Stochastic Optimal Control for Continuous-Time fMRI Representation Learning
Brain Dynamics with Optimal Control (BDO) uses a continuous-discrete State Space Model to decode fMRI signals. It employs Stochastic Optimal Control (SOC) for efficient inference and a locally linear approximation to avoid slow simulations. The model learns transferable representations by optimizing an objective function that aggregates control policies into a universal feature, achieving state-of-the-art performance.