FURI | Spring 2020

Learning Unknown Physics Behind Cellular Dynamics Using Time-Variant Neural Network

The objective of the research is to develop a modified Hamiltonian ordinary differential equation neural network (MHODE) that models the dynamics of a multi body system with non-conservative energy. This model will be validated through the comparison of the total energy profile over time between the learned physics and the ground truth. The proposed machine learning framework will be used to demonstrate effective learning of the dynamics of (1) a pendulum experiencing friction and (2) cell migration, both through existing experimental data. This research will allow for improved learning of physics in which non-conservative forces exist.