BENJAMIN NAYLOR

In quantum computing, matrix product states (a type of tensor train) are used to compactly represent high-dimensional systems. We took that idea and applied it to options pricing. When Fourier pricing, TTs allow us to efficiently store the payoff and characteristic function, cutting the complexity from $\mathcal{O}(N^d)$ to $\mathcal{O}(dN)$. This is especially useful for highly parameterised models and multi-asset cases. Our innovation was using a HyperNetwork to construct the TTs, which is significant because creating the TTs directly isn't always feasible—traditional methods often fail for non-linear systems due to rapid growth in the bond dimension. (Paper coming soon!)

Throughout my Master's degree I've been working on the Outlier platform helping to train AI models via RLHF. An expert in Mathematics, Finance, Coding, and Mechanical Engineering, I have worked on a range of projects and continue to do so. I often use AI to help me understand things so it's fun to be on the other side of that process, using my knowledge to try and outsmart it. (At least for the next few weeks before we all get replaced...)

I've been lucky to work on some really fun projects over the course of this year over a wide range of topics. From developing algorithmic trading strategies for treasury bonds to analysing the impact of tariffs using global trade atlas data to using dimensionality reduction for quantifying stock correlations, I've done a lot! My favourite was creating a transformer-based neural network for predicting stock returns. I used short term price features and wider market context features with two different transformer encoders to better capture market dynamics. It worked really well and it's here on my GitHub (along with most of my other projects).

I started here part-time during the final year of my undergrad, then joined full time after. I started as one of only five on the dev team, so was tasked with everything from creating a real-time speech-to-text model to implementing and deploying an entire serverless MLOps framework to fixing backend bugs in languages I didn't yet know. Being in a start-up was a great experience, you have to learn fast and keep moving because the company depends on it and no-one else will do it for you. I never thought I'd ever know anything about DevOps but here we are.