Duke Researchers Build AI That Finds Simple Rules in Complex Systems

1/1/26 News 1 min read

PreMiEr researcher Boyuan Chen and his Duke General Robotics Lab have developed an AI framework that turns complex, changing systems into compact, readable equations.

Duke Chapel — Duke Researchers Build AI That Finds Simple Rules in Complex Systems

Published in npj Complexity, the framework studies time series data from experiments and reduces systems with hundreds or thousands of interacting variables to a much smaller set that still captures their essential behavior. Building on mathematician Bernard Koopman’s idea that nonlinear systems can be represented with linear models, the team tested it on pendulums, electrical circuits, and climate and neural circuit models, often producing models more than 10 times smaller than earlier machine learning methods. The framework can also identify attractors, the stable states a system tends toward, supporting the lab’s goal of building “machine scientists” that help researchers move from raw data to testable rules.