Scalable Gaussian Processes
Kernel Packet transforms Matérn kernels with half-integer smoothness into compactly supported functions via a sparse linear transformation, turning GP regression into banded linear solves that require only 𝒪(𝑛) time for Matérn GP regression.
Sampling with Sparse Grid under Additive Schwarz preconditioner generates prior and posterior GP sample paths by coupling inducing points on sparse grids with additive Schwarz preconditioning, yielding scalable iterative samplers whose convergence is theoretically guaranteed.
Optimal aggregation for distributed GPs introduces correlation-aware weighting schemes for both exact and variational GPs so that distributed GP predictions remain accurate, stable, and efficient for large-scale applications.
Haoyuan Chen
Research Fellow
My research interests include Gaussian processes, uncerntainty quantification and probabilistic machine learning.