Geometric Machine Learning with Symmetries and Graphs
Details
Title: Geometric Machine Learning with Symmetries and Graphs: Applications in Molecular Problems
Date: Jan 31 2026 Noon - 14:00 EST
Abstract: For geometric problems, symmetry equivariant machine learning has emerged as the modus operandi. In physics, symmetry is treated as the north star, the guiding principle for problem characterization and method optimization. Geometric features are inherently compatible with continuous Lie groups (and therefore symmetries), leading to continuous parametrization and successful training methodologies. The general mapping of a classical embedding and a quantum mechanical embedding will be discussed in terms of group theoretical concepts. In the context of embeddings, graph embedding is also a powerful method for machine learning problems taking better care of inter-relations among nodal variables. Classical and quantum neural networks with different subsets of the concepts mentioned will be defined and learning attempted with. The dataset under study is geometric molecular data (from quantum mechanical models) with varying geometric complexity. Performance comparison will include accuracy metrics, generalizability metrics, and.general remarks about expressivity.
Bio: Dr Saumya Biswas completed a Master's degree in electrical engineering from University of California Riverside in 2016, a PhD in Physics from University of Oregon in 2021. He has been serving as a PostDoc at University of Maryland from 2022. He is also active in the technology subspace, working with the companies Apsidal LLC and Sapienly for AI and quantum information innovations.
Moderators: Dr. Sebastian Zajac, member of QPoland, Dr. Pawel Gora, CEO of Quantum AI Foundation