System Identification with Regularized Neural Networks

Elisa Negrini
University of California, Los Angeles (UCLA)
Mathematical Sciences

In this talk, I will present a neural network-based approach for learning governing equations from data, also known as system identification. I will begin with an overview of feedforward networks and their key properties, emphasizing the role of regularization when dealing with noisy data. Next, I will introduce Lipschitz regularization and demonstrate its application in neural networks for system identification. Finally, I will showcase results where a Lipschitz-regularized neural network reconstructs the right-hand side of an ODE system x'(t) = f(t,x) directly from observed data, and discuss potential future research directions in this area.


Back to PUMA: Practicum for Undergraduate MAthematicians in Inverse Problems and Data Assimilation