91Ö±²¥

Using Gaussian Processes to discover hidden correlations in experimental data by Professor Nigel Clarke

Date:  Wednesday 25th September

Location:  Workroom 1 (Ground Floor), The Diamond, 32 Leavygreave Road, S3 7RD

Agenda: 

13:10 - Networking 

13:15 - Refreshments 

13:20 - Talk

13:40 - Q&A's

Abstract:

De-mixing is one of the most ubiquitous examples of material self-assembly, occurring frequently in complex fluids and living systems, as well as being of great importance to the development of metallic alloys. It has enabled the development of multi-phase polymer blends and composites for use in sophisticated applications, including structural aerospace components, flexible solar cells, and filtration membranes. Even though superior functionality is derived from the microstructure, our understanding of the correlations between microstructure characteristics and material properties remains largely empirical.

We have explored the use of machine learning as a promising route to the model-free extraction of microstructure characteristics from experimental data. We have focussed on the process of spinodal decomposition in binary polymer blends as an example, using numerically generated data to test our approach. The use of Gaussian process regression has enabled us to discover previously unknown predictors of structural characteristics hidden in the experimental data. I will also discuss progress and challenges in our use of equation learning ML techniques to describe dynamic processes such as phase separation and pattern formation.