The Belt and Road global navigation satellite system(B&R GNSS)network is the first large-scale deployment of Chinese GNSS equipment in a seismic system.Prior to this,there have been few systematic assessments of t...The Belt and Road global navigation satellite system(B&R GNSS)network is the first large-scale deployment of Chinese GNSS equipment in a seismic system.Prior to this,there have been few systematic assessments of the data quality of Chinese GNSS equipment.In this study,data from four representative GNSS sites in different regions of China were analyzed using the G-Nut/Anubis software package.Four main indicators(data integrity rate,data validity ratio,multi-path error,and cycle slip ratio)used to systematically analyze data quality,while evaluating the seismic monitoring capabilities of the network based on earthquake magnitudes estimated from high-frequency GNSS data are evaluated by estimating magnitude based on highfrequency GNSS data.The results indicate that the quality of the data produced by the three types of Chinese receivers used in the network meets the needs of earthquake monitoring and the new seismic industry standards,which provide a reference for the selection of equipment for future new projects.After the B&R GNSS network was established,the seismic monitoring capability for earthquakes with magnitudes greater than M_(W)6.5 in most parts of the Sichuan-Yunnan region improved by approximately 20%.In key areas such as the Sichuan-Yunnan Rhomboid Block,the monitoring capability increased by more than 25%,which has greatly improved the effectiveness of regional comprehensive earthquake management.展开更多
【目的】围绕AI场景下科学数据的共享与利用问题,针对现有FAIR原则不足以指导科学数据满足AI就绪的现状,构建面向AI就绪的科学数据共享与利用原则框架。【方法】通过系统梳理传统机器学习、大模型预训练、大模型微调、检索增强生成及智...【目的】围绕AI场景下科学数据的共享与利用问题,针对现有FAIR原则不足以指导科学数据满足AI就绪的现状,构建面向AI就绪的科学数据共享与利用原则框架。【方法】通过系统梳理传统机器学习、大模型预训练、大模型微调、检索增强生成及智能体等5类典型AI任务的数据需求,在传统FAIR“四可”维度的基础上,提出面向AI就绪(即For AI Ready)的科学数据共享与利用原则框架FAIR×FAIR,进而提出与框架相适应的层次化技术栈。【结果】FAIR×FAIR框架明确了13项科学数据满足AI就绪的技术要求,为弥合AI任务与科学数据之间的语义鸿沟提供了系统化方案。【局限】本研究提出的原则框架其实施效果仍需通过后续领域应用案例进一步验证。【结论】FAIR×FAIR框架为AI时代的科学数据共享与高效利用提供了理论依据和实践路径,对推动数据驱动型科研范式的演进具有重要意义。展开更多
The seismological observation system in China has experienced rapid development over the Tenth Five-year Plan period. China Earthquake Administration (CEA) has completed the establishment of China digital seismologica...The seismological observation system in China has experienced rapid development over the Tenth Five-year Plan period. China Earthquake Administration (CEA) has completed the establishment of China digital seismological observation network. CEA has accomplished analog-to-digital conversion of the existing seismological observation systems and set up a number of new digital seismic stations. This indicates full digitization of seismological ob-servation in China. This paper presents an overview of the scale,layout principle and major functions of the up-dated national digital seismograph network,regional digital seismograph network,and digital seismograph net-work for volcano monitoring and mobile digital seismograph networks in China.展开更多
逻辑回归是一种广泛应用于现实分类任务的机器学习模型。随着数据孤岛问题的涌现,如何针对多参与主体非贯通数据联合构建逻辑回归模型成为一个关键问题。纵向联邦学习可实现数据明文不暴露前提下多主体跨样本特征的联合机器学习模型训...逻辑回归是一种广泛应用于现实分类任务的机器学习模型。随着数据孤岛问题的涌现,如何针对多参与主体非贯通数据联合构建逻辑回归模型成为一个关键问题。纵向联邦学习可实现数据明文不暴露前提下多主体跨样本特征的联合机器学习模型训练。然而,现有纵向联邦逻辑回归方法主要基于同态加密技术,具有计算和通信开销大的短板。针对逻辑回归模型,研究安全高效的纵向联邦学习算法,目标实现数据隐私保护和模型学习效率的较优权衡。具体地,基于秘密共享提出了一种面向逻辑回归模型的高效率纵向联邦学习算法(Vertical Federated Logistic Regression algorithm based on Secret Sharing, VFLR-SS),通过将跨域分析过程中的中间数据随机分解为多个秘密份额进行交互从而实现隐私保护,同时避免了同态加密引发的计算和通信开销。对VFLR-SS的安全性进行了分析,并基于真实数据对算法进行了验证。实验结果表明VFLR-SS可实现与集中式逻辑回归算法可比的效用和性能,大幅降低了传统同态加密方法中的计算及通信开销。展开更多
基金supported by grants from the National Natural Science Foundation of China(No.42004010)the B&R Seismic Monitoring Network Project of the China Earthquake Networks Center(No.5007).
文摘The Belt and Road global navigation satellite system(B&R GNSS)network is the first large-scale deployment of Chinese GNSS equipment in a seismic system.Prior to this,there have been few systematic assessments of the data quality of Chinese GNSS equipment.In this study,data from four representative GNSS sites in different regions of China were analyzed using the G-Nut/Anubis software package.Four main indicators(data integrity rate,data validity ratio,multi-path error,and cycle slip ratio)used to systematically analyze data quality,while evaluating the seismic monitoring capabilities of the network based on earthquake magnitudes estimated from high-frequency GNSS data are evaluated by estimating magnitude based on highfrequency GNSS data.The results indicate that the quality of the data produced by the three types of Chinese receivers used in the network meets the needs of earthquake monitoring and the new seismic industry standards,which provide a reference for the selection of equipment for future new projects.After the B&R GNSS network was established,the seismic monitoring capability for earthquakes with magnitudes greater than M_(W)6.5 in most parts of the Sichuan-Yunnan region improved by approximately 20%.In key areas such as the Sichuan-Yunnan Rhomboid Block,the monitoring capability increased by more than 25%,which has greatly improved the effectiveness of regional comprehensive earthquake management.
文摘【目的】围绕AI场景下科学数据的共享与利用问题,针对现有FAIR原则不足以指导科学数据满足AI就绪的现状,构建面向AI就绪的科学数据共享与利用原则框架。【方法】通过系统梳理传统机器学习、大模型预训练、大模型微调、检索增强生成及智能体等5类典型AI任务的数据需求,在传统FAIR“四可”维度的基础上,提出面向AI就绪(即For AI Ready)的科学数据共享与利用原则框架FAIR×FAIR,进而提出与框架相适应的层次化技术栈。【结果】FAIR×FAIR框架明确了13项科学数据满足AI就绪的技术要求,为弥合AI任务与科学数据之间的语义鸿沟提供了系统化方案。【局限】本研究提出的原则框架其实施效果仍需通过后续领域应用案例进一步验证。【结论】FAIR×FAIR框架为AI时代的科学数据共享与高效利用提供了理论依据和实践路径,对推动数据驱动型科研范式的演进具有重要意义。
文摘The seismological observation system in China has experienced rapid development over the Tenth Five-year Plan period. China Earthquake Administration (CEA) has completed the establishment of China digital seismological observation network. CEA has accomplished analog-to-digital conversion of the existing seismological observation systems and set up a number of new digital seismic stations. This indicates full digitization of seismological ob-servation in China. This paper presents an overview of the scale,layout principle and major functions of the up-dated national digital seismograph network,regional digital seismograph network,and digital seismograph net-work for volcano monitoring and mobile digital seismograph networks in China.
文摘逻辑回归是一种广泛应用于现实分类任务的机器学习模型。随着数据孤岛问题的涌现,如何针对多参与主体非贯通数据联合构建逻辑回归模型成为一个关键问题。纵向联邦学习可实现数据明文不暴露前提下多主体跨样本特征的联合机器学习模型训练。然而,现有纵向联邦逻辑回归方法主要基于同态加密技术,具有计算和通信开销大的短板。针对逻辑回归模型,研究安全高效的纵向联邦学习算法,目标实现数据隐私保护和模型学习效率的较优权衡。具体地,基于秘密共享提出了一种面向逻辑回归模型的高效率纵向联邦学习算法(Vertical Federated Logistic Regression algorithm based on Secret Sharing, VFLR-SS),通过将跨域分析过程中的中间数据随机分解为多个秘密份额进行交互从而实现隐私保护,同时避免了同态加密引发的计算和通信开销。对VFLR-SS的安全性进行了分析,并基于真实数据对算法进行了验证。实验结果表明VFLR-SS可实现与集中式逻辑回归算法可比的效用和性能,大幅降低了传统同态加密方法中的计算及通信开销。
