Blog

Do Diffusion Models Dream of Electric Planes?



Have you ever wondered what it might be like to let diffusion models go rogue and come up with crazy aircraft designs? Well, now is your chance to see what happens. In this blog post I summarize our recent paper: “Do Diffusion Models Dream of Electric Planes?” Discrete and Continuous Simulation-Based Inference for Aircraft Design.1 Motivation... Read more 06 Mar 2026 - 12 minute read

Training Neural Networks using Fomoh



This blog post focuses on training neural networks using the Fomoh library. Fomoh has a direct interface with PyTorch. We will see how to pass weights between PyTorch modules and Fomoh Models. This tutorial also shows you how to train a simple neural network. For those of you looking for an introduction into how to use Fomoh, please refer to my... Read more 06 Aug 2024 - 11 minute read

Second-Order Optimization without Backpropagation



Can we perform optimization without backpropagation? Yes. Recent work by Baydin et al. 2022 has shown that there is promise in using forward-mode automatic differentiation to optimize functions (including neural networks). This is interesting because it removes the need for a backwards pass and opens up the possibility of using energy-efficient ... Read more 24 Jul 2024 - 14 minute read

Scaling HMC to larger data sets



In this blog post I will demonstrate how to use hamiltorch for inference in Bayesian neural networks with larger data sets. Further details can be found in “Scaling Hamiltonian Monte Carlo Inference for Bayesian Neural Networks with Symmetric Splitting”. hamiltorch for Bayesian neural networks Towards the end of my previous post I gave a basic... Read more 14 Oct 2020 - 19 minute read

hamiltorch: a PyTorch Python package for sampling



What is hamiltorch? hamiltorch is a Python package that uses Hamiltonian Monte Carlo (HMC) to sample from probability distributions. As HMC requires gradients within its formulation, we built hamiltorch with a PyTorch backend to take advantage of the available automatic differentiation. Since hamiltorch is based on PyTorch, we ensured that ha... Read more 15 Jul 2019 - 13 minute read

3D Shape Modelling of Asteroids



This article describes my summer at the NASA Frontier Development Lab. Machine Learning and Planetary Defence As a researcher at NASA Frontier Development Lab (FDL), I spent my summer working alongside a team of experts with the aim of using machine learning to improve the process of modelling the 3D shapes of asteroids. FDL is an experimental... Read more 11 Sep 2017 - 4 minute read