Challenge: Developing deep meta-learning algorithms to model and understand learning curve patterns in machine learning.
Impact: Faster, better, more cost-efficient training and tuning of learning
Read the vacancy textIn my lab, we often give short talks about selected papers we like. You can find the papers I've given talks about below.
My focus is on education innovation and learning curve research. With regards to education innovation, we are working to create a community for machine learning teachers in TU Delft to improve collaborations in teaching and to kickstart developing open education material for reuse. In my teaching I like to employ interactive Python widgets in my class to stimulate students' understanding and am interested in developing more interactive education.
I helped develop the new AI minor program which started in September 2021. For this program I have developed two courses from scratch and lectured them: 'Introduction to Machine Learning' (TI3145TU) and 'Capstone Applied AI project' (TI3150TU). In this program engineers with various backgrounds learn the basics of AI and machine learning, and apply the learned techniques in the field for their major.
I am also involved in the Master elective Fundamentals of Artificial Intelligence, and have developed a MOOC (Massive Online Open Course) on Supervised Machine Learning together with Hanne Kekkonen. In collaboration with other lecturers we have designed a second MOOC on unsupervised learning, reinforcement learning and deep learning.
Supervisor: Marco Loog, Promotor: Elmar Eiseman.
My PhD focused on three theoretical machine learning topics: explainability, active learning and learning curves. The main take-aways are (TLDR):
- Strictly tighter generalization bounds do not imply better performance.
- Explanations provided by Grad-CAM can be misleading.
- Even in simple settings more data can lead to worse performance.
- We provide ideas to construct learners that always improve with more data.
Supervisor: Marco Loog.
During my masters in Delft I discovered my passion for Machine Learning.
I spent a long time on my masters project, simply because I loved it so much.
My supervisors were Marco Loog and Jesse Krijthe,
and we studied the problem of Active Learning using generalization bounds,
in particular using the Discrepancy measure.
While I really enjoyed physics in my bachelor, in the end of my bachelor I fell in love with computer science (CS). During my bachelor project I built an application to control an electron microscope (EM) to record a giant mosaic of images as fast and accurate as possible. I also worked together with Frank Faas to develop a basic application to annotate and view gigabyte-size EM images. You can view the zebrafish dataset of the KosterLab research group here, which was in part annotated with help of software that Frank and I wrote. I spent the fourth year of my (physics) bachelor studying Computer Science in order to switch to my CS master at TU Delft.