- This event has passed.
Spring 2022 GRASP SFI: Marc Finzi, New York University, “Embedding Symmetries and Conservation Laws in Deep Learning Models for Dynamical Systems”
February 23 at 3:00 PM - 4:00 PM
*This will be a HYBRID Event with in-person attendance in Levine 512 and Virtual attendance via Zoom
In contrast to traditional control systems where detailed dynamics models are constructed from a mix of physical understanding and empirical data, machine learning for intuitive physics, reinforcement learning, and robotics often takes a hands off approach treating the dynamics as a black box with little to no assumed structure. We show how desirable high level properties like symmetries, energy and momentum conservation, and other constraints can be reintroduced into these models to improve generalization. These high level attributes represent prior knowledge about the underlying physics of the system in the Bayesian sense, and can even be incorporated in a way that does not limit the flexibility of the model.
New York University
Marc is a PhD candidate in computer science at New York University advised by Andrew Wilson. Marc has a masters in operations research at Cornell and previously obtained a bachelors in physics at Harvey Mudd College. His research interests have been on incorporating high level properties of data like symmetries, locality, and differential equations into the structure of deep learning models to improve generalization as well as pushing towards more general machine learning models that can be applied on many data types simultaneously. In his work, Marc often draws on ideas from physics and differential geometry, which he finds as a helpful guide for intuition.