Simon Ghyselincks

PhD student at UBC Computer Science — research in scientific machine learning and generative AI

Welcome to my personal site! I’m Simon Ghyselincks, a PhD-track student at the University of British Columbia (UBC) in the Department of Computer Science.

I completed my BASc in Engineering Physics with a minor in Computer Science at UBC in 2025. I am currently working with Professor Eldad Haber from UBC Earth and Ocean Sciences and Professor Evan Shelhamer from UBC Computer Science on scientific machine learning and generative AI with broad applications to geophysics and geospatial data. You can find my recent publications here.

Feel free to connect with me on LinkedIn or check out my GitHub.

Recent

Structural Geology Meets Deep Learning

A 3d geological simulation engine that can produce fast computation samples is presented. The synthetic dataset is used to train a generative AI model that can be used as a prior for solving difficult inverse problems in geophysics.

Numerical Methods for Deep Learning

Ordinary Differential Equations

A guest lecture and workshop exploring the intersection of deep learning and numerical analysis. Neuro differential equations with architectures such as ResNet and RevNet. Neural ODEs where a continuous time dynamical system is modeled by a neural network, with an application to generative AI and inverse problems.

Synthetic Geology: Structural Geology Meets Deep Learning

AIM-SI Optimization, AI, and Inverse Problems Workshop

Lightning talk at the AIM-SI Optimization, AI, and Inverse Problems Workshop at UBC, presenting our work on generative modeling for structural geology.

Trace Estimation for Implicit Matrices

A Hutchinson’s and Hutch++ Survey

Implicit matrix trace estimation using only matrix-vector products is a problem that commonly arises in many areas of scientific computing and machine learning. This work explores randomized algorithms including Hutchinson’s and Hutch++.

Dimensionality Reduction with the Johnson-Lindenstrauss Lemma

A randomized approach to preserving distances in high-dimensional data

Randomized algorithms provide an alternative approach to improving computational efficiency in high-dimensional data processing where full accuracy can be traded for speed.