Actionable Machine Learning for Tackling Distribution Shift - Huaxiu Yao

Explore cutting-edge approaches to tackling distribution shift in machine learning with this 58-minute lecture by Stanford University's Huaxiu Yao. Dive into two paradigms for addressing subpopulation and domain shifts, learning how to build robust models and adapt them to test distributions with minimal labeled data. Gain insights into real-world applications, challenges, and future research directions in this field. Benefit from Yao's expertise as he shares findings published in top-tier venues and discusses practical implementations for solving problems with limited data.

Introduction
Machine Learning Systems
Model Failure
Goal
Motivation
Representations
Datasets
Predicting Domain Information
Two Domains
spurious correlation
LISA
Copy Paste
How Human Learn
MLTI
Hidden Partition
Threeway Classification
NonLabel Sharing
Limited Tasks
Data Augmentation