Abstract
Swarming behavior is prevalent throughout the natural world. The field of ac-
tive matter investigates swarm dynamics using methods from condensed matter
physics, including symmetry breaking, phase transitions, and statistical dynamics.
The Vicsek model is a numerical model used to simulate swarm behavior. This
thesis uses a machine learning model originally developed for traffic prediction to
forecast the future states of the Vicsek model. The machine learning model is
a Spatio-Temporal Graph Convolutional Network (STGCN) which is designed to
make short term forecasts on graph-structures time series.
Using a coarse-grained approach, this thesis demonstrates how the overall dy-
namics of a system can be assessed at a macroscopic level by forecasting the future
density and orientation of a Vicsek model at a macroscopic level. The STGCN
model is capable of making forecasts on systems with varying noise and density
levels. This approach has potential applications in real-life swarms and other
swarm models. One possible extension of this work could involve mapping phase
transitions in active matter systems.
i
