研究・研究者
研究会・セミナー
| 演 題 | 「Machine Learning for Molecular Properties and Chemistry」 |
|---|---|
| 日 時 | 2019年02月26日(火) 13:30 |
| 講演者 | Sergei Tretiak博士 （ロスアラモス国立研究所） |
| 場 所 | 分子科学研究所 研究棟301室 |
| 概 要 |
Computer simulation is foundational to modern theoretical chemistry. Calculating physical properties of molecular systems is paramount to drug development, bio-molecular simulation and materials design. These calculations must be fast and accurate to allow for high-throughput studies of small molecules or the simulation of large systems of many thousands of atoms. Traditional methods for obtaining these properties are based on classical physics or quantum mechanical (QM) methods. Classical techniques are computationally efficient but have questionable accuracy when used outside of their direct parametrization sets. QM based methods tend to be more accurate than their classical counterparts, however, their computational scaling is frequently prohibitively expensive to treat realistic systems. Machine learning-based (ML) QM property predictors are capable of fitting directly to QM data with low error while remaining computationally as fast as classical techniques. We present our work on developing and applying various models for QM property prediction, which are trained to large QM datasets then shown to generalize well outside of the training set. The targeted properties are include ground state potential energy and forces, various atomic charge schemes, dipoles and quadruples, Infrared spectra, reduced Hamiltonians and a single model for singlet and triplet state energies and forces. Our results show the applicability of these accurate ML property predictors to systems many times larger than those in the training set with a several magnitude speedup over reference QM methods, an exciting prospect for computational sciences.
References:
Dr. Sergei Tretiak |
| お問合せ先 | 石崎 章仁 |
