Fully Bayesian differential Gaussian processes through stochastic differential equations

Jian Xu, Zhiqi Lin, Min Chen, Junmei Yang, Delu Zeng, John W. Paisley

Abstract: Highlights•Fully Bayesian DiffGPs: Treats kernel hyperparameters as random variables via coupled SDEs.•Adaptive SDE: Uses neural networks as black-box solvers for time-varying dynamics.•Superior performance: Outperforms traditional methods in flexibility and accuracy.