AI Paper Atlas

Research map / Graph learning / GNN architectures

GemNet: Universal Directional Graph Neural Networks for Molecules

Johannes Gasteiger, Florian Becker, Stephan Günnemann

NeurIPS 2021 · Graph learning › GNN architectures · approach: Equivariant networks for molecules · 123 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
GemNet-Q vs DimeNet++DimeNet++ · COLLforce MAE: 40 -> 26.4Improved
add symmetric message passing (ablation)GemNet-Q · COLLforce MAE: 28.5 -> 27.0Improved
use bilinear layer instead of Hadamard product (ablation)GemNet-Q · COLLforce MAE: 29.3 -> 27.0Improved
add atom embedding updates (ablation)GemNet-Q · COLLforce MAE: 28.3 -> 27.0Improved
add one-hop message passing (ablation)GemNet-Q · COLLforce MAE: 31.3 -> 27.0Improved
add two-hop message passing (ablation)GemNet-Q · COLLforce MAE: 32.4 -> 27.0Improved
add two-hop message passing (GemNet-Q vs GemNet-T) (ablation)GemNet-T · MD17force MAE: Aspirin 9.5 -> 9.4No clear effect
add precomputed variance scaling factors (ablation)GemNet-Q · COLLforce MAE: 29.1 -> 27.0Improved
use layer normalization instead of scaling factors (ablation)GemNet-Q · COLLforce MAE: 27.0 -> 33.3Hurt
add biases (ablation)GemNet-Q · COLLforce MAE: 27.0 -> 27.2No clear effect
predict forces directly instead of via energy gradient (ablation)GemNet-Q · COLLforce MAE: 26.4 -> 38.1Hurt
predict forces directly instead of via energy gradient (ablation)GemNet · OC20 S2EFforce MAEImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy