在传统的联邦学习系统中,中央服务器在接收到参与方上传的本地模型后,没有对本地模型的合法性进行验证,可能导致后续模型不可靠即模型存储可靠性问题。针对该问题提出一种基于区块链隐私增强的联邦学习可验证安全存储方案(blockchain-ba...在传统的联邦学习系统中,中央服务器在接收到参与方上传的本地模型后,没有对本地模型的合法性进行验证,可能导致后续模型不可靠即模型存储可靠性问题。针对该问题提出一种基于区块链隐私增强的联邦学习可验证安全存储方案(blockchain-based federated learning for model verified security storage, BFL4MVSS)。利用区块链技术,提出验证者投票机制,保证存储模型的可靠性;引入一种奖励机制,允许系统在一定数量的恶意节点存在下正常运行。采用PyTorch框架在MNIST数据集上进行仿真验证,结果表明,其能够在15%的恶意节点的参与下确保模型具有良好的表现。展开更多
The bistatic Synthetic Aperture Radar (SAR) systems with separate transmitter and receiver antennas provide a new potential to imaging in the forward-looking geometry. Analysis of the Doppler property in this paper in...The bistatic Synthetic Aperture Radar (SAR) systems with separate transmitter and receiver antennas provide a new potential to imaging in the forward-looking geometry. Analysis of the Doppler property in this paper indicates the feasibility of Bistatic Forward-Looking (BFL) SAR imaging. Considering the different Doppler property determined by the two platforms in BFL SAR, a new 2-D point target spectrum is derived in our study. Based on the spectrum, an imaging method is chosen for the configuration, and the point target simulation validates the analysis.展开更多
文摘在传统的联邦学习系统中,中央服务器在接收到参与方上传的本地模型后,没有对本地模型的合法性进行验证,可能导致后续模型不可靠即模型存储可靠性问题。针对该问题提出一种基于区块链隐私增强的联邦学习可验证安全存储方案(blockchain-based federated learning for model verified security storage, BFL4MVSS)。利用区块链技术,提出验证者投票机制,保证存储模型的可靠性;引入一种奖励机制,允许系统在一定数量的恶意节点存在下正常运行。采用PyTorch框架在MNIST数据集上进行仿真验证,结果表明,其能够在15%的恶意节点的参与下确保模型具有良好的表现。
基金Supported by the National Natural Science Foundation of China (No. 61071165)the Aviation Science Foundation (No. 20102052024)
文摘The bistatic Synthetic Aperture Radar (SAR) systems with separate transmitter and receiver antennas provide a new potential to imaging in the forward-looking geometry. Analysis of the Doppler property in this paper indicates the feasibility of Bistatic Forward-Looking (BFL) SAR imaging. Considering the different Doppler property determined by the two platforms in BFL SAR, a new 2-D point target spectrum is derived in our study. Based on the spectrum, an imaging method is chosen for the configuration, and the point target simulation validates the analysis.
