Safe Bayesian Optimization for High-Dimensional Neural Control of Movement

The neural control of movement is a fundamental aspect of understanding how the nervous system coordinate and regulate motor behavior. Procedural safety is a critical concern for movement control, especially for online tasks in high-dimensional control spaces. Current safe exploration algorithms exhibit inefficiency and may even become infeasible with large high-dimensional input spaces. Furthermore, existing high-dimensional constrained optimization methods neglect safety in the search process. In this extended abstract, we introduce High-dimensional Safe Bayesian Optimization with local optimistic exploration (HdSafeBO), which is capable of handling high-dimensional neural control of movement under probabilistic safety constraints. We propose a local optimistic strategy to efficiently and safely optimize the objective function with theoretical safety guarantee. We use isometric embedding to handle problems ranging from a few hundred to several thousand dimensions while maintaining safety guarantee. HdSafeBO outperforms representative high-dimensional safe and constrained optimization algorithms in the control of a human musculoskeletal model. We also show its clinical application for safe and efficient online optimization of neural stimulation.