Efficient methods for chemical space explorations and high-fidelity predictions are essential to the discovery of new materials. Fundamental bottlenecks to new chemical and material discoveries include: (1) our inability to rapidly and accurately determine the properties of new materials, and (2) a failure to recognize how different properties are correlated. Here, we usher in a paradigm based on �machine learning� (familiar to those in the cognitive sciences arena) to address both the above aspects to a startling level of accuracy and efficiency. Machine learning is an automated pathway for learning effectively from past knowledge, and significantly alleviates the burden of expensive and laborious computations and experimentation. While it has already been enormously successful in domains such as game theory, facial/fingerprint recognition, weather forecasting, etc., it has barely been exploited in the materials sciences. The present development thus constitutes a paradigm shift in how past data can be mined and understood, and can lead to the automated discovery of hidden rules, laws and correlations in material systems. We anticipate this contribution to impact conventional materials chemistry as well as physics, nanotechnology and engineering design.

Efficient chemical space exploration and accelerated property predictions using machine learning

246th National Meeting (2013)

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.