About

I am a lecturer in Applied Mathematics at Stellenbosch University, South Africa. I recently completed my PhD in Applied Mathematics (dissertation), where I focused on computational efficiency and likelihood reliability in continuous-time normalising flows. My research interests lie in deep generative models and their applications, and I enjoy asking questions about reliability, robustness, and generalisation in modern machine learning systems.

Before transitioning to academia, I worked as a software engineer in telecommunications, where I developed complex event processing applications and helped improve the frameworks that supported them. I also worked on database integration projects, focusing on optimising data workflows and system efficiency.

News

• May 2026 From July to August 2026, I’ll be joining the University of Glasgow as a visiting researcher, working with Dr Paul Henderson on generative models for computer vision and planning.

• May 2026 I’ve been invited by the Swedish Collegium for Advanced Study to join a new international research initiative, Human Futures - AI Transitions in a Global Context, in collaboration with Stellenbosch Institute for Advanced Study and Tokyo College.

• Feb 2026 New paper on the reliability of likelihoods from conditional flow matching has been published in IET Computer Vision.

Teaching

Below is a list of courses I am currently teaching and have taught in the past.

Publication and preprints

1. On the reliability of likelihoods from conditional flow matching generative models trained in feature space
   Shane Josias, Willie Brink
   IET Computer Vision, 20(1):e70061, 2026
   html

2. Multimodal base distributions in conditional flow matching generative models
   Shane Josias, Willie Brink
   British Machine Vision conference (BMVC), paper 492, 2024
   html | video | poster | supplementary
   

3. Multimodal base distributions for continuous-time normalising flows
   Shane Josias, Willie Brink
   The Symbiosis of Deep Learning and Differential Equations Workshop at NeurIPS, 2023
   html

4. Deep learning approaches to landmark detection in tsetse wing images
   Dylan Geldenhuys, Shane Josias, Willie Brink, and Mulanga Makhubele, Cang Hui, Pietro Landi, Jeremy Bingham, John Hargrove, Marijn Hazelbag
   PLOS Computational Biology, 2023
   html | data
   

5. Jacobian norm regularisation and conditioning in neural ODEs
   Shane Josias, Willie Brink
   Communications in Computer and Information Science, 2022
   html | slides | poster
   

6. Class-selective mini-batching and multitask learning for visual relationship recognition
   Shane Josias, Willie Brink
   SAIEE Africa Research Journal, 2021
   html
   

7. Batch construction and multitask learning in visual relationship recognition
   Shane Josias, Willie Brink
   International SAUPEC/RobMech/PRASA Conference, 2020
   html | slides