This website requires JavaScript.

Neural Structure Fields with Application to Crystal Structure Autoencoders

Naoya ChibaYuta SuzukiTatsunori Taniai ...+3 Kanta Ono
Dec 2022
摘要
Representing crystal structures of materials to facilitate determining themvia neural networks is crucial for enabling machine-learning applicationsinvolving crystal structure estimation. Among these applications, the inversedesign of materials can contribute to next-generation methods that explorematerials with desired properties without relying on luck or serendipity. Wepropose neural structure fields (NeSF) as an accurate and practical approachfor representing crystal structures using neural networks. Inspired by theconcepts of vector fields in physics and implicit neural representations incomputer vision, the proposed NeSF considers a crystal structure as acontinuous field rather than as a discrete set of atoms. Unlike existinggrid-based discretized spatial representations, the NeSF overcomes the tradeoffbetween spatial resolution and computational complexity and can represent anycrystal structure. To evaluate the NeSF, we propose an autoencoder of crystalstructures that can recover various crystal structures, such as those ofperovskite structure materials and cuprate superconductors. Extensivequantitative results demonstrate the superior performance of the NeSF comparedwith the existing grid-based approach.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答