My research focuses on neural architectures, specifically what makes which architecture successful when. It turns out that geometry plays an important role: probing the neural network to see what inputs it finds indistinguishable characterises important qualities of the network. I’m interested in finding smart ways to find these invariances autonomously. In short, I use applied category theory with geometric deep learning to develop new theories and techniques in neural architecture search.
The blend of a cohort-based experience and frequent touchpoints with industry through the centre’s partners are the two main reasons that I gravitated towards the CDT. King’s and Imperial are two exciting intellectual and cultural hubs to be part of and I’ve learned a lot just from interacting with my peers.
I was drawn to this research project after having seen, both in my masters and in my previous work in consulting, how overbearing the issue of interpretability was to the deployment of AI systems. My hunch was that the definitive answer lies close to deep learning theory and understanding what happens at a neural level – and I was soon captivated by the problem of architecture.
Undergraduate Qualification: BSc Hons Mathematics @ University of Nottingham
Masters Qualification: MSc Banking and International Finance @ Bayes Business School (formerly Cass) – City University London