Dissertation Award 2023: Daniel Johansen Trosten
The Norwegian Artificial Intelligence Society presents the NAIS Dissertation Award to Daniel Johansen Trosten for his PhD thesis Improving Representation Learning for Deep Clustering and Few-shot Learning. The thesis was supervised by Michael C. Kampffmeyer at the University in Tromsø. Dr. Trosten studied two areas of deep learning that deal with limited data. First, he explored how to improve the quality of representations learned in deep clustering, especially for complex data with multiple views. Second, he investigated few-shot learning, where models learn from very few examples. The focus was on the role of the geometrical shape of the data representation in this context. Many recent methods use a specific shape called a hypersphere, and Trosten aimed to understand why this is a good choice for few-shot learning tasks. The thesis stood out because the work delves into critical aspects like representational alignment and the geometry of embeddings, leading to groundbreaking and state-of-the-art approaches in both domains. The impact is evident not only in the impressive number of publications in top conferences and journals – including first-authored works at Conference on Computer Vision and Pattern Recognition (CVPR) and International Journal of Computer Vision (IJCV) – but also in the significant citations Read more…