Network embedding(NE)tries to learn the potential properties of complex networks represented in a low-dimensional feature space.However,the existing deep learningbased NE methods are time-consuming as they need to tra...Network embedding(NE)tries to learn the potential properties of complex networks represented in a low-dimensional feature space.However,the existing deep learningbased NE methods are time-consuming as they need to train a dense architecture for deep neural networks with extensive unknown weight parameters.A sparse deep autoencoder(called SPDNE)for dynamic NE is proposed,aiming to learn the network structures while preserving the node evolution with a low computational complexity.SPDNE tries to use an optimal sparse architecture to replace the fully connected architecture in the deep autoencoder while maintaining the performance of these models in the dynamic NE.Then,an adaptive simulated algorithm to find the optimal sparse architecture for the deep autoencoder is proposed.The performance of SPDNE over three dynamical NE models(i.e.sparse architecture-based deep autoencoder method,DynGEM,and ElvDNE)is evaluated on three well-known benchmark networks and five real-world networks.The experimental results demonstrate that SPDNE can reduce about 70%of weight parameters of the architecture for the deep autoencoder during the training process while preserving the performance of these dynamical NE models.The results also show that SPDNE achieves the highest accuracy on 72 out of 96 edge prediction and network reconstruction tasks compared with the state-of-the-art dynamical NE algorithms.展开更多
With the development of science, economy and society, the needs for research and exploration of deep space have entered a rapid and stable development stage. Deep Space Optical Network(DSON) is expected to become an i...With the development of science, economy and society, the needs for research and exploration of deep space have entered a rapid and stable development stage. Deep Space Optical Network(DSON) is expected to become an important foundation and inevitable development trend of future deepspace communication. In this paper, we design a deep space node model which is capable of combining the space division multiplexing with frequency division multiplexing. Furthermore, we propose the directional flooding routing algorithm(DFRA) for DSON based on our node model. This scheme selectively forwards the data packets in the routing, so that the energy consumption can be reduced effectively because only a portion of nodes will participate the flooding routing. Simulation results show that, compared with traditional flooding routing algorithm(TFRA), the DFRA can avoid the non-directional and blind transmission. Therefore, the energy consumption in message routing will be reduced and the lifespan of DSON can also be prolonged effectively. Although the complexity of routing implementation is slightly increased compared with TFRA, the energy of nodes can be saved and the transmission rate is obviously improved in DFRA. Thus the overall performance of DSON can be significantly improved.展开更多
Deep space communications has played an important role in deep space exploration. Compared with common satellite and terrestrial communications, deep space communications faces more challenging environment. The paper ...Deep space communications has played an important role in deep space exploration. Compared with common satellite and terrestrial communications, deep space communications faces more challenging environment. The paper investigated the unique features of deep space communica-tions in detail, discussed the key technologies and its development trends for deep space communica-tions.展开更多
This research proposes a method called enhanced collaborative andgeometric multi-kernel learning (E-CGMKL) that can enhance the CGMKLalgorithm which deals with multi-class classification problems with non-lineardata d...This research proposes a method called enhanced collaborative andgeometric multi-kernel learning (E-CGMKL) that can enhance the CGMKLalgorithm which deals with multi-class classification problems with non-lineardata distributions. CGMKL combines multiple kernel learning with softmaxfunction using the framework of multi empirical kernel learning (MEKL) inwhich empirical kernel mapping (EKM) provides explicit feature constructionin the high dimensional kernel space. CGMKL ensures the consistent outputof samples across kernel spaces and minimizes the within-class distance tohighlight geometric features of multiple classes. However, the kernels constructed by CGMKL do not have any explicit relationship among them andtry to construct high dimensional feature representations independently fromeach other. This could be disadvantageous for learning on datasets with complex hidden structures. To overcome this limitation, E-CGMKL constructskernel spaces from hidden layers of trained deep neural networks (DNN).Due to the nature of the DNN architecture, these kernel spaces not onlyprovide multiple feature representations but also inherit the compositionalhierarchy of the hidden layers, which might be beneficial for enhancing thepredictive performance of the CGMKL algorithm on complex data withnatural hierarchical structures, for example, image data. Furthermore, ourproposed scheme handles image data by constructing kernel spaces from aconvolutional neural network (CNN). Considering the effectiveness of CNNarchitecture on image data, these kernel spaces provide a major advantageover the CGMKL algorithm which does not exploit the CNN architecture forconstructing kernel spaces from image data. Additionally, outputs of hiddenlayers directly provide features for kernel spaces and unlike CGMKL, do notrequire an approximate MEKL framework. E-CGMKL combines the consistency and geometry preserving aspects of CGMKL with the compositionalhierarchy of kernel spaces extracted from DNN hidden layers to enhance the predictive performance of CGMKL significantly. The experimental results onvarious data sets demonstrate the superior performance of the E-CGMKLalgorithm compared to other competing methods including the benchmarkCGMKL.展开更多
In order to solve the problem that the star point positioning accuracy of the star sensor in near space is decreased due to atmospheric background stray light and rapid maneuvering of platform, this paper proposes a s...In order to solve the problem that the star point positioning accuracy of the star sensor in near space is decreased due to atmospheric background stray light and rapid maneuvering of platform, this paper proposes a star point positioning algorithm based on the capsule network whose input and output are both vectors. First, a PCTL (Probability-Coordinate Transformation Layer) is designed to represent the mapping relationship between the probability output of the capsule network and the star point sub-pixel coordinates. Then, Coordconv Layer is introduced to implement explicit encoding of space information and the probability is used as the centroid weight to achieve the conversion between probability and star point sub-pixel coordinates, which improves the network’s ability to perceive star point positions. Finally, based on the dynamic imaging principle of star sensors and the characteristics of near-space environment, a star map dataset for algorithm training and testing is constructed. The simulation results show that the proposed algorithm reduces the MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) of the star point positioning by 36.1% and 41.7% respectively compared with the traditional algorithm. The research results can provide important theory and technical support for the scheme design, index demonstration, test and evaluation of large dynamic star sensors in near space.展开更多
正交时频空(Orthogonal Time Frequency Space, OTFS)调制技术凭借对多普勒频移的优良抗性,保证了高动态场景下的可靠性通信。与大多数OTFS信号检测方案相比,基于深度学习(Deep Learning, DL)的OTFS检测器不需要耗费高额的导频能量,以...正交时频空(Orthogonal Time Frequency Space, OTFS)调制技术凭借对多普勒频移的优良抗性,保证了高动态场景下的可靠性通信。与大多数OTFS信号检测方案相比,基于深度学习(Deep Learning, DL)的OTFS检测器不需要耗费高额的导频能量,以此获得精确的信道状态信息。基于多维输入的卷积神经网络(Convolutional Neural Networks, CNN)和一维输入的深度神经网络(Deep Neural Networks, DNN),搭建了OTFS信号检测模型,并结合OTFS的输入输出关系,以模型驱动,提出一种部分输入方法。与数据驱动DL相比,该方法沿时延轴截断输入数据,仅向网络输入与待检测信号相关性强的部分接收信号。该方法不仅减小了数据驱动CNN和DNN的训练参数量,降低了训练复杂度,而且检测性能也不弱于传统的线性最小均方误差(Linear Minimum Mean Square Error, LMMSE)算法。展开更多
基金National Natural Science Foundation of China,Grant/Award Numbers:62173236,61876110,61806130,61976142,82304204.
文摘Network embedding(NE)tries to learn the potential properties of complex networks represented in a low-dimensional feature space.However,the existing deep learningbased NE methods are time-consuming as they need to train a dense architecture for deep neural networks with extensive unknown weight parameters.A sparse deep autoencoder(called SPDNE)for dynamic NE is proposed,aiming to learn the network structures while preserving the node evolution with a low computational complexity.SPDNE tries to use an optimal sparse architecture to replace the fully connected architecture in the deep autoencoder while maintaining the performance of these models in the dynamic NE.Then,an adaptive simulated algorithm to find the optimal sparse architecture for the deep autoencoder is proposed.The performance of SPDNE over three dynamical NE models(i.e.sparse architecture-based deep autoencoder method,DynGEM,and ElvDNE)is evaluated on three well-known benchmark networks and five real-world networks.The experimental results demonstrate that SPDNE can reduce about 70%of weight parameters of the architecture for the deep autoencoder during the training process while preserving the performance of these dynamical NE models.The results also show that SPDNE achieves the highest accuracy on 72 out of 96 edge prediction and network reconstruction tasks compared with the state-of-the-art dynamical NE algorithms.
基金supported by National Natural Science Foundation of China (61471109, 61501104 and 91438110)Fundamental Research Funds for the Central Universities ( N140405005 , N150401002 and N150404002)Open Fund of IPOC (BUPT, IPOC2015B006)
文摘With the development of science, economy and society, the needs for research and exploration of deep space have entered a rapid and stable development stage. Deep Space Optical Network(DSON) is expected to become an important foundation and inevitable development trend of future deepspace communication. In this paper, we design a deep space node model which is capable of combining the space division multiplexing with frequency division multiplexing. Furthermore, we propose the directional flooding routing algorithm(DFRA) for DSON based on our node model. This scheme selectively forwards the data packets in the routing, so that the energy consumption can be reduced effectively because only a portion of nodes will participate the flooding routing. Simulation results show that, compared with traditional flooding routing algorithm(TFRA), the DFRA can avoid the non-directional and blind transmission. Therefore, the energy consumption in message routing will be reduced and the lifespan of DSON can also be prolonged effectively. Although the complexity of routing implementation is slightly increased compared with TFRA, the energy of nodes can be saved and the transmission rate is obviously improved in DFRA. Thus the overall performance of DSON can be significantly improved.
基金Supported by the National Natural Science Foundation of China (No. 60972061,60972062,and 61032004)the National High Technology Research and Development Program of China ("863" Program) (No. 2008AA12A204)
文摘Deep space communications has played an important role in deep space exploration. Compared with common satellite and terrestrial communications, deep space communications faces more challenging environment. The paper investigated the unique features of deep space communica-tions in detail, discussed the key technologies and its development trends for deep space communica-tions.
文摘This research proposes a method called enhanced collaborative andgeometric multi-kernel learning (E-CGMKL) that can enhance the CGMKLalgorithm which deals with multi-class classification problems with non-lineardata distributions. CGMKL combines multiple kernel learning with softmaxfunction using the framework of multi empirical kernel learning (MEKL) inwhich empirical kernel mapping (EKM) provides explicit feature constructionin the high dimensional kernel space. CGMKL ensures the consistent outputof samples across kernel spaces and minimizes the within-class distance tohighlight geometric features of multiple classes. However, the kernels constructed by CGMKL do not have any explicit relationship among them andtry to construct high dimensional feature representations independently fromeach other. This could be disadvantageous for learning on datasets with complex hidden structures. To overcome this limitation, E-CGMKL constructskernel spaces from hidden layers of trained deep neural networks (DNN).Due to the nature of the DNN architecture, these kernel spaces not onlyprovide multiple feature representations but also inherit the compositionalhierarchy of the hidden layers, which might be beneficial for enhancing thepredictive performance of the CGMKL algorithm on complex data withnatural hierarchical structures, for example, image data. Furthermore, ourproposed scheme handles image data by constructing kernel spaces from aconvolutional neural network (CNN). Considering the effectiveness of CNNarchitecture on image data, these kernel spaces provide a major advantageover the CGMKL algorithm which does not exploit the CNN architecture forconstructing kernel spaces from image data. Additionally, outputs of hiddenlayers directly provide features for kernel spaces and unlike CGMKL, do notrequire an approximate MEKL framework. E-CGMKL combines the consistency and geometry preserving aspects of CGMKL with the compositionalhierarchy of kernel spaces extracted from DNN hidden layers to enhance the predictive performance of CGMKL significantly. The experimental results onvarious data sets demonstrate the superior performance of the E-CGMKLalgorithm compared to other competing methods including the benchmarkCGMKL.
文摘In order to solve the problem that the star point positioning accuracy of the star sensor in near space is decreased due to atmospheric background stray light and rapid maneuvering of platform, this paper proposes a star point positioning algorithm based on the capsule network whose input and output are both vectors. First, a PCTL (Probability-Coordinate Transformation Layer) is designed to represent the mapping relationship between the probability output of the capsule network and the star point sub-pixel coordinates. Then, Coordconv Layer is introduced to implement explicit encoding of space information and the probability is used as the centroid weight to achieve the conversion between probability and star point sub-pixel coordinates, which improves the network’s ability to perceive star point positions. Finally, based on the dynamic imaging principle of star sensors and the characteristics of near-space environment, a star map dataset for algorithm training and testing is constructed. The simulation results show that the proposed algorithm reduces the MAE (Mean Absolute Error) and RMSE (Root Mean Square Error) of the star point positioning by 36.1% and 41.7% respectively compared with the traditional algorithm. The research results can provide important theory and technical support for the scheme design, index demonstration, test and evaluation of large dynamic star sensors in near space.
