This course teaches learners how to achieve provable robustness in deep learning through adversarially trained smoothed classifiers. The course covers topics such as definitions, randomization, the Zico idea, experimental and semi-supervised results, training techniques, notation, gradients, optimal gradients, the full algorithm, parameters, and a summary of results. The teaching method involves a 47-minute lecture by Jerry Li from Microsoft Research. The course is designed for individuals interested in advancing their knowledge of deep learning and enhancing the robustness of their models.
Overview
Syllabus
Intro
Definition
Randomization
Zico
Idea
Experimental Results
SemiSupervised Results
Training
Notation
Gradients
Optimal Gradient
Full Algorithm
Parameters
Results
Summary
Taught by
Simons Institute