Gilmer et al., "Neural Message Passing for Quantum Chemistry" (2017)

• https://arxiv.org/abs/1704.01212
• Justin Gilmer, Samuel S. Schoenholz, Patrick F. Riley, Oriol Vinyals, George E. Dahl

Graph neural network, Message passing, Quantum chemistry, Molecular property prediction

The paper reformulates several neural models for graph-structured data within a common Message Passing Neural Network framework: learned messages update node states for a fixed number of rounds, followed by a permutation-invariant readout. It develops variants for molecular property prediction on QM9, providing a common language for comparing models such as GCNs and later studying the expressive limits of aggregation in GINs.