![Queens tower](http://www.imperial.ac.uk/ImageCropToolT4/imageTool/uploaded-images/newseventsimage_1691407340914_mainnews2012_x1.jpg)
![](http://www.imperial.ac.uk/assets/news/img/main_image_shd.png)
PhD student Tycho van der Ouderaa, emerged victorious in the 14th edition of Qualcomm Innovation Fellowship (QIF) Europe
Tycho van der Ouderaa, a PhD student from Imperial College London’s Department of Computing, emerged victorious in the 14th edition of Qualcomm Innovation Fellowship (QIF) Europe. Presented by Qualcomm Technologies, Inc., the award recognises and mentors exceptional engineering PhD students across Europe, India, and the United States for their innovative contributions to artificial intelligence and cybersecurity.
Out of the twelve finalists representing leading institutions such as ETH Zurich, Tübingen University, University of Cambridge, University of Oxford, CISPA, and EPF Lausanne, Tycho van der Ouderaa was selected as one of the four winners. The award includes $40,000 of additional research and maintenance funds, as well as a mentor from the Qualcomm Technologies team who will follow the research closely, and provide guidance. The recognition and support from this fellowship are set to accelerate his research efforts at Imperial College London’s Department of Computing.
Tycho’s proposal, “Learning Equivariances from Data,” focuses on leveraging large neural networks to address real-world challenges by enhancing their performance and generalisation abilities through the automatic discovery of compute and parameter-efficient architectures. This research delves into the crucial role of inductive biases, such as invariance and symmetries, in these neural network models. While existing works have explored extensions to various symmetry groups and domains, the architecture and group structures they employ are often fixed, necessitating manual selection or costly cross-validation processes.
Tycho’s innovative approach, however, proposes a method of learning equivariances and the corresponding neural structures directly from training data, which eliminates manual selection. Through a combination of flexible parameterisations and an amenable objective function capable of learning symmetries, his research aims to simplify the process of finding suitable architectures, making it as straightforward as learning weights.
The research will be carried out in the Occam Machine Learning Group, together with Tycho’s supervisor Dr Mark van der Wilk. The research group focusses on developing machine learning methods that can spot general patterns that help to make better predictions in new and unseen situations. The techniques Tycho is developing may also help other’s in the group to find causally significant structure, or to help reduce the data needed for predictive models in science and engineering applications.
The Department of Computing congratulates Tycho van der Ouderaa on this well-deserved achievement and celebrates his contributions to the cutting-edge advancements in the field of machine learning. His success further solidifies the college’s position as a hub of excellence for fostering groundbreaking research and innovation in the realm of computing and artificial intelligence.
Article text (excluding photos or graphics) © Imperial College London.
Photos and graphics subject to third party copyright used with permission or © Imperial College London.
Reporter
![Mr Ahmed Idle](http://www.imperial.ac.uk/ImageCropToolT4/imageTool/uploaded-images/img-20200913-195942_1647961157692_x1.jpg)
Mr Ahmed Idle
Department of Computing