In the financial sector, data are highly confidential and sensitive,and ensuring data privacy is critical. Sample fusion is the basis of horizontalfederation learning, but it is suitable only for scenarios where custo...In the financial sector, data are highly confidential and sensitive,and ensuring data privacy is critical. Sample fusion is the basis of horizontalfederation learning, but it is suitable only for scenarios where customershave the same format but different targets, namely for scenarios with strongfeature overlapping and weak user overlapping. To solve this limitation, thispaper proposes a federated learning-based model with local data sharing anddifferential privacy. The indexing mechanism of differential privacy is used toobtain different degrees of privacy budgets, which are applied to the gradientaccording to the contribution degree to ensure privacy without affectingaccuracy. In addition, data sharing is performed to improve the utility ofthe global model. Further, the distributed prediction model is used to predictcustomers’ loan propensity on the premise of protecting user privacy. Usingan aggregation mechanism based on federated learning can help to train themodel on distributed data without exposing local data. The proposed methodis verified by experiments, and experimental results show that for non-iiddata, the proposed method can effectively improve data accuracy and reducethe impact of sample tilt. The proposed method can be extended to edgecomputing, blockchain, and the Industrial Internet of Things (IIoT) fields.The theoretical analysis and experimental results show that the proposedmethod can ensure the privacy and accuracy of the federated learning processand can also improve the model utility for non-iid data by 7% compared tothe federated averaging method (FedAvg).展开更多
In this paper,we propose a symmetric difference data enhancement physics-informed neural network(SDE-PINN)to study soliton solutions for discrete nonlinear lattice equations(NLEs).By considering known and unknown symm...In this paper,we propose a symmetric difference data enhancement physics-informed neural network(SDE-PINN)to study soliton solutions for discrete nonlinear lattice equations(NLEs).By considering known and unknown symmetric points,numerical simulations are conducted to one-soliton and two-soliton solutions of a discrete KdV equation,as well as a one-soliton solution of a discrete Toda lattice equation.Compared with the existing discrete deep learning approach,the numerical results reveal that within the specified spatiotemporal domain,the prediction accuracy by SDE-PINN is excellent regardless of the interior or extrapolation prediction,with a significant reduction in training time.The proposed data enhancement technique and symmetric structure development provides a new perspective for the deep learning approach to solve discrete NLEs.The newly proposed SDE-PINN can also be applied to solve continuous nonlinear equations and other discrete NLEs numerically.展开更多
A new concept is suggested on tectonomagnetic research about the noise in simultaneous geomagnetic difference data caused by the effect of Sq local-time variation, together with the method of theoretical calculation. ...A new concept is suggested on tectonomagnetic research about the noise in simultaneous geomagnetic difference data caused by the effect of Sq local-time variation, together with the method of theoretical calculation. The level of the noise and its contribution to the total noises of the differences data are analyzed. The result indicates that the noise increases linearly with the increase of the distance between the two stations in the range of 40° longitude-difference, and its increasing rate is about 0.4 nT/（°）at latitude 40°N. The example calculated at a pair of sites with longitude-difference 0.357°, shows that the noise is about one fifth of the total noises of the difference data on geomagnetic quiet-day.展开更多
A novel weather radar system with distributed phased-array front-ends was developed. The specifications and preliminary data synthesis of this system are presented, which comprises one back-end and three or more front...A novel weather radar system with distributed phased-array front-ends was developed. The specifications and preliminary data synthesis of this system are presented, which comprises one back-end and three or more front-ends. Each front-end, which utilizes a phased-array digital beamforming technology, sequentially transmits four 22.5°-width beams to cover the 0°–90° elevational scan within about 0.05 s. The azimuthal detection is completed by one mechanical scan of0°–360° azimuths within about 12 s volume-scan update time. In the case of three front-ends, they are deployed according to an acute triangle to form a fine detection area(FDA). Because of the triangular deployment of multiple phased-array front-ends and a unique synchronized azimuthal scanning(SAS) rule, this new radar system is named Array Weather Radar(AWR). The back-end controls the front-ends to scan strictly in accordance with the SAS rule that assures the data time differences(DTD) among the three front-ends are less than 2 s for the same detection point in the FDA. The SAS can maintain DTD < 2 s for an expanded seven-front-end AWR. With the smallest DTD, gridded wind fields are derived from AWR data, by sampling of the interpolated grid, onto a rectangular grid of 100 m ×100 m ×100 m at a 12 s temporal resolution in the FDA. The first X-band single-polarized three-front-end AWR was deployed in field experiments in 2018 at Huanghua International Airport, China. Having completed the data synthesis and processing, the preliminary observation results of the first AWR are described herein.展开更多
基金supported by the National Natural Science Foundation (NSFC),China,under the National Natural Science Foundation Youth Fund program (J.Hao,No.62101275).
文摘In the financial sector, data are highly confidential and sensitive,and ensuring data privacy is critical. Sample fusion is the basis of horizontalfederation learning, but it is suitable only for scenarios where customershave the same format but different targets, namely for scenarios with strongfeature overlapping and weak user overlapping. To solve this limitation, thispaper proposes a federated learning-based model with local data sharing anddifferential privacy. The indexing mechanism of differential privacy is used toobtain different degrees of privacy budgets, which are applied to the gradientaccording to the contribution degree to ensure privacy without affectingaccuracy. In addition, data sharing is performed to improve the utility ofthe global model. Further, the distributed prediction model is used to predictcustomers’ loan propensity on the premise of protecting user privacy. Usingan aggregation mechanism based on federated learning can help to train themodel on distributed data without exposing local data. The proposed methodis verified by experiments, and experimental results show that for non-iiddata, the proposed method can effectively improve data accuracy and reducethe impact of sample tilt. The proposed method can be extended to edgecomputing, blockchain, and the Industrial Internet of Things (IIoT) fields.The theoretical analysis and experimental results show that the proposedmethod can ensure the privacy and accuracy of the federated learning processand can also improve the model utility for non-iid data by 7% compared tothe federated averaging method (FedAvg).
基金supported by the National Natural Science Foundation of China(Grant No.12071042)the Beijing Natural Science Foundation(Grant No.1202004)Promoting the Development of University Classification-Student Innovation and Entrepreneurship Training Programme(Grant No.5112410857)。
文摘In this paper,we propose a symmetric difference data enhancement physics-informed neural network(SDE-PINN)to study soliton solutions for discrete nonlinear lattice equations(NLEs).By considering known and unknown symmetric points,numerical simulations are conducted to one-soliton and two-soliton solutions of a discrete KdV equation,as well as a one-soliton solution of a discrete Toda lattice equation.Compared with the existing discrete deep learning approach,the numerical results reveal that within the specified spatiotemporal domain,the prediction accuracy by SDE-PINN is excellent regardless of the interior or extrapolation prediction,with a significant reduction in training time.The proposed data enhancement technique and symmetric structure development provides a new perspective for the deep learning approach to solve discrete NLEs.The newly proposed SDE-PINN can also be applied to solve continuous nonlinear equations and other discrete NLEs numerically.
基金Joint Seismological Science Foundation of China (198009).
文摘A new concept is suggested on tectonomagnetic research about the noise in simultaneous geomagnetic difference data caused by the effect of Sq local-time variation, together with the method of theoretical calculation. The level of the noise and its contribution to the total noises of the differences data are analyzed. The result indicates that the noise increases linearly with the increase of the distance between the two stations in the range of 40° longitude-difference, and its increasing rate is about 0.4 nT/（°）at latitude 40°N. The example calculated at a pair of sites with longitude-difference 0.357°, shows that the noise is about one fifth of the total noises of the difference data on geomagnetic quiet-day.
基金supported by Natural Science Foundation of China(NSFC)(Grant No.31727901)。
文摘A novel weather radar system with distributed phased-array front-ends was developed. The specifications and preliminary data synthesis of this system are presented, which comprises one back-end and three or more front-ends. Each front-end, which utilizes a phased-array digital beamforming technology, sequentially transmits four 22.5°-width beams to cover the 0°–90° elevational scan within about 0.05 s. The azimuthal detection is completed by one mechanical scan of0°–360° azimuths within about 12 s volume-scan update time. In the case of three front-ends, they are deployed according to an acute triangle to form a fine detection area(FDA). Because of the triangular deployment of multiple phased-array front-ends and a unique synchronized azimuthal scanning(SAS) rule, this new radar system is named Array Weather Radar(AWR). The back-end controls the front-ends to scan strictly in accordance with the SAS rule that assures the data time differences(DTD) among the three front-ends are less than 2 s for the same detection point in the FDA. The SAS can maintain DTD < 2 s for an expanded seven-front-end AWR. With the smallest DTD, gridded wind fields are derived from AWR data, by sampling of the interpolated grid, onto a rectangular grid of 100 m ×100 m ×100 m at a 12 s temporal resolution in the FDA. The first X-band single-polarized three-front-end AWR was deployed in field experiments in 2018 at Huanghua International Airport, China. Having completed the data synthesis and processing, the preliminary observation results of the first AWR are described herein.
