Tradeoffs Between Robustness and Accuracy - Percy Liang

Explore the complex relationship between robustness and accuracy in machine learning models through this insightful lecture by Percy Liang at the Institute for Advanced Study. Delve into topics such as the general robustness problem, robust objectives, and strategies for addressing challenges in natural language inference and spurious correlations. Learn about various approaches including regularization, complexity modulation, and the impact of model size on performance. Gain valuable insights into the trade-offs involved in developing robust and accurate machine learning systems, and participate in the discussion through audience questions.

Introduction
General Robustness Problem
Robust Objective
Discussion
Audience Question
spurious correlations
natural language inference
general setup
training error
brief interlude
what do you do
Regularization
Complexity Modulation
Story
Toy Model
DL vs Upwait
Average Error
Bigger Models
Final remarks
Conclusion
Questions