Machine Learning: the ultimate tool for predicting the fascinating properties of Active Matter.

Active systems are groups of interacting, self-driven "machines" that self-organize on multiple scales with no predefined and in the presence of noise. Recently with the development of new algorithms and new open areas to predict equations from datasets, we've been able to think about the possibility of determining the governing equations for such complex phenomena. In this work, we use Machine learning to predict the dynamic behavior of these active systems. Our aim is to develop artificial intelligence algorithms to predict the fundamental equations that describe the behavior of active matter. Our results show that we can predict equations of motion for several types of time-series data from simulations, from numerical data, and from microscopy video recorder images. Interestingly, our predicted equations have a low level of complexity and a reduced number of parameters. This provides a combined computational and machine learning framework that sheds light on the physical underpinnings of active systems and provides suggestions on how to engineer the assembly of smart active materials.