Although much buzz exists around machine learning due to the fact it has been used to build state of the art systems for solving many complicated problems; computer vision, natural language processing, machine translation etc. machine learning has, yet, not led to such advances within physics.
Whilst the algorithms behind these advances are relatively general we suffer from the problem of how to correctly arrange our input data to make use of them. I work on how to construct descriptors that suitably describe physical systems in the hope that we will then be able to utilise the demonstrated potential of machine learning algorithms in materials design problems.
I investigate the inference of structure function relationships in cuprate superconductors. Uncovering the dependencies of these systems on their structural and electronic properties will hopefully enable us to tune their properties and produce materials with higher critical temperatures.
In Plain English
I roll balls down complicated high dimensional hills via the magic of machine learning to try uncovering useful relationships between the structures and properties of different materials.
Currently I am looking to construct balls that are more suitable for rolling as in general we are not very good at making balls that roll well for different materials.
I am particularly focussed on making superconductor themed balls as finding the bottom of the high dimensional superconductor hill would have significant impact on transport and power distribution.