Causal Inference
Causal inference aims to determine cause-and-effect relationships from data, going beyond mere correlations to understand how interventions impact outcomes. Current research heavily focuses on addressing challenges like confounding (the influence of unobserved variables), particularly in high-dimensional data and complex treatments (e.g., text, sequences of actions), employing methods such as structural causal models, Bayesian Additive Regression Trees (BART), and various neural network architectures including Graph Neural Networks (GNNs). These advancements are crucial for improving the reliability of causal conclusions across diverse fields, from medicine and economics to personalized interventions and policy-making.
Papers
Causal modelling without introducing counterfactuals or abstract distributions
Benedikt Höltgen, Robert C. Williamson
Establishing Truly Causal Relationship Between Whole Slide Image Predictions and Diagnostic Evidence Subregions in Deep Learning
Tianhang Nan, Yong Ding, Hao Quan, Deliang Li, Mingchen Zou, Xiaoyu Cui
Probabilistic Temporal Prediction of Continuous Disease Trajectories and Treatment Effects Using Neural SDEs
Joshua Durso-Finley, Berardino Barile, Jean-Pierre Falet, Douglas L. Arnold, Nick Pawlowski, Tal Arbel
LLMs Are Prone to Fallacies in Causal Inference
Nitish Joshi, Abulhair Saparov, Yixin Wang, He He