Machine Learning for Adaptive Discretization in Massive Multiscale Biomedical Modeling

Authors: Changnian Han (Stony Brook University), Prachi Gupta (Stony Brook University), Peng Zhang (Stony Brook University), Danny Bluestein (Stony Brook University), Yuefan Deng (Stony Brook University)

Abstract: For multiscale problems, traditional time stepping algorithms use a single smallest time stepsize in order to capture the finest details; using this scale leads to a significant waste of computing resources for simulating coarse-grained portion of the problem. To improve computing efficiency for multiscale modeling, we propose a novel state-driven adaptive time stepping (ATS) algorithm to automatically adapt the time stepsizes to the underlying biophysical phenomena at multiple scales. In this, we use a machine-learning based solution framework to classify and label these states for regulating the time stepsizes. We demonstrate the values of our ATS algorithm by assessing the accuracy and efficiency of a multiscale two-platelet aggregation simulation. By comparing with traditional algorithm for this simulation, our ATS algorithm significantly improves the efficiency while maintaining accuracy. Our novel ATS algorithm presents a more efficient framework for solving massive multiscale biomedical problems.

Best Poster Finalist (BP): no

Poster: pdf
Poster summary: PDF


Back to Poster Archive Listing