About me: I'm a DPhil candidate in the CDT for Random Systems at the University of Oxford, visiting student at Imperial College London, and part-time quantitative researcher at J.P. Morgan Chase. My current research concerns the robustification of deep hedging algorithms, which overlaps with many areas of mathematics, finance, and computer science: see keywords below. Previous interests include activation sparsity and initialisation problems for very deep neural networks, and ergodic theory.

Keywords in no particular order: Ambiguity Aversion, Deep Learning, Deep Hedging, Distributional Robustness, Quantitative Finance, Mathematical Finance, Hedging Problems, Exotic Derivatives and Structured Products.

Publications:

• Deep Neural Network Initialization with Sparsity Inducing Activations;

with I. Price, N. Daultry Ball, S. C. H. Lam & J. Tanner;

Published in the International Conference on Learning Representations (ICLR) 2024.

Shifting ReLU is an intuitive way to encourage sparsity, however we find that it also needs to be clipped to have any chance of training. By initialising on the edge of chaos, we observe almost the same accuracy as our benchmarks, but with 85% activation sparsity.

See its webpage here.

• On Counting Cards and Learning Optimal Deviations from Blackjack Strategies;

solo author;

Published in SIGBOVIK 2025. Won the SIGBOVIK 2025 Spirit Award.

Optimal blackjack strategies are easily computed with modern software. However, their objective is to maximise expected value, not maximise profit. Since rent can't be paid in edge, we modify the optimal strategy to become optimaler. (NB: This is for a satirical journal.)