I am an associate professor of Applied Mathematics in the Faculty of Science at Stellenbosch University, South Africa. My research is primarily focused around applied machine learning. I have worked in the domains of probabilistic robotics and computer vision for many years, have some recent experience in computational linguistics and reinforcement learning, and currently have an interest mainly in representation learning and resource-efficient computer vision. I like working on applications for nature conservation and social impact, and I am passionate about building and serving machine learning communities all across the African continent.

I completed an MSc in Applied Mathematics at Stellenbosch in 2005, and a PhD in Computer Science at Sheffield Hallam University in the UK in 2008. After a short postdoc at South Africa's Council for Scientific and Industrial Research, I returned to Stellenbosch in 2009 where I have been teaching a variety of undergrad and postgrad courses in Applied Mathematics and Computer Science, and working with many wonderful grad students.

Office:  Room A311, General Engineering Building, Banghoek Road        Email:  wbrink [at] sun [dot] ac [dot] za        Phone:  +27 (0)21 808 4218

Maties Machine Learning
Herman Kamper and I created MML in 2018, in an effort to bring together researchers and students from across departments and faculties of Stellenbosch University. It is primarily a seminar series and discussion forum, and anyone at the university interested in machine learning is most welcome to join us.

Deep Learning Indaba
The Deep Learning Indaba is an organisation whose mission is to strengthen machine learning and artificial intelligence in Africa. We work towards the goal of Africans being not only observers and receivers of ongoing advances, but active shapers and owners of these technologies. Visit our website to find out more!

DeepMind Scholarships
I coordinate the DeepMind scholarship programme at Stellenbosch University, and am proud to represent the first African university on this programme. Applications for the 2022 scholarships at SU have closed, but do keep an eye out for future opportunities. More information here and here.

Courses I am teaching in 2022:

•   •   Mathematics for Machine Learning 811  (1st semester)

•   •   Computer Vision 842  (1st semester)

•   •   Computer Vision 792  (2nd semester)

•   •   Modelling in Mechanics 144  (2nd semester)

Other courses I've taught in the past:

•   •   Probability Theory and Statistics 114

•   •   Applied Differential Equations 244

•   •   Numerical Methods 262

•   •   Numerical Analysis 324

•   •   Applied Fourier Analysis 364

•   •   Computer Vision 364

•   •   Digital Image Processing 793

I am also the programme coordinator of a new one-year structured MSc in Machine Learning and Artificial Intelligence. Follow the link for more information!

•   •   Samuel Mensah (PhD, co-supervised with Bubacarr Bah)

•   •   Shane Josias (PhD)

•   •   Jama Mohamud (PhD, co-supervised with Ulrich Paquet)

•   •   Fred de Villiers (PhD, co-supervised with Benjamin Rosman)

•   •   Milanto Rasolofohery (MSc, co-supervised with Emmanuel Dufourq)

1. Jacobian norm regularisation and conditioning in neural ODEs
   Shane Josias, Willie Brink
   Southern African Conference for Artificial Intelligence Research (SACAIR), accepted, 2022

2. Learning to pay multiple attention with fully convolutional Transformers
   Samuel Mensah, Bubacarr Bah, Willie Brink
   Southern African Conference for Artificial Intelligence Research (SACAIR), accepted, 2022

3. Improving the performance of image captioning models trained on small datasets [link] [pdf]
   Mikkel du Plessis, Willie Brink
   Artificial Intelligence Research, CCIS vol. 1551, pp. 77-91, 2022

4. Machine learning models for stomatal conductance in multiple tree species across different forest biomes [link]
   Alta Saunders, David Drew, Willie Brink
   Trees, Forests and People, vol. 6, art. 100139, 2021

5. Mava: a research framework for distributed multi-agent reinforcement learning [arXiv]
   Arnu Pretorius, Kale-ab Tessera, Andries Smit, Claude Formanek, St John Grimbly, Kevin Eloff, Siphelele Danisa, Lawrence Francis, Jonathan Shock, Herman Kamper, Willie Brink, Herman Engelbrecht, Alexandre Laterre, Karim Beguir
   arXiv preprint, arXiv:2107.01460, 2021

6. Class-selective mini-batching and multitask learning for visual relationship recognition [link]
   Shane Josias, Willie Brink
   SAIEE Africa Research Journal, vol. 112, no. 2, pp. 99-109, 2021

7. BINet: a binary inpainting network for deep patch-based image compression [link] [arXiv]
   André Nortje, Willie Brink, Herman Engelbrecht, Herman Kamper
   Signal Processing: Image Communication, vol. 92, art. 116119, 2021

8. Towards the localisation of lesions in diabetic retinopathy [link] [arXiv]
   Samuel Mensah, Bubacarr Bah, Willie Brink
   Lecture Notes in Networks and Systems: Intelligent Computing, vol. 285, pp. 100-107, 2021

9. Link prediction in knowledge graphs using latent feature modelling and neural tensor factorisation [pdf]
   Luyolo Magangane, Willie Brink
   Southern African Conference for Artificial Intelligence Research (SACAIR), pp. 335-348, 2020

10. Image identification of Protea species with attributes and subgenus scaling [link] [pdf]
    Peter Thompson, Willie Brink
    IEEE Winter Conference on Applications of Computer Vision (WACV), pp. 2094-2102, 2020

11. Learning fine-grained control for mapless navigation [pdf]
    Fred de Villiers, Willie Brink
    SAUPEC/RobMech/PRASA Conference, pp. 666-671, 2020

12. Batch construction and multitask learning in visual relationship recognition [pdf]
    Shane Josias, Willie Brink
    SAUPEC/RobMech/PRASA Conference, pp. 713-718, 2020

13. Towards automating healthcare question answering in a noisy multilingual low-resource setting [link] [pdf]
    Jeanne Daniel, Willie Brink, Ryan Eloff, Charles Copley
    Meeting of the Association for Computational Linguistics (ACL), pp. 948-953, 2019

14. Analysis, prediction and comparison algorithms for water quality variables
    Reem Elmahdi, Willie Brink, Josefine Wilms
    Black in AI Workshop at the Conference on Neural Information Processing Systems (NeurIPS), 2018

15. Short-term stream flow forecasting at Australian river sites using data-driven regression techniques [pdf]
    Melise Steyn, Josefine Wilms, Willie Brink, Francois Smit
    International Work-Conference on Time Series (ITISE), pp. 865-876, 2017

16. A method for 3D stem analysis and its application on the occurrence of resin pockets in Pinus patula [link]
    Frederick Lerm, Melanie Blumentritt, Willie Brink, Brand Wessels
    European Journal of Forest Research, vol. 136, no. 3, pp. 411-420, 2017

17. A probabilistic graphical model approach to the structure-and-motion problem [arXiv] [pdf]
    Simon Streicher, Willie Brink, Johan du Preez
    PRASA-RobMech International Conference, paper 8, 2016

18. Text detection in natural images with convolutional neural networks and synthetic training data [pdf]
    Marco Grond, Willie Brink, Ben Herbst
    PRASA-RobMech International Conference, paper 20, 2016

19. The application of support vector regression for stream flow prediction on the Amazon basin [pdf]
    Melise du Toit, Josefine Wilms, Francois Smit, Willie Brink
    Conference of the South African Society for Atmospheric Sciences, pp. 25-28, 2016

20. Pose uncertainty in occupancy grids through Monte Carlo integration [link]
    Daniek Joubert, Willie Brink, Ben Herbst
    Journal of Intelligent & Robotic Systems, vol. 77, no. 1, pp. 5-16, 2015

21. Long-term tracking of multiple interacting pedestrians using a single camera [pdf]
    Mogomotsi Keaikitse, Willie Brink, Natasha Govender
    PRASA, RobMech and AfLaT International Joint Symposium, pp. 59-65, 2014

22. Pose uncertainty in occupancy grids through Monte Carlo integration [pdf]
    Daniek Joubert, Willie Brink, Ben Herbst
    International Conference on Advanced Robotics (ICAR), paper 90, 2013

23. FastSLAM with stereo vision [pdf]
    Wikus Brink, Willie Brink, Corné van Daalen
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 24-30, 2012

24. Probabilistic outlier removal for robust landmark identification in stereo vision based SLAM [pdf]
    Wikus Brink, Willie Brink, Corné van Daalen
    IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 2822-2827, 2012

25. Scene reconstruction from uncontrolled motion using a low cost 3D sensor [pdf]
    Pierre Joubert, Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 13-18, 2011

26. Stereo vision as a sensor for EKF SLAM [pdf]
    Wikus Brink, Willie Brink, Corné van Daalen
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 19-24, 2011

27. A mesh-based approach to incremental range image integration [pdf]
    Daniek Joubert, Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 74-79, 2011

28. Graph cut segmentation of range images into planar regions [pdf]
    Simon Muller, Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 108-113, 2011

29. Gain scheduling control of monocular vision-based human-following robot [pdf]
    Michael Burke, Willie Brink
    World Congress of the International Federation of Automatic Control (IFAC), pp. 8177-8182, 2011

30. Dense stereo correspondence for uncalibrated images in multiple view reconstruction [pdf]
    Willie Brink, Daniek Joubert, Francois Singels
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 39-44, 2010

31. Estimating target orientation with a single camera for use in a human-following robot [pdf]
    Michael Burke, Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 51-56, 2010

32. Multi-view 3D position estimation of sports players [pdf]
    Robbie Vos, Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 299-304, 2010

33. Applying Bayesian segmentation in volumetric silhouette-based reconstruction [pdf]
    Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 17-22, 2009

34. Real-time stereo reconstruction through hierarchical DP and LULU filtering [pdf]
    Francois Singels, Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 41-46, 2009

35. Combining motion detection and hierarchical particle filter tracking in a multi-player sports environment [pdf]
    Robbie Vos, Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 65-70, 2009

36. Real-time surface tracking with uncoded structured light [pdf]
    Willie Brink
    Annual Symposium of the Pattern Recognition Association of South Africa (PRASA), pp. 91-95, 2008

37. Indexing uncoded stripe patterns in structured light systems by maximum spanning trees [pdf]
    Willie Brink, Alan Robinson, Marcos Rodrigues
    British Machine Vision Conference (BMVC), pp. 575-584, 2008

38. Willmore energy vs area as criteria for mesh optimisation [pdf]
    Willie Brink, Lyuba Alboul
    Numerical Geometry, Grid Generation and High Performance Computing, pp. 6-12, 2006

39. Mesh optimisation based on Willmore energy [pdf]
    Lyuba Alboul, Willie Brink, Marcos Rodrigues
    European Workshop on Computational Geometry, pp. 133-136, 2006

• Mikkel du Plessis (MSc 2022): Low-resource image captioning

• Christiaan Louw (MSc 2022): Semi-supervised learning in computer vision

• Andries Smit (PhD 2022): Scaling multi-agent RL to 11-aside simulated robot soccer

• Wian Crous (MSc 2022): The class imbalance problem in computer vision [link]

• Evander Nyoni (MSc 2021): Neural machine translation for Southern African languages [link]

• Mulanga Makhubele (MSc 2021): FCNs for landmark detection in tsetse wing images [link]

• Luyolo Magangane (MSc 2020): Link prediction in knowledge graphs [link]

• Gregory Newman (MSc 2020): Video classification using deep learning [link]

• Reem Elmahdi (MSc 2020): Predicting water quality variables [link]

• Shane Josias (MSc 2020): Visual relationship recognition [link]

• Jeanne Daniel (MSc 2020): NLP for low-resource languages in healthcare [link]

• Peter Thompson (MSc 2020): Image identification of Protea species [link]

• Simbarashe Nyatsanga (MSc 2020): Automatic video captioning [link]

• Belinda Matebese (PhD 2019): Path planning using optimal control [link]

• Jaco Briers (MSc 2019): River flow routing using deep learning [link]

• Russell Kingwill (MSc 2019): Forecasting South African basic fuel prices [link]

• Jacques Marais (MSc 2018): Elephant detection in aerial images [link]

• Melise Steyn (MSc 2018): Stream flow forecasting using machine learning [link]

• Marco Grond (MSc 2017): Text detection in natural images [link]

• Simon Streicher (MSc 2016): Structure-from-motion with PGMs [link]

• Gideon Zuurmond (MSc 2015): Camera calibration through moiré pattern analysis [link]

• Alwyn Burger (MEng 2015): Occupancy grid mapping using stereo vision [link]

• Pieter Holtzhausen (PhD 2015): Video surveillance incorporating PTZ cameras [link]

• Lloyd Hughes (MSc 2014): Mobile camera pose estimation [link]

• Pierre Joubert (MSc 2014): People detection and tracking in RGB and IR [link]

• Mogomotsi Keaikitse (MSc 2014): Long-term tracking of multiple pedestrians [link]

• Simon Muller (MSc 2013): Planar segmentation of range images [link]

• Wikus Brink (MEng 2012): Stereo vision for SLAM [link]

• Daniek Joubert (MSc 2012): Occupancy grid mapping with pose uncertainty [link]

• Michael Burke (MEng 2011): Visual control for a human-following robot [link]

• Robbie Vos (MSc 2010): Multi-view tracking of sports players

• Francois Singels (MSc 2010): Real-time stereo reconstruction [link]