Our new X account is live! Follow @wizwand_team for updates
WorkDL logo mark

Equivariant Matrix Function Neural Networks

About

Graph Neural Networks (GNNs), especially message-passing neural networks (MPNNs), have emerged as powerful architectures for learning on graphs in diverse applications. However, MPNNs face challenges when modeling non-local interactions in graphs such as large conjugated molecules, and social networks due to oversmoothing and oversquashing. Although Spectral GNNs and traditional neural networks such as recurrent neural networks and transformers mitigate these challenges, they often lack generalizability, or fail to capture detailed structural relationships or symmetries in the data. To address these concerns, we introduce Matrix Function Neural Networks (MFNs), a novel architecture that parameterizes non-local interactions through analytic matrix equivariant functions. Employing resolvent expansions offers a straightforward implementation and the potential for linear scaling with system size. The MFN architecture achieves stateof-the-art performance in standard graph benchmarks, such as the ZINC and TU datasets, and is able to capture intricate non-local interactions in quantum systems, paving the way to new state-of-the-art force fields.

Ilyes Batatia, Lars L. Schaaf, Huajie Chen, G\'abor Cs\'anyi, Christoph Ortner, Felix A. Faber• 2023

Related benchmarks

TaskDatasetResultRank
Graph ClassificationMUTAG (10-fold cross-validation)
Accuracy91.5
206
Graph ClassificationPROTEINS (10-fold cross-validation)
Accuracy76.18
197
Graph ClassificationIMDB-B (10-fold cross-validation)
Accuracy74.1
148
Graph ClassificationPTC (10-fold cross-validation)
Accuracy68.9
115
Graph ClassificationENZYMES (10-fold cross-validation)
Accuracy72.9
64
Energy and force predictionGuaranteed non-local cumulene nc 3-10, 13, 14 (train)
Energy RMSE (meV/atom)2
3
Energy and force predictionGuaranteed non-local cumulene In-Domain, nc 3-10, 13, 14 (test)
Energy RMSE (meV/atom)2.6
3
Energy and force predictionGuaranteed non-local cumulene Out-of-Domain nc 11 12 (test)
Energy RMSE (meV/atom)0.9
3
Energy and force predictionGuaranteed non-local cumulene Out-of-Domain nc 15 16 (test)
Energy RMSE (meV/atom)2.6
3
Showing 9 of 9 rows

Other info

Follow for update