CURV Institute publishes research on representational stability, routing mechanisms, and control methods for complex systems. Our work focuses on diagnosing and correcting failures that arise when representations are composed across depth, time, or multiple sources—problems that manifest as training instability, brittle inference, and unreliable routing in modern architectures.
All publications include reproducible code and data. We prioritize clear problem statements, controlled experiments, and practical applicability over novelty for its own sake.