Designing Robust Learners

Explore the challenges and strategies for designing robust machine learning models in this 53-minute lecture by Jacob Steinhardt from Stanford University. Delve into the emerging challenges in deep learning, focusing on the importance of creating learners that can withstand various perturbations and maintain performance across different contexts. Examine key concepts such as total variation (TV) and mean estimation, modulus examples, finite-sample estimation, and resilience in the face of arbitrary loss functions. Investigate specific applications in linear regression and covariance estimation, and learn about extending these principles to other types of perturbations. Gain insights into the formal definition of friendly perturbations and understand how resilience applies to different scenarios in machine learning.