Reproduce “A Survey of Learning Causality with Data: Problems and Methods” on OpenResearch

About this paper

Ruocheng Guo, Lu Cheng, Jundong Li, P. Richard Hahn, Huan Liu

This survey paper from Arizona State University and University of Virginia examines how the availability of copious data impacts the ability to learn causal effects and relations. It distinguishes between traditional and big data challenges and opportunities, categorizing existing methods for causal inference and discovery, and highlighting their interplay with machine learning.

Key insights

• Structural Causal Models (SCMs) and the Potential Outcome Framework serve as the two foundational pillars for formally defining and reasoning about causality.
• Big data introduces both new opportunities, such as mitigating unmeasured confounders, and unique challenges, including increased complexity in understanding biases and statistical issues with high-dimensionality.
• Machine learning techniques can enhance causal inference by providing powerful estimation tools, while causal principles can improve machine learning models by enabling invariant prediction, robustness, and fairness.

Summary by alphaXiv.