Gaussian Process Surrogate Modeling
In modern scientific computing and engineering, evaluating complex models or simulators can often be computationally expensive. Each simulation may take minutes, hours, or even days to complete making tasks like optimization, uncertainty quantification, or sensitivity analysis extremely challenging.
To overcome this, researchers increasingly rely on surrogate models (also called emulators or metamodels). These models aim to approximate the behavior of a costly function using a much cheaper-to-evaluate statistical representation. Among the various approaches, Gaussian Process (GP) modeling has become one of the most popular and powerful techniques.
A Gaussian Process provides a probabilistic framework to represent uncertainty about the function being modeled. It not only predicts the function’s value at unseen points but also quantifies the confidence in each prediction. This property makes GPs particularly suited for Bayesian optimization, sequential experimental design, and uncertainty propagation.
In this article, we introduce the key concepts behind Gaussian Process surrogate modeling — explaining how it works, why it is so effective, and how it can be applied to real-world problems where exhaustive simulation is computationally infeasible.➡️ Continue reading…
Design of experiments
In scientific research, engineering, and data-driven modeling, exploring the relationship between inputs and outputs often requires running experiments or simulations. However, when each experiment is costly or time-consuming, it becomes crucial to plan them strategically. This is where the Design of Experiments (DoE) methodology comes into play.
Design of Experiments provides a systematic and quantitative framework for selecting informative combinations of input variables. Instead of performing random or exhaustive trials, DoE aims to maximize information gain while minimizing the number of experiments or simulations required.
Originally developed for physical experiments in agriculture and industry, DoE has evolved into a cornerstone of computational science, machine learning, and uncertainty quantification. It enables researchers to build accurate surrogate models, perform sensitivity analyses, and guide sequential optimization processes efficiently.
A well-designed experimental plan not only improves model accuracy but also enhances interpretability and robustness, allowing better understanding of complex systems under uncertainty.
In this article, we introduce the fundamental principles of Design of Experiments — what it is, why it matters, and how it can be leveraged to accelerate scientific discovery and data-driven modeling.➡️ Continue reading…
Academic Projects
During my studies at ENSAI (École Nationale de la Statistique et de l’Analyse de l’Information), I had the opportunity to work on several academic projects combining statistics, data science, and machine learning.
These projects allowed me to apply theoretical concepts to real-world problems — from data exploration and predictive modeling.➡️ Continue reading…