AIM-3D Team Presents Research at KDD 2026
The AIM-3D team participated in KDD 2026, the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining, held August 9–13, 2026, in Jeju, South Korea.
Michael Coppedge presented the paper “From Causal Discovery to Dynamic Causal Inference in Neural Time Series,” co-authored with Dmitry Zaytsev and Valentina Kuskova.
The paper introduces Dynamic Causal Network Autoregression (DCNAR), a two-stage neural framework that combines data-driven causal discovery with time-varying causal inference. Rather than assuming that the underlying causal network is already known, DCNAR first learns causal structure from multivariate time-series data and then uses that structure to estimate how causal influences change over time.
The study evaluates the framework using multi-country panel time-series data and shows that learned causal networks can produce more stable and behaviorally meaningful dynamic causal inferences than coefficient-based or structure-free alternatives, even when their forecasting performance is comparable. The findings demonstrate the potential of AI as a scientific instrument for dynamic causal reasoning, particularly in settings where causal structures are complex or uncertain.
This research contributes to AIM-3D’s broader effort to integrate artificial intelligence, causal inference, and social science theory to advance the modeling and understanding of complex social and political systems.
Authors: Dmitry Zaytsev, Valentina Kuskova, and Michael Coppedge
Presented by: Michael Coppedge
Conference: KDD 2026, August 9–13, 2026, Jeju, South Korea
Paper: From Causal Discovery to Dynamic Causal Inference in Neural Time Series
