The idea of versatile models able to identify and utilise complex patterns in data has always been an intriguing part of Information Engineering. Whilst studying at the University of Cambridge, I conducted multiple research projects with a special focus on a family of such models called deep neural networks. Applied to areas ranging from speaker diarisation to active learning, I slowly started to understand the wide ranging applicability, but also drawbacks of this family of approaches. One of these drawbacks is the lack of uncertainty estimation, the ability to account a model’s prediction to various causes of uncertainty, and this is where I will focus my research. I plan on developing efficient neural networks that will be able to quantify different types of uncertainties, and apply them in both high-risk and sensitive domains, where previously, such models would not have been applicable due to a limited understanding of how and when they work.
University of Cambridge Eng. Tripos (Information Eng.) 2021