PE-MHL: Physics-Encoded Modular Hybrid Layers for Scalable Learning of Complex Systems
Researchers have introduced a new framework called Physics-Encoded Modular Hybrid Layers (PE-MHL) designed to enhance the scalability and robustness of hybrid models that combine physics-based and data-driven approaches. This framework incrementally refines a physics-based model by adding new sub-models, ensuring that previously learned information is preserved. Theoretical guarantees show that this construction leads to monotonically non-increasing training errors and provable convergence. Empirical results on benchmarks like the nonlinear NARX and the Quanser Aero 2 platform indicate that PE-MHL surpasses monolithic networks in accuracy and generalization. AI
IMPACT Introduces a novel architecture for hybrid AI models, potentially improving performance and scalability in complex system learning.