Switched-system identification
Learning switching surfaces and vector fields from trajectory data through bilevel convex optimization, for the hybrid-system pipeline developed for L4DC 2026.
Software
Topological tools provide a way to describe the global behavior of a dynamical system without tracking every trajectory in detail. The goal is to identify attractors and recurrent structure, understand how they are connected, and determine which conclusions persist under perturbation or model uncertainty.
Timeseries data from the Lorenz system
Persistence diagram of the data (H_1)
Projects
LatentDynamics
Topological analysis of learned high-dimensional dynamics on low-dimensional latent spaces.