||
为推动物理学计算程序代码的共享与合作,CPB最新开辟“Computational Programs for Physics”栏目。本文是此栏目的第一篇发文,欢迎关注!
MatChat: A large language model and application service platform for materials science
Zi-Yi Chen(陈子逸), Fan-Kai Xie(谢帆恺), Meng Wan(万萌), Yang Yuan(袁扬),Miao Liu(刘淼), Zong-Guo Wang(王宗国), Sheng Meng(孟胜), and Yan-Gang Wang(王彦棡)
Chin. Phys. B, 2023, 32 (11): 118104
Abstract
2:17
研究背景
材料合成路径预测是物质科学领域的重要课题。1990年诺贝尔化学奖授予了美国有机化学家Elias James Corey教授,表彰他开发了计算机辅助有机合成的理论和方法,将人类300多条经验写进计算机编码,成为制药领域的重要工具软件。无机材料受限于合成路径的复杂度和缺乏数据集等因素,尚未有清晰的化学合成路径预测机制。
人工智能(AI)大模型的崛起是近几年人工智能领域的一项重大突破。人工智能大模型是指具有大规模参数数量的人工智能模型。这些模型通常使用深度学习算法,并且由数百万到数千亿个参数组成。这些参数用于表示模型在学习任务中的知识和能力。人工智能大模型的目标是通过学习大规模数据集中的模式和规律来实现强大的图像识别、自然语言处理、语音识别等人工智能任务。大模型的参数数量可以帮助它们更好地捕捉输入数据的复杂特征,并进行更精确的推断和预测。一些著名的人工智能大模型,如BERT(Bidirectional Encoder Representations from Transformers)和GPT(Generative Pre-trained Transformer),在自然语言处理领域取得了显著的成果。
图1 MatChat人工智能模型基本框架及功能细节
研究内容
本工作将AI大模型应用于材料科学领域,将数万个化学合成路径数据投喂给大语言模型LLAMA2-7b,从而获得了MatChat模型,可用来预测无机材料的合成路径。该模型可根据所询问的结构进行逻辑推理,并输出相应的制备工艺和配方。目前已部署上线(http://chat.aicnic.cn/onchat),并向所有材料科研人员开放使用,为材料研究和创新带来了新启发和新思路。
图2 在线对话场景 (http://chat.aicnic.cn/onchat)
研究意义和重要性
该工作为大语言模型在细分科学领域的应用提出了一种有可能的解决思路,并初步展示出了方法的可行性。通过自然语言模型提取文献数据,进而将文献数据用于语言模型训练是发展细分科学领域的一条可行路径。这项工作展示了无机材料合成路径预测领域的“莱特兄弟一分钟飞翔“,受限于数据集的数量和质量,模型预测精度尚受到一定限制。未来合作团队计划将更多的文献数据、Atomly.net 数据库等诸多高质量数据集用于模型训练,让无机材料合成路径预测的大飞机飞得更高更远。
表1 Matchat表现与其他大语言模型比较实例
本文作为"绿色通道”稿件在CPB最新开辟的"Computational Program for Physics" 栏目发表,开源代码发布在Science Data Bank。由物理所孟胜研究员、中国科学院计算机网络信息中心王彦棡研究员领衔。刘淼研究员、王宗国副研究员为本文的通讯作者,谢帆恺为本文的第一作者。该工作得到中国科学院网信专项(CAS-WX2023SF-0101)应用示范项目的支持。
SPECIAL TOPIC — Valleytronic
SPECIAL TOPIC — Plasma disruption
SPECIAL TOPIC — Smart design of materials and design of smart materials
SPECIAL TOPIC — Celebrating the 100th Anniversary of Physics Discipline of Xiamen University
TOPICAL REVIEW — Physics in micro-LED and quantum dots devices
TOPICAL REVIEW — Celebrating 30 Years of Chinese Physics B
TOPICAL REVIEW — The third carbon: Carbyne with one-dimensional sp-carbon
SPECIAL TOPIC — Fabrication and manipulation of the second-generation quantum systems
SPECIAL TOPIC — Celebrating the 70th Anniversary of the Physics of Jilin University
TOPICAL REVIEW—Laser and plasma assisted synthesis of advanced nanomaterials in liquids
TOPICAL REVIEW — Progress in thermoelectric materials and devices
SPECIAL TOPIC — Emerging photovoltaic materials and devices
SPECIAL TOPIC — Organic and hybrid thermoelectrics
SPECIAL TOPIC — Superconductivity in vanadium-based kagome materials
SPECIAL TOPIC— Interdisciplinary physics: Complex network dynamics and emerging technologies
SPECIAL TOPIC — Non-Hermitian physics
SPECIAL TOPIC — Unconventional superconductivity
SPECIAL TOPIC — Two-dimensional magnetic materials and devices
SPECIAL TOPIC — Ion beam modification of materials and applications
SPECIAL TOPIC — Quantum computation and quantum simulation
SPECIAL TOPIC —Twistronics
SPECIAL TOPIC — Machine learning in condensed matter physics
SPECIAL TOPIC — Phononics and phonon engineering
SPECIAL TOPIC — Water at molecular level
SPECIAL TOPIC — Optical field manipulation
SPECIAL TOPIC — Modeling and simulations for the structures and functions of proteins and nucleic acids
SPECIAL TOPIC —Terahertz physics
SPECIAL TOPIC — Ultracold atom and its application in precision measurement
SPECIAL TOPIC — Topological 2D materials
SPECIAL TOPIC — Active matters physics
SPECIAL TOPIC — Physics in neuromorphic devices
SPECIAL TOPIC — Advanced calculation & characterization of energy storage materials & devices at multiple scale
TOPICAL REVIEW — Advanced calculation & characterization of energy storage materials & devices at multiple scale
TOPICAL REVIEW — Quantum dot displays
TOPICAL REVIEW — CALYPSO structure prediction methodology and its applications to materials discovery
SPECIAL TOPIC — A celebration of the 100th birthday of Kun Huang
TOPICAL REVIEW — A celebration of the 100th birthday of Kun Huang
SPECIAL TOPIC — Strong-field atomic and molecular physics
TOPICAL REVIEW — Strong-field atomic and molecular physics
TOPICAL REVIEW — Topological semimetals
SPECIAL TOPIC — Topological semimetals
SPECIAL TOPIC — Photodetector: Materials, physics, and applications
TOPICAL REVIEW — Photodetector: Materials, physics, and applications
TOPICAL REVIEW — Fundamental research under high magnetic fields
Virtual Special Topic — High temperature superconductivity
Virtual Special Topic — Magnetism and Magnetic Materials
长按二维码,关注我们
Archiver|手机版|科学网 ( 京ICP备07017567号-12 )
GMT+8, 2024-10-20 03:56
Powered by ScienceNet.cn
Copyright © 2007- 中国科学报社