Ghifari Adam Faza

Ghifari Adam Faza

PhD Student at KU Leuven

Publications

Direct interval propagation methods using neural-network surrogates for uncertainty quantification in physical systems surrogate model

Published in Knowledge-Based Systems, 2026

We reformulate interval uncertainty propagation as an interval-valued regression problem to avoid costly optimisation-based methods. Using neural network surrogates, including MLPs and DeepONets, we evaluate direct interval prediction via naive propagation, bound propagation (IBP, CROWN), and interval neural networks. The proposed approaches significantly improve computational efficiency while maintaining accurate interval estimates.

Download here

Most likely heteroscedastic Gaussian process via kernel smoothing

Published in Knowledge-Based Systems, 2025

We propose a computationally efficient heteroscedastic Gaussian process (HGP) approach that reduces training cost by replacing the second GP model with a kernel smoothing–based noise estimator. This reduces complexity from $\mathcal{O}(2\mathcal{N}^3)$ to $\mathcal{O}(\mathcal{N}^3 + \mathcal{N}^2)$, yielding roughly a 2× speed-up while maintaining or improving predictive performance. The method also demonstrates strong performance in Bayesian optimisation, particularly in low-data regimes.

Download here

Interval Reduced Order Surrogate Modelling Framework for Uncertainty Quantification

Published in AIAA SciTech 2024 , 2024

This paper introduces a non-intrusive framework for epistemic surrogate modeling, leveraging interval proper orthogonal decomposition (interval POD) and interval polynomial chaos expansion (interval PCE) to handle interval observations, addressing a major limitation in existing frameworks. By integrating POD for interval data with PCE for interval observations, the framework allows for the consideration of non-scalar data, such as intervals, providing a more comprehensive approach to physical system modeling that captures additional information.

Download here