Explore measures of relatedness in domain adaptation and related problems in this 1 hour 15 minute lecture by Samory Kpotufe from Columbia University, part of the Modern Paradigms in Generalization Boot Camp at the Simons Institute. Delve into unstructured distribution shifts, focusing on domain adaptation and transfer learning settings with minimal assumptions about the nature of distribution shifts. Examine discrepancy measures between source and target distributions, learn adaptation techniques for unknown measures, and understand statistical limits in various scenarios including multitask and model selection. If time allows, discover new unifying principles that encompass multiple discrepancy measures simultaneously.
Measures of Relatedness in Domain Adaptation and Related Problems
Overview
Syllabus
Measures of relatedness in domain adaptation and related problems.
Taught by
Simons Institute
Related Courses
-
A Discrepancy-Based Theory of Domain Adaptation
-
Domain Adaptation - 20 Years of Theory Chasing Practice
-
Domain Adaptation on Wheels: Closing the Gap to the Open-world
-
Domain Adaptation and Related Areas in Machine Learning
-
Pseudo-Labeling for Covariate Shift Adaptation in Machine Learning
-
Controlling Distribution Shifts in Language Models: A Data-Centric Approach
