All Causal DAGs Are Wrong but Some Are Useful
Offered By: Uncertainty in Artificial Intelligence via YouTube
Course Description
Overview
Explore a thought-provoking keynote talk from the Uncertainty in Artificial Intelligence (UAI) 2024 conference, delivered by Dominik Janzing from Amazon Research, Germany. Delve into the concept that "All causal DAGs are wrong but some are useful," challenging the assumption of a "true DAG" in causal discovery. Examine over five different tasks where causal Directed Acyclic Graphs (DAGs) can be beneficial and consider how causal discovery research could be reinvigorated through benchmarking against these tasks. Investigate the meaning of causality through various perspectives, including self-compatibility evaluation, the causal marginal problem, and predicting unobserved joint statistics. Gain insights from related works spanning topics such as causal inference, phenomenological accounts of causality, causal and anticausal learning, and root cause analysis of outliers with missing structural knowledge.
Syllabus
UAI 2024 Keynote Talk 2: Dominik Janzing
Taught by
Uncertainty in Artificial Intelligence
Related Courses
Macroeconometric ForecastingInternational Monetary Fund via edX Machine Learning With Big Data
University of California, San Diego via Coursera Data Science at Scale - Capstone Project
University of Washington via Coursera Structural Equation Model and its Applications | 结构方程模型及其应用 (粤语)
The Chinese University of Hong Kong via Coursera Data Science in Action - Building a Predictive Churn Model
SAP Learning