Human Biophysics as Network Weights: Conditional Generative Models for Dynamic Simulation

Simulations of biophysical systems are fundamental for studying physiological mechanisms and developing human machine interfaces. Whilst advanced numerical methods, such as finite element models, can excel in this task, they are extremely computationally expensive to use when generating a large number of simulations or simulating dynamic events with continuously changing structural parameters. We propose an architecture that uses a conditional generative model to interpolate between the numerical model states, dramatically lowering the modeling time while maintaining a high generation accuracy. As a demonstration of this concept, we present BioMime, a hybrid-structured generative model that enables an accurate, ultra-fast, and arbitrarily high temporal-resolution simulation of a specific biophysical system during dynamic changes. This methodology has wide applications in physiological and clinical research as well as in supporting data augmentation strategies for signal analysis, representing a computationally efficient and highly accurate model for biophysical simulations.