APIC: Amortized Physics-Informed Calibration using Neural Processes
Researchers have developed APIC, a new method for calibrating physics models that suffer from discrepancies with real-world data. This approach extends the Kennedy-O'Hagan framework by using Neural Processes to enable scalable, population-level Bayesian inference. APIC's architecture separates instance-specific physical parameters from shared discrepancy structures, allowing for rapid calibration and uncertainty quantification of new, unseen systems. AI
IMPACT Introduces a novel amortized inference technique for improving physics model accuracy and uncertainty quantification.