Energy landscape analysis of multivariate time series via the Ising model | Winter Seminar Series

I present energy landscape analysis for multivariate time series. We infer an effective energy landscape from the data by fitting the inverse Ising model (also called a Boltzmann machine and pairwise maximum entropy model) and represent each observed system state as the position of a "ball" constrained to move on that surface. From the estimated landscape we compute, statistical-physics‑inspired indices, such as basin structure, barrier heights, dwell times, transition rates, and susceptibilities, to characterize collective organization, metastability, and transition dynamics in the original time series. I illustrate the approach with neuroimaging examples in health and disease.