How can AI be used to search for new materials?
Machine learning and Artificial Intelligence (AI) are transforming all aspects of daily life, but how can we use this powerful technology to search for new materials that will transform industries? The answer to this question was addressed at The Advanced Materials Show 2019 by Dr. Gareth Conduit, Chief Technology Officer at Intellegens - the Cambridge-based company with an artificial intelligence solution for the materials sector.
Taking part in The Advanced Materials Show 2019 Conference, Dr. Conduit explained how deep learning algorithms can help organisations drive more value from their experimental data. Using real life examples, Dr. Conduit demonstrated how existing data can be merged with computer simulations and information available in the public domain, to drastically cut the amount of time and money needed to develop new materials.
As a case study, Dr. Conduit looked at how AI played a part in the design of a nanoscale material for thermometry, which resulted in a new material that has target properties that are two orders of magnitude greater than any commercially available alternative.
Commenting, Dr. Conduit said: “Worldwide there are millions of materials available commercially that are characterised by hundreds of different properties. Using traditional techniques to explore the information we know about these materials, to come up with new substances, substrates and systems, is a painstaking process that can take months if not years. Learning the underlying correlations in existing materials data, to estimate missing properties, our Alchemite AI engine can quickly, efficiently and accurately propose new materials with target properties - speeding up the development process. The potential for this technology in the advanced materials sector is huge.”
Intellegens’ Alchemite engine is designed to work with sparse or noisy data and is capable of learning from datasets as little as 0.05% complete. The Alchemite algorithm has proven commercial applications in materials design and can create trained models that can be used for predictions, error detection and parameter optimisation (design).