ML Session: Improving the Generalization Behavior of ML based Surrogate Models (Mathias Niepert)

We are happy to announce the next presentation in the ML-Session series:

Mathias Niepert (University of Stuttgart) will present on Wednesday 17 April 2023 at 2pm in PWR 57, 8.122 a lecture IN PERSON (no Webex) on "Improving the Generalization Behavior of ML based Surrogate Models"

Abstract: The talk will provide a short introduction and overview of equivariant neural networks, that is, neural networks that are designed to be equivariant (or invariant) to operations of particular application-dependent groups. Examples are SE(2)-equivariant CNNs that are equivariant to translations and rotations, graph neural networks invariant to permutations of the nodes, and equivariant message-passing neural networks equivariant to the group SO(3). The core idea of equivariant neural networks is improved generalization behavior and data efficiency. We will also touch on recent work on equivariant neural networks for modeling interatomic potentials.