Subscribe to Events

Conley's topological approach to dynamics provides a framework for deducing qualitative and global information about dynamics that is robust with respect to perturbations of the underlying system. Current methods for applications begin with a uniform discretization of the phase space into cubes, followed by problem-specific subdivisions if needed. At fine enough levels of resolution, the computational cost of these methods can become prohibitively expensive as dimension increases. We will introduce a machine learning approach for discretizing the phase space that aims to overcome the dimension problem. Preliminary experimental results using this method will be discussed.