English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Graph Convolutional Neural Network (GCNN) is a popular class of deep learning(DL) models in material science to predict material properties from the graphrepresentation of molecular structures. Training an accurate and comprehensiveGCNN surrogate for molecular design requires large-scale graph datasets and isusually a time-consuming process. Recent advances in GPUs and distributedcomputing open a path to reduce the computational cost for GCNN trainingeffectively. However, efficient utilization of high performance computing (HPC)resources for training requires simultaneously optimizing large-scale datamanagement and scalable stochastic batched optimization techniques. In thiswork, we focus on building GCNN models on HPC systems to predict materialproperties of millions of molecules. We use HydraGNN, our in-house library forlarge-scale GCNN training, leveraging distributed data parallelism in PyTorch.We use ADIOS, a high-performance data management framework for efficientstorage and reading of large molecular graph data. We perform parallel trainingon two open-source large-scale graph datasets to build a GCNN predictor for animportant quantum property known as the HOMO-LUMO gap. We measure thescalability, accuracy, and convergence of our approach on two DOEsupercomputers: the Summit supercomputer at the Oak Ridge Leadership ComputingFacility (OLCF) and the Perlmutter system at the National Energy ResearchScientific Computing Center (NERSC). We present our experimental results withHydraGNN showing i) reduction of data loading time up to 4.2 times comparedwith a conventional method and ii) linear scaling performance for training upto 1,024 GPUs on both Summit and Perlmutter.

Scalable training of graph convolutional neural networks for fast and accurate predictions of HOMO-LUMO gap in molecules