Research

My research interests include surrogate modeling and uncertainty quantification for computer experiments, with particular interests in deep Gaussian processes and sequential learning.

Preprints

Efficient Deep Gaussian Process Surrogates via Mini-Batch Latent MAP Estimation.
Chang, C.-Y. and Sung, C.-L. (2026). Submitted.

We propose LaMP-DGP, a deep Gaussian process surrogate that combines latent maximum-a-posteriori estimation with efficient probabilistic prediction. A mini-batch optimization algorithm enables training on larger datasets, with competitive predictive performance across numerical experiments and real-data applications.

Publications

Deep Intrinsic Coregionalization Multi-Output Gaussian Process Surrogate with Active Learning.
Chang, C.-Y. and Sung, C.-L. (2026). International Journal for Uncertainty Quantification.

We develop deepICMGP, a multi-output deep Gaussian process surrogate that uses hierarchical coregionalization to capture nonlinear dependencies among simulation outputs. We incorporate active learning strategies to sequentially select informative input locations, enabling efficient surrogate modeling for multi-output computer experiments.