| 研究実績の概要 |
We continued our research into developing physics informed machine learning methods to tackle complex soft matter problems. Our work on learning constitutive relations for accelerating multi-scale-simulations, originally developed for non-entangled polymer melts, has been extended and applied to the canonical Doi-Takimoto polymer entanglement model with considerable success. This work has been carried out with a Master course student I have supervised, who has now entered the doctoral course, where we will continue this research.
In addition, we have also developed a Bayesian Stokes flow solver, based on Gaussian Processes, which allows us to simultaneously solve for the fluid velocity and pressure fields, given only knowledge of the velocity/pressure at the boundaries. This method will be useful when traditional methods fail. In particular, since it allows for incorporating missing and/or noisy data, it will be helpful when analyzing experimental data.
Finally, we have also developed Machine Learning techniques to tackle (inverse) optimal control problems, allowing us to infer hidden utilities from observed behaviour. This particular study comes out of a collaboration to understand the optimal government intervention strategy during an epidemic, but it has broad applications in science, engineering, and biology.
The lessons learned from all these studies will be invaluable to learn the constitutive relations of cellular tissues from a given microscopic model, which will allow us to bride the gap between detailed cell-level models, and coarse-grained continuum models.
|
