GSNS-masterscriptie winnaar: Wouter Ubbink (Artificial Intelligence)

Wouter, your thesis is entitled: Improving image recognition for species identification by modeling ecological context.

One of the typical uses of AI nowadays is image recognition, i.e., understanding what items are shown in an image. Novel computer vision techniques are steadily improving the recognition of species from images. Despite these advances, several bottlenecks remain. An important challenge is to deal with strongly skewed distributions of species, which brings what we call bias in the data. This bias results into reduced accuracy in the identification of rare plants and animals.

Yet, those rare species and their occurrence is of more interest when studying species distribution. Wouter has introduced creative new ways of combatting this issue by incorporating (spatial) ecological context, providing a solution that is orthogonal to computer vision advances. He used the observation location to provide the Machine learning model with variables on the local habitat and the species in neighboring observations. Then he employed a transformer architecture to train the model on that data.

Experimental comparison with existing baselines has indicated that the developed solution produces significantly better results. Wouter's innovative technical solutions will benefit both scientists and the general public. The thesis work took place at Naturalis, the owner of a phone app called ObsIdentify that on a weekly basis is used by more than 10K nature enthusiasts to report findings of (rare) plants and animals through photos.

AI technology consumes these photos and provide back the enthusiasts with a classification of their findings. Wouter used this data, and his solution is now on its way to be incorporated into the app. In his thesis document, he provides an elegant motivation, and a comprehensive detailed and complete description of the related work and his solutions. 

The thesis is well structured and the method of presentation is such that even less technical readers can appreciate the impact that this work will have to both science and society. Currently, Wouter is preparing a journal publication, and Wouter was offered an extension to his appointment at Naturalis, where he is preparing the inclusion of his innovative algorithm into the widely deployed ObsIdentify app.

Congratulations!

Marien, you are the second of our two runners up. Your thesis is entitled: Computation in Nested Closed Timelike Curves

Computers nowadays are based on a computational model that has been introduced by Turing, known as Turing machine, which is based on classical intuitions on how the universe works. But the new physical theories of the 20th-century—quantum mechanics and general relativity—have shown that the universe fundamentally behaves differently from how it seems to work in ordinary situations. It may be possible to use this behavior to create computers that are more

powerful than Turing machines. Yet these machines do not seem to solve new problems. In this thesis, it is argued that much harder problems can be solved on computers in certain situations. It is shown that the problems a computer can solve depend on how many time loops there are in the universe and how these time loops are connected. The thesis gives more information about what can and cannot be calculated in our universe and under which conditions.

The thesis is an excellent piece of work where complex concepts are explained naturally and often illustrated nicely. The reader is left without gaps and the text flows smoothly. At the end of the thesis, it becomes clear to the reader what contribution the thesis has made and that it has advanced the existing state of the art.

Congratulations!

Annemijn, you are one of the two runners up. Your thesis is entitled: 

Escaping the flatland: Design and synthesis of 3-dimensional multinuclear complexes for catalysis.

Metal complexes containing multiple metal centers can be very efficient homogeneous catalysts and have thus attracted increasing interest in recent years. However, the influence of the number of metal centers in a complex on its reactivity is still poorly understood. In her MSc research project, Annemijn van Koten addressed this challenge by designing and synthesizing a series of similar ligands, which can hold two, three or four metal centers.

This allowed her to systematically investigate the influence of the number of metal centers in the complex on its reactivity. The work described in the thesis is thus original and contributes toward a deeper understanding of multinuclear homogeneous catalysts. In the thesis, Annemijn demonstrated excellent experimental skills,  she independently analyzed data with a critical and reflective attitude, and related it to the bigger picture in a way which enabled her to effectively come up with follow-up experiments.

The thesis is very well-written, clearly organized, and of a very high scientific level. The Introduction provides an excellent description of the background knowledge and of the importance of the research carried out during the MSc project. The theory section reflects a very deep understanding of the fundamental concepts behind the project and of the state-of-the-art in the field. The Experimental work is described in a clear and comprehensive way, ensuring that it can be reproduced by others.

The Results are presented and discussed in a clear and scientifically rigorous way, making use of high-quality and insightful figures, and demonstrating excellent command and a deep understanding of the current body of knowledge. The conclusions are sound and fully supported by the data presented, and they provide significant advances in the field of homogeneous catalysis.

Congratulations!