May 8, 2018
Keio University
A research group at Keio University, consisting of Katsuhiro Endo (a first-year master's student at the Graduate School of Science and Technology), Katsufumi Tomobe (who completed the Doctoral Programs in March 2018), and Professor Kenji Yasuoka of the Faculty of Science and Technology, has proposed a new model that can generate long-term molecular simulation data by learning from short-term molecular simulation data using deep learning. They have demonstrated its usefulness through several verification experiments.
Molecular simulation is a method that can reproduce the movement of molecules. Its applications are extremely diverse, spanning biomaterials, polymers, and other materials, and it is used in the development of new materials and the elucidation of pathologies. However, simulating large molecules or for long periods requires extensive computational resources, which has been a drawback due to the difficulty of the calculations. In this study, we have proposed a new model that uses deep learning, a form of artificial intelligence (AI), to predict long-term simulation data using only short-term simulation data. This enables companies and research institutions that use molecular simulations to significantly improve their research and development efficiency by reducing the amount of simulation they need to perform. Furthermore, this proposal can be applied not only to molecular simulation data but also to general time-series data, with potential applications in various fields such as natural language processing, economic data, and motion data.
The results of this research were published on the website of the 32nd AAAI Conference on Artificial Intelligence (AAAI-18) on April 26, 2018 (local time).
For the full press release, please see below.