing foundational aspects of ML methodologies.
In this seminar, I will present my recent research through two vignettes: 1) developing useful prediction methods in the presence of corrupted data and complex response variables, and 2) improving decision reliability through calibrated predictions. The first theme addresses errors-in-variables regression with a metric-space-valued response, illustrating a combined ML and statistical modeling approach for versatile prediction with error mitigation. The second theme focuses on classifier calibration, introducing a distribution-free probability calibration method with a principled parameter choice.
These demonstrate how blending statistics and ML approaches can enhance insights into improving predictive methods and decision-making practices. I will conclude the talk with a brief overview of my broader research agenda and future directions.
Seminar Date/Time: Friday January 26, 2024 at 11:00am
Location: MSB 1147