MINDS CDT student explains all about Dynamic Causal Modelling

MINDS CDT student Elliot Stein, completed a 3-month internship at Embecosm in summer 2022, and wrote an article effectively communicating to non-experts all about Dynamic Causal Modelling and how it can be optimised using Variational Laplace and Bayesian Model Selection to solve a dynamical systems problem. The article is published on the online publishing platform medium.com.

Elliot’s PhD project is about ‘Convolutional Neural Networks for Auditory Attention in Robots’ and is supervised by Dr Christine Evers, Dr Jonathon Hare and Prof Chris Freeman. This 3-month internship allowed Elliot to dive deeper into Bayesian techniques, both in theory and practise, in a different setting to his typical PhD work. The project, called dcEmb, sought to take advances in Bayesian techniques (Variational Laplace) from the field of computational neuroscience, where they originated, into the mainstream. The motivation being that Bayesian methods address two key limitations of the deep learning methods that dominate the headlines (and Elliot’s PhD): uncertainty and interpretability.

Additionally, working side by side with industry professionals in order to build up an open-source library of C++ code, served to improve his skills as a researcher, programmer, and team player, which will benefit his own PhD work immeasurably.