Search papers, labs, and topics across Lattice.
The paper investigates the design of foundation models for constrained scientific systems, focusing on AC optimal power flow (ACOPF) as a representative problem. Through controlled experiments, the authors identify three key design principles for building scientific foundation models that balance physics-invariance, constraint satisfaction, and reliability. They then introduce LUMINA, a framework with data processing and training pipelines, to facilitate reproducible research on physics-informed foundation models.
Foundation models can accelerate scientific computation, but only if they respect physical laws and safety limits, a challenge this paper tackles by deriving design principles for constrained scientific foundation models.
Foundation models in general promise to accelerate scientific computation by learning reusable representations across problem instances, yet constrained scientific systems, where predictions must satisfy physical laws and safety limits, pose unique challenges that stress conventional training paradigms. We derive design principles for constrained scientific foundation models through systematic investigation of AC optimal power flow (ACOPF), a representative optimization problem in power grid operations where power balance equations and operational constraints are non-negotiable. Through controlled experiments spanning architectures, training objectives, and system diversity, we extract three empirically grounded principles governing scientific foundation model design. These principles characterize three design trade-offs: learning physics-invariant representations while respecting system-specific constraints, optimizing accuracy while ensuring constraint satisfaction, and ensuring reliability in high-impact operating regimes. We present the LUMINA framework, including data processing and training pipelines to support reproducible research on physics-informed, feasibility-aware foundation models across scientific applications.
