EMR

arXiv paper proposes a graph-based “probabilistic compositional inference” method to solve inverse problems in large coupled engineered systems (notably power grids + embedded turbine multiphysics) with sparse/noisy sensing. Key claimed advantage is uncertainty-aware state/parameter inference with scaling improving from ~cubic to ~linear by avoiding global augmented state/covariance, enabling hierarchical subsystem composition and mixed mechanistic/learned components.

Paper proposes a fully automated resonant core-loss measurement setup for sub‑MHz magnetics using digitally controlled switched-capacitor sequences plus onboard signal processing, replacing manual tuning + heavy FFT workflows. If commercialized, it reduces magnetics characterization time (1000+ points/20s) and labor, potentially accelerating development cycles for high‑frequency power magnetics used in EV/inverter, data-center/AI power, and industrial supplies. Near-term investability hinges on