Anastasios N. Angelopoulos, “Prediction-Powered Inference”
In-person in Clark Hall, Room 110
OR
virtually over Zoom
Join Zoom Meeting:
https://wse.zoom.us/j/97055652302?pwd=dWFUUHRHS1lna2h5K0U1cEt4RDRrQT09
If you would like to meet with Anastasios Angelopoulos before or after the seminar, please sign up for a time at this link:
“Prediction-Powered Inference”
Anastasios Angelopoulos
PhD Student
University of California, Berkley
Abstract: I will discuss prediction-powered inference—a framework for performing valid statistical inference when an experimental data set is augmented using predictions from a machine-learning system such as AlphaFold.
Our framework yields provably valid conclusions without making any assumptions on the machine-learning algorithm that supplies the predictions. Higher accuracy of the predictions translates to smaller confidence intervals, permitting more powerful inference. Prediction-powered inference yields simple algorithms for computing valid confidence intervals for statistical objects such as means, quantiles, linear and logistic regression coefficients. I will demonstrate the benefits of prediction-powered inference with data sets from proteomics, genomics, electronic voting, remote sensing, census analysis, and ecology.
This will be a chalk-talk, and we will discuss both the fundamental statistical principles underlying this framework and also its practical application, including plenty of code.
Biography: Anastasios Angelopoulos is a rising fourth-year PhD student at the University of California, Berkeley, advised by Michael I. Jordan and Jitendra Malik. From 2016 to 2019, Anastasios was an electrical engineering student at Stanford University advised by Gordon Wetzstein and Stephen P. Boyd.
Anastasios works on theoretical machine learning with applications in vision and healthcare. His goal is to apply modern statistical ideas to increase robustness of black-box models like deep neural networks. Anastasios is motivated by medical diagnostics: statistical reliability will become paramount as computer vision and machine learning become ubiquitous in such high-risk settings. His other applied interests include computational imaging and ophthalmology.