The rich properties of nanomaterials can be a bane as much as a bonus to researchers keen to put them to good use. “Nanomaterials have all the challenges of molecules (such as finite sizes, surfaces and chemical functionalization), combined with the complexity of materials (such as defects, impurities and disorder),” explains Amanda Barnard, Chief Research Scientist in Data61 at the Commonwealth Scientific and Industrial Research Organisation (CSIRO) in Australia. “Machine learning is a powerful way to navigate that complexity and include all of these features in our predictions.”
Reporting in Journal of Physics: Materials Barnard and CSIRO colleague Baichuan Sun show that dimension reduction algorithms used in bioinformatics, finance, transport and social science can also help materials scientists to visualize data with a large number of defining features – descriptors – and so help to identify patterns in the structure-function relationships. These algorithms have largely gone below the radar of materials scientists as they were not developed with these applications in mind. However as the materials science specialists of a group specializing in data science and machine learning, Barnard and Sun are exposed to a far wider range of data tools than most researchers in groups specializing in materials science.
To read more, click here.