Variational Approach for Markov Processes (VAMP)
A mathematical framework used to learn optimal low-dimensional representations and models of complex dynamical systems from time-series data.
- Built on the principle that the best linear model for a nonlinear system can be derived from the top singular components of the Koopman Operator
- Introduces a family of score functions (e.g.,VAMP-E, VAMP-1, VAMP-2) that measure how well a set of features (or “reaction coordinates”) represents the slow, essential dynamics of the system.
- Valid for both reversible and non-reversible processes, as well as stationary and non-stationary dynamics.
Works by identifying a low-dimensional Feature Space where the complex, nonlinear dynamics of a system can be accurately represented as a simple linear model.