Bio
Abhra Chaudhuri is a PhD student in Computer Science at the University of Exeter. He is also an Intern at the Fujitsu Research of Europe, working in the Autonomous Learning group.
In general, his interest lies in understanding the representational properties of learning algorithms from both theoretical and empirical perspectives. Specifically, he works on leveraging ideas from the theory of metric spaces, information theory, and graph theory to analyse deep neural networks.
His PhD research is on Multi-View Representation Learning in Computer Vision. At Fujitsu, he is currently working on making Vision Transformers generalize to out-of-distribution segmentation tasks.
Research interests
At the Alan Turing Institute, Abhra Chaudhuri aims to investigate the geometric properties of self-supervised representation spaces. He wishes to expand on his PhD research on relational representations by exploring ways of interpreting the behaviour of models that use multi-view relational invariances to generalize across domains. He also wants to understand the theoretical properties of such algorithms that give them the versatility of being transferable to a diverse variety of downstream tasks.