Upstream ML stack integration
Helped ship bleeding-edge ML framework builds for Arm CPUs by pulling together upstreamed changes across the stack (PyTorch, oneDNN, Arm Compute Library, OpenBLAS) into Tool Solutions.
Senior Software Engineer in AI Frameworks.
I work on the optimisation of machine-learning workloads for Arm CPUs...
...with deep roots in imaging algorithms and applied mathematics (computational fluid dynamics).
Introduction
I work on optimising machine learning (ML) workloads for Arm CPUs. Much of my work sits in and around open-source ML frameworks such as PyTorch, oneDNN, and Arm Compute Library, spanning performance optimisation, build and release tooling, and upstream enablement.
Before moving to the AI Frameworks team, I worked on imaging algorithms across the ISP pipeline and embedded ML, which inspired my love for photography.
Before Arm, I completed a PhD in Applied Mathematics, focused on computational fluid dynamics and large-scale numerical methods.
Outside of work, I'm a part-time maintainer of
oomph-lib, an open-source finite-element library for PDEs, where I rewrote the project's Autotools
build system in CMake from scratch.
Recent work
Restoring external Docker delivery
Unblocked external delivery of Docker images by working through compliance requirements (SBOM maintenance, release process, security expectations) and then automating the boring parts.
Faster builds and developer workflows
Made builds faster (including ~2.7× CI improvements in places) by improving build tooling and contributing upstream where needed.
Fixing real model bottlenecks
Investigated and fixed performance issues in the PyTorch → oneDNN → Arm Compute path (e.g. enabling dispatch paths) to improve real model workloads.
Research & Publications
I completed my PhD in applied mathematics at the University of Manchester, with a focus on fluid dynamics and the numerical solution of time-periodic solutions to partial differential equations (otherwise known as PDEs).
For modest Reynolds numbers (Re ≤ 100), a fixed cylinder sheds vortices in a classical 2S pattern, known widely as the Kármán vortex street. When the cylinder oscillates with a period close to the natural shedding frequency, increasing the oscillation amplitude leads to a transition to a different, asymmetric wake pattern (the P+S pattern). A central question of the thesis was whether this transition arises through a continuous (topological) evolution of the flow or via bifurcations of the Navier–Stokes equations.
ORCID: 0000-0001-9359-9814
Thesis Direct computation of time-periodic solutions of PDEs University of Manchester PhD Publication Topological bifurcations in the transition from two single vortices to a pair and a single vortex in the periodic wake behind an oscillating cylinder Journal of Fluid Mechanics (2022) Publication Spatio-temporal symmetry breaking in the flow past an oscillating cylinder Journal of Fluid Mechanics (2021)
Get in Touch
Looking to collaborate? Let's talk.