We explore a powerful decomposition of popular QML models as Fourier series.
More specifically, we break down the loss function into its frequency components and show (in experiment) that gradients at each frequency follow a property of that frequency called "redundancy".
That is, where frequencies have high redundancy, models learn them faster!
You can kind of control the redundancy by choosing how you embed your data into the quantum circuit, which, combined with the above observation, makes it a simple yet powerful tool for model design.
We then turn to explaining our results with some maths.
2025, UCL
We consider an important detail in quantum error correction experiments - the reset after an error correction round.
This paper offers important insights into when experiments should or should not opt to reset their qubits.
My first (intense) exposure to the world of fault-tolerant quantum computing.
2024 (published 2025 in npj Quantum Information), Riverlane
We use a quantum autoencoder to look for a needle (evidence of new physics) in a haystack (simulated events at the LHC).
We improve on two existing studies and begin to investigate the magic (! this is a real metric used in quantum information) and entanglement of the circuits used.
2024 (published 2025 in Quantum Machine Intelligence), UCL
A collaboration with the Astrophysics Group at UCL.
First time we got results from a real quantum device.
2023 (published in RASTI), UCL
We calculate what is called the 'kernel matrix' on a quantum device and use that to define
a machine learning model (a support vector machine) which is used to classify small connections
between hits left by particles in a detector as belonging (or not) to a particle track.
2022 (published 2024 in Physical Review D), UCL
Masters project at Lancaster University
2021, Lancaster University
3rd year python project at Lancaster University
2020, Lancaster University