| Budget Amount *help |
¥43,420,000 (Direct Cost: ¥33,400,000、Indirect Cost: ¥10,020,000)
Fiscal Year 2021: ¥10,530,000 (Direct Cost: ¥8,100,000、Indirect Cost: ¥2,430,000)
Fiscal Year 2020: ¥10,400,000 (Direct Cost: ¥8,000,000、Indirect Cost: ¥2,400,000)
Fiscal Year 2019: ¥10,400,000 (Direct Cost: ¥8,000,000、Indirect Cost: ¥2,400,000)
Fiscal Year 2018: ¥6,630,000 (Direct Cost: ¥5,100,000、Indirect Cost: ¥1,530,000)
|
|---|
| Outline of Final Research Achievements |
We studied the discrete preimage problem, in which prediction functions for discrete data are obtained using machine learning methods and then novel discrete data are obtained by computing the preimages for given properties. In this project, we developed methods for the problem based on mixed integer linear programming with focusing on design of chemical structures. As for the prediction functions, we mainly used artificial neural networks, and developed novel representation models such as the two-layered model for efficiently handle chemical structures. As a result, our developed methods could compute preimages for moderate-size chemical structures. From a theoretical viewpoint, we obtained several results on discrete models, which include analysis of relations between the compression ratio and the numbers of layers and nodes in autoencoders using linear threshold activation functions.
|
