| 刘小平,郑企雨,朱相丽.化学通报,2024,87(11):1309-1318. |
| 图神经网络在材料化学中的应用研究 |
| Graph Neural Networks Applied in Materials Chemistry |
| 投稿时间：2024-06-11 修订日期：2024-08-29 |
| DOI： |
| 中文关键词: 机器学习 图神经网络 材料化学 |
| 英文关键词:machine learning, graph neural networks, materials chemistry |
| 基金项目:国家自然科学基金项目(22342011)资助 |
|
| 摘要点击次数: 269 |
| 全文下载次数: 81 |
| 中文摘要: |
| 图神经网络(Graph Neural Network,GNN)作为一类快速发展的机器学习模型,能够直接处理分子或者材料的图形或结构表示,可以完全表征材料,建立不同层面的图神经网络任务,并且通过神经网络的信息传递过程,最大限度地获得分子和材料的“构效关系”。因而GNN在材料化学领域广受关注。人们已经开发了多种图神经网络模型或框架用于分子性质的预测、分子动力学模拟、用于激发态动力学及反应预测、用于材料性能预测和材料的分子设计等。本文概述了当前广泛使用的GNN模型、展示了最新的GNN架构、广泛使用的数据集以及GNN在材料化学中的最新应用成果。最后提出了GNN进一步在材料化学中的发展和应用前景。 |
| 英文摘要: |
| Graph Neural Networks (GNN), a rapidly growing class of machine learning models, are able to directly deal with graphical or structural representations of molecules or materials, to fully characterize materials, to build graph neural network tasks at different levels, and to maximize access to molecules and materials through the neural network information transfer process of the "conformational relationships". Thus, GNNs have gained wide attention in the field of materials chemistry. A variety of graph neural network models or frameworks have been developed for the prediction of molecular properties, molecular dynamics simulations, for excited state dynamics and reaction prediction, for materials properties prediction and molecular design of materials. This paper provides an overview of the current widely used GNN models, shows the latest GNN architectures, widely used datasets, and recent results of GNN applications in materials chemistry. Finally, the prospects for further development and application of GNN in materials chemistry are presented. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
|
|
|