High Dimensional Model Explanations - An Axiomatic Approach

Explore an axiomatic approach to high dimensional model explanations in this 20-minute conference talk from FAccT 2021. Delve into the research presented by N. Patel, M. Strobel, and Y. Zick, which covers key concepts such as value functions, design approaches, efficiency, monotonicity, and limit conditions. Examine the characterization results and a case study on movie reviews, while also considering potential pitfalls and future research directions in this field. Gain insights into the complexities of explaining high dimensional models and the importance of developing robust methodologies for interpreting machine learning outcomes.

Introduction
Value Function
Design Approach
Efficiency
Monotonicity
Limit Condition
Characterization Results
Case Study Movie Reviews
What Can Go Wrong
Future Work