
Geometric deep learning models leveraging symmetry group equivariance can effectively represent and process non-Euclidean data like graphs and manifolds, achieving improved statistical efficiency, interpretability, and generalization.


coremsg

equivariant-convolution-frameworks-for-representation-learning-on-non-euclidean-domains


Equivariant Convolution Frameworks for Representation Learning on Non-Euclidean Domains


title_rewrite


MAgNET is a novel graph U-Net framework that extends convolutional neural networks to handle arbitrary graph-structured data, enabling efficient surrogate modeling for computationally expensive mesh-based simulations.


magnet-a-novel-graph-u-net-architecture-for-efficient-mesh-based-simulations


MAgNET: A Novel Graph U-Net Architecture for Efficient Mesh-Based Simulations



Leveraging differential k-forms in Rn for efficient and interpretable geometric representation learning without message passing.


simplicial-representation-learning-with-neural-k-forms-at-iclr-2024


Simplicial Representation Learning with Neural k-Forms at ICLR 2024



Leveraging differential k-forms in Rn for geometric deep learning without message passing.


simplicial-representation-learning-with-neural-k-forms-a-geometric-deep-learning-approach


Simplicial Representation Learning with Neural k-Forms: A Geometric Deep Learning Approach
