Prediction of repeat unit of optimal polymer by Bayesian optimization
- PDF / 304,784 Bytes
- 6 Pages / 432 x 648 pts Page_size
- 69 Downloads / 209 Views
MRS Advances © 2019 Materials Research Society DOI: 10.1557/adv.2019.57
Prediction of repeat unit of optimal polymer by Bayesian optimization Takuya Minami1,2, Masaaki Kawata3, Toshio Fujita1,2, Katsumi Murofushi2, Hiroshi Uchida2, Kazuhiro Omori2, and Yoshishige Okuno2 1
Research Association of High-Throughput Design and Development for Advanced Functional Materials, Tsukuba, Ibaraki 305-8568, Japan
2
Showa Denko K.K., Minato-ku, Tokyo 105-8518, Japan.
3 National Institute of Advanced Industrial Science and Technology (AIST), Tsukuba, Ibaraki 305-8568, Japan
Abstract
Design processes of functional polymers were accelerated by adopting the Bayesian optimization; the number of trials in the process was substantially reduced. The optimization process was more than forty time accelerated to find out the target polymer compared to the random selection. The optimization efficiency was found to be successfully improved by utilizing the standard deviation of predicted probability distribution of objective function. The performance of the method was robust for dataset size in the analysis; the target polymer could be found even for a small training dataset. The proposed method is a promising tool for the high-performance polymer design, and a wide range of its applications will be expected in the polymer industry.
INTRODUCTION: Development time and cost of new functional materials are important issues affecting industrial competitiveness. Recently, materials informatics has attracted much attention as an approach to accelerate the design process of advanced functional materials [1] [2] [3] [4]. This data-driven approach is expected to reduce the number of unnecessary experiments by preliminary predicting promising candidate materials.
1125
Downloaded from https://www.cambridge.org/core. Columbia University Libraries, on 03 Aug 2019 at 19:03:08, subject to the Cambridge Core terms of use, available at https://www.cambridge.org/core/terms. https://doi.org/10.1557/adv.2019.57
Bayesian optimization is one of the global optimization methods [5]. This method predicts next candidate materials to be examined by utilizing both of mean and standard deviation of predicted Gaussian distribution of objective function. The Bayesian optimization has ever been employed in many researches and succeeded to accelerate design processes of functional materials such as inorganic low-thermal-conductivity compound [6] and rechargeable battery [7]. However, there are unfortunately few reports on the Bayesian optimization applied to functional polymers, which are ones of the important materials for industry. In the present study, we aimed to numerically demonstrate the effectiveness of Bayesian optimization in polymer design. To this end, the virtual polymer design experiment maximizing glass transition temperature (Tg) was performed. The efficiency of the Bayesian optimization was numerically discussed based on the number of trials required to find out a target polymer. To clarify the advantage of the Bayesian optimization, the result
Data Loading...
