Simulation-based inference for scientific discovery

Computational modeling plays a crucial role in both generating and testing hypotheses in the natural sciences. Across a diverse array of scientific disciplines -- ranging from particle physics and biological sciences to climate science and cosmology --, the scientific workflow shares key commonalities. Typically, models are designed to capture certain aspects of empirical data. These models are then analysed for a deeper understanding of the system of study and to provide predictions for further experimental validation. Despite its importance, however, this process is not without its hurdles. Notably, the quantitative linking of models to data -- also known as statistical inference --, is a task that frequently proves elusive.

In this project, the student will be trained on simulation-based inference, a rapidly growing field in machine learning dealing with statistical inference on simulator-based models. The student will build on the latest advancements in probabilistic deep learning to develop an algorithm that tackles current limitations in simulation-based inference, in particular regarding the scalability to high-dimensional parameter spaces and/or computationally expensive simulators.

Type of project: Thesis

Required degree: Master of Science, Master of Engineering Science

Supervising scientist(s): For more information on this topic, please contact Pedro Goncalves (pedro.goncalves@nerf.be)

Only for self-supporting students.