Accelerated search for materials with targeted properties by adaptive design - 0 views
-
jcunha on 10 May 16There has been much recent interest in accelerating materials discovery. High-throughput calculations and combinatorial experiments have been the approaches of choice to narrow the search space. The emphasis has largely been on feature or descriptor selection or the use of regression tools, such as least squares, to predict properties. The regression studies have been hampered by small data sets, large model or prediction uncertainties and extrapolation to a vast unexplored chemical space with little or no experimental feedback to validate the predictions. Thus, they are prone to be suboptimal. Here an adaptive design approach is used that provides a robust, guided basis for the selection of the next material for experimental measurements by using uncertainties and maximizing the 'expected improvement' from the best-so-far material in an iterative loop with feedback from experiments. It balances the goal of searching materials likely to have the best property (exploitation) with the need to explore parts of the search space with fewer sampling points and greater uncertainty.
-
jcunha on 11 May 16Can AI create the next wonder material? http://www.nature.com/news/can-artificial-intelligence-create-the-next-wonder-material-1.19850?WT.ec_id=NATURE-20160505&spMailingID=51301208&spUserID=MTEzODM0NjYzMzgS1&spJobID=920498312&spReportId=OTIwNDk4MzEyS0