ISSN 
2023 Vol. 52, No. 1
2023, 52(1): 2-7.
doi: 10.12178/1001-0548.2022187
Abstract:
Shared randomness and quantum entanglement are important resources for many information processing tasks, where the former is also called classical correlation. In the classical correlation generation problem, we study the minimum amount of shared randomness or quantum entanglement needed to produce a target classical correlation. Here we review classical protocol, quantum protocol, and classical and quantum hybrid protocols for generating classical correlations. First, in classical protocol and quantum protocol the minimum amount of shared randomness and quantum entanglement required are characterized by nonnegative rank and positive semidefinite rank, respectively. Based on these results, sharing prior quantum entanglement shows exponentially advantage over sharing prior randomness in such a task. Second, since it is hard to access large-scale quantum system in the near future, classical-quantum hybrid protocol is also introduced to produce large scale classical correlations. When the size of manipulable quantum systems is limited, the minimum amount of extra classical resources needed to generate a target classical correlation is characterized by the concept of\begin{document}$ k $\end{document} ![]()
![]()
Shared randomness and quantum entanglement are important resources for many information processing tasks, where the former is also called classical correlation. In the classical correlation generation problem, we study the minimum amount of shared randomness or quantum entanglement needed to produce a target classical correlation. Here we review classical protocol, quantum protocol, and classical and quantum hybrid protocols for generating classical correlations. First, in classical protocol and quantum protocol the minimum amount of shared randomness and quantum entanglement required are characterized by nonnegative rank and positive semidefinite rank, respectively. Based on these results, sharing prior quantum entanglement shows exponentially advantage over sharing prior randomness in such a task. Second, since it is hard to access large-scale quantum system in the near future, classical-quantum hybrid protocol is also introduced to produce large scale classical correlations. When the size of manipulable quantum systems is limited, the minimum amount of extra classical resources needed to generate a target classical correlation is characterized by the concept of
2023, 52(1): 8-13.
doi: 10.12178/1001-0548.2022429
Abstract:
In the quantum many-body system, the calculation of the ground state is the key target problem. Variational quantum eigensolver (VQE) is a variational ground state solution algorithm based on quantum computation. However, since it requires combining quantum circuit and classical variational algorithm, the complexity of the quantum circuit and the choice of variational algorithm becomes extremely important. This paper focuses on quantum molecular systems and proposes a variational ground state solver. It uses the single-electron reduced density matrix analysis to obtain the number of electron occupations under the appearance of natural molecular orbitals. According to the number of occupations, the Hamiltonian of the system and the corresponding unitary couple cluster (UCC) ansatz circuit are greatly simplified. Secondly, the variational quantum imaginary time evolution algorithm is used to replace the commonly used gradient algorithm in VQE, which is not easily influenced by the gradient distribution of the parameter space, causing the variational process to converge more quickly and robustly.
In the quantum many-body system, the calculation of the ground state is the key target problem. Variational quantum eigensolver (VQE) is a variational ground state solution algorithm based on quantum computation. However, since it requires combining quantum circuit and classical variational algorithm, the complexity of the quantum circuit and the choice of variational algorithm becomes extremely important. This paper focuses on quantum molecular systems and proposes a variational ground state solver. It uses the single-electron reduced density matrix analysis to obtain the number of electron occupations under the appearance of natural molecular orbitals. According to the number of occupations, the Hamiltonian of the system and the corresponding unitary couple cluster (UCC) ansatz circuit are greatly simplified. Secondly, the variational quantum imaginary time evolution algorithm is used to replace the commonly used gradient algorithm in VQE, which is not easily influenced by the gradient distribution of the parameter space, causing the variational process to converge more quickly and robustly.
2023, 52(1): 14-22.
doi: 10.12178/1001-0548.2022292
Abstract:
The decoding analysis of gait features based on electroencephalogram (EEG) and the reliable recognition and prediction of motion intention are the core problems of brain-computer interface (BCI) based human-machine hybrid rehabilitation training system and intelligent walking robot. In order to realize the recognition of the most basic gait processes such as standing, sitting and resting states, this study proposes a feature representation method based on multi-layer functional brain network using EEG. Combined with the statistical analysis of various network features, these parameters sensitive to different movements are determined, and support vector machine, linear discriminant analysis, logistic regression and naive bayes algorithms are applied to complete the classification of different actions. Experiment results show the proposed method can realize the recognition of the three actions, and the recognition accuracy of standing, sitting and resting state is higher than 71% and the highest accuracy is 77% for 13 subjects. Multi-layer brain network analysis shows the motion action of lower limb can weaken the interdependence between brain regions, resulting the sparsification of the topology structure. This study has certain reference value for understanding the changes of brain cognitive process during lower limb movement, carrying out BCI based rehabilitation strategies, and developing corresponding rehabilitation systems.
The decoding analysis of gait features based on electroencephalogram (EEG) and the reliable recognition and prediction of motion intention are the core problems of brain-computer interface (BCI) based human-machine hybrid rehabilitation training system and intelligent walking robot. In order to realize the recognition of the most basic gait processes such as standing, sitting and resting states, this study proposes a feature representation method based on multi-layer functional brain network using EEG. Combined with the statistical analysis of various network features, these parameters sensitive to different movements are determined, and support vector machine, linear discriminant analysis, logistic regression and naive bayes algorithms are applied to complete the classification of different actions. Experiment results show the proposed method can realize the recognition of the three actions, and the recognition accuracy of standing, sitting and resting state is higher than 71% and the highest accuracy is 77% for 13 subjects. Multi-layer brain network analysis shows the motion action of lower limb can weaken the interdependence between brain regions, resulting the sparsification of the topology structure. This study has certain reference value for understanding the changes of brain cognitive process during lower limb movement, carrying out BCI based rehabilitation strategies, and developing corresponding rehabilitation systems.
2023, 52(1): 23-29.
doi: 10.12178/1001-0548.2022200
Abstract:
Digital modeling of 3D bone porous structure based on bone slice images is the technical basis of bone tissue engineering and a research hotspot in the field of biomedical engineering. The quality of bone slice image determines the accuracy of the digital model of bone porous structure. However, problems such as data loss, image damage or too small image size may occur in the process of acquiring bone section images, resulting in only partial slice images, and thus such incomplete image information seriously affects the accuracy of 3D porous bone structure modeling. In order to solve this problem, an improved conditional generative adversarial network for complete reconstruction and repair of partial bone slice images was proposed, that is, on the basis of conditional generative adversarial network, the nested residual dense block is added to the generator, and the polarized self-attention module is added to the discriminator. Morphological function analysis and partial porosity distribution study were performed to evaluate the similarity between the reconstructed images and the real bone porous images. The results show that the network can accurately and stably reconstruct complete bone porous slice images.
Digital modeling of 3D bone porous structure based on bone slice images is the technical basis of bone tissue engineering and a research hotspot in the field of biomedical engineering. The quality of bone slice image determines the accuracy of the digital model of bone porous structure. However, problems such as data loss, image damage or too small image size may occur in the process of acquiring bone section images, resulting in only partial slice images, and thus such incomplete image information seriously affects the accuracy of 3D porous bone structure modeling. In order to solve this problem, an improved conditional generative adversarial network for complete reconstruction and repair of partial bone slice images was proposed, that is, on the basis of conditional generative adversarial network, the nested residual dense block is added to the generator, and the polarized self-attention module is added to the discriminator. Morphological function analysis and partial porosity distribution study were performed to evaluate the similarity between the reconstructed images and the real bone porous images. The results show that the network can accurately and stably reconstruct complete bone porous slice images.
2023, 52(1): 30-37.
doi: 10.12178/1001-0548.2022162
Abstract:
Distributed through-the-wall imaging radar (DTWIR) deploys multiple through-the-wall radar nodes in multiple viewpoints for cooperative detection of indoor targets, which can effectively compensate for the shortage of traditional single-view through-the-wall target detection. In this paper, a registration algorithm for DTWIR is proposed, which calculates the positions of the radar nodes based on their detection results. Firstly, the measurements of each radar node are modeled as Gaussian mixture model. Then an optimization problem which solves the optimal coordinate transformation parameters is formulated for radar registration. Finally, the optimization problem is solved via the particle swarm optimization method. The proposed algorithm is validated by simulation and experimental results.
Distributed through-the-wall imaging radar (DTWIR) deploys multiple through-the-wall radar nodes in multiple viewpoints for cooperative detection of indoor targets, which can effectively compensate for the shortage of traditional single-view through-the-wall target detection. In this paper, a registration algorithm for DTWIR is proposed, which calculates the positions of the radar nodes based on their detection results. Firstly, the measurements of each radar node are modeled as Gaussian mixture model. Then an optimization problem which solves the optimal coordinate transformation parameters is formulated for radar registration. Finally, the optimization problem is solved via the particle swarm optimization method. The proposed algorithm is validated by simulation and experimental results.
2023, 52(1): 38-43.
doi: 10.12178/1001-0548.2022294
Abstract:
The human brain is a highly complex and large-scale nonlinear dynamic system, and its dynamic behavior is closely related to human intelligent activities. The artificial neural network based on memristors can not only better simulate the working mechanism of human brain, but also its nonlinear characteristics can bring richer dynamic behavior to the neural network. In order to further exploit the advantages of neural networks, a new memristor model with negative resistance is introduced in this paper. This model breaks the restriction of the resistance state polarity of the original memristor, and provides a richer variety of performance for the memristor to act as a neural network synaptic bionic device. A new Hopfield neural network (HNN) based on the memristor model is constructed, which further strengthens the negative feedback function of the Hopfield neural network and makes it exhibit richer and more complex dynamic behaviors. The experimental results show that the new memristive Hopfield neural network has rich dynamic behavior characteristics and some chaotic phenomena. Under the conditions of different values of memristor’s parameters and weight matrix, the changes of phase trajectory and Lyapunov exponent of the system are observed, and comparison with the same type of networks are done, which further proves the effectiveness of the proposed neural network. At the same time, the complex dynamic characteristics also provide research support for applications in data processing and image encryption.
The human brain is a highly complex and large-scale nonlinear dynamic system, and its dynamic behavior is closely related to human intelligent activities. The artificial neural network based on memristors can not only better simulate the working mechanism of human brain, but also its nonlinear characteristics can bring richer dynamic behavior to the neural network. In order to further exploit the advantages of neural networks, a new memristor model with negative resistance is introduced in this paper. This model breaks the restriction of the resistance state polarity of the original memristor, and provides a richer variety of performance for the memristor to act as a neural network synaptic bionic device. A new Hopfield neural network (HNN) based on the memristor model is constructed, which further strengthens the negative feedback function of the Hopfield neural network and makes it exhibit richer and more complex dynamic behaviors. The experimental results show that the new memristive Hopfield neural network has rich dynamic behavior characteristics and some chaotic phenomena. Under the conditions of different values of memristor’s parameters and weight matrix, the changes of phase trajectory and Lyapunov exponent of the system are observed, and comparison with the same type of networks are done, which further proves the effectiveness of the proposed neural network. At the same time, the complex dynamic characteristics also provide research support for applications in data processing and image encryption.
2023, 52(1): 44-53.
doi: 10.12178/1001-0548.2022039
Abstract:
To solve the problem of low accuracy and high false alarm rate of synthetic aperture radar (SAR) nearshore area vessel detection, a new SAR nearshore area vessel detection method based on improved attention mechanism and rotating frame is proposed. Firstly, the feature extraction capability of the network was enhanced by improving the coordinate attention mechanism and introducing it into the feature extraction network. Secondly, the angle classification head was added and the two-dimensional Gaussian distribution was introduced to calculate the KL divergence between the prediction distribution and the target distribution, so as to evaluate the loss value of the rotating frame and complete the angle information extraction of the target. Then, based on the anchor frameless (AF) mechanism of YOLOX algorithm, the model can be made lightweight and the positioning accuracy can be further improved by reducing the redundancy of candidate frames. Finally, the model was tested on the open dataset Offical - SSDD, and the inference verification was performed on the embedded platform (NVIDIA Jetson AGX Xavier). The calculation parameter of the algorithm model is only 1.14M, and the average detection accuracy of the algorithm model is 18.77%, higher than that of the YOLOX model in the nearshore condition, and the overall detection accuracy reaches 94.2%. The verification results show that the algorithm is suitable for dense ship target detection in any direction in complex scenes and can meet the requirements of real-time processing.
To solve the problem of low accuracy and high false alarm rate of synthetic aperture radar (SAR) nearshore area vessel detection, a new SAR nearshore area vessel detection method based on improved attention mechanism and rotating frame is proposed. Firstly, the feature extraction capability of the network was enhanced by improving the coordinate attention mechanism and introducing it into the feature extraction network. Secondly, the angle classification head was added and the two-dimensional Gaussian distribution was introduced to calculate the KL divergence between the prediction distribution and the target distribution, so as to evaluate the loss value of the rotating frame and complete the angle information extraction of the target. Then, based on the anchor frameless (AF) mechanism of YOLOX algorithm, the model can be made lightweight and the positioning accuracy can be further improved by reducing the redundancy of candidate frames. Finally, the model was tested on the open dataset Offical - SSDD, and the inference verification was performed on the embedded platform (NVIDIA Jetson AGX Xavier). The calculation parameter of the algorithm model is only 1.14M, and the average detection accuracy of the algorithm model is 18.77%, higher than that of the YOLOX model in the nearshore condition, and the overall detection accuracy reaches 94.2%. The verification results show that the algorithm is suitable for dense ship target detection in any direction in complex scenes and can meet the requirements of real-time processing.
2023, 52(1): 54-63.
doi: 10.12178/1001-0548.2022018
Abstract:
The low frequency radio frequency identification (LF RFID)-based detection technology is popular for access detection systems because it is convenient and low-cost. However, the performance of LF RFID technology is easily affected by complex environmental factors, especially in the metal environment, leading to low detection accuracy and poor stability. To tackle this problem, we propose an access detection method with an accurate positioning algorithm based on improved maximum likelihood in this paper. The positioning algorithm consists of two stages: training stage and positioning stage. In the training stage, a correction fingerprint database is established. The database is used to correct the ranging error due to environmental factors. In the positioning stage, after correcting the ranging measurement error, the maximum likelihood algorithm is used to locate tags carried by subjects for detecting the entry/exit status. Some experimental results show that the detection accuracy can reach 3 cm and the accuracy of detecting entering/exiting status under two walk strategies is up to 99% or more, which is enough for application.
The low frequency radio frequency identification (LF RFID)-based detection technology is popular for access detection systems because it is convenient and low-cost. However, the performance of LF RFID technology is easily affected by complex environmental factors, especially in the metal environment, leading to low detection accuracy and poor stability. To tackle this problem, we propose an access detection method with an accurate positioning algorithm based on improved maximum likelihood in this paper. The positioning algorithm consists of two stages: training stage and positioning stage. In the training stage, a correction fingerprint database is established. The database is used to correct the ranging error due to environmental factors. In the positioning stage, after correcting the ranging measurement error, the maximum likelihood algorithm is used to locate tags carried by subjects for detecting the entry/exit status. Some experimental results show that the detection accuracy can reach 3 cm and the accuracy of detecting entering/exiting status under two walk strategies is up to 99% or more, which is enough for application.
2023, 52(1): 64-73.
doi: 10.12178/1001-0548.2022102
Abstract:
Wireless communication network traffic prediction is of great significance to operators in network construction, base station wireless resource management and user experience improvement. However, the existing centralized algorithm models face the problems of complexity and timeliness, thus difficult to meet the traffic prediction of the whole city scale. A distributed urban global traffic prediction algorithm Fed-DenseNet is proposed in this paper. Each edge computing server of the algorithm performs collaborative training under the coordination of the central server, and the central server uses KL (Kullback-Leibler) divergence to select regional traffic models with similar traffic distribution and uses the federated average algorithm to fuse the parameters of these regional traffic models. In this way, the urban global traffic prediction can be realized with lower complexity and communication cost. In addition, the traffic in different areas within the city is highly differentiated, so how to improve the accuracy of model prediction is also facing challenges. Based on Fed-Densenet algorithm, a personalized federated learning algorithm p-Fed-DenseNet based on cooperative game is proposed. Each regional data feature in the region is taken as a participant of cooperative game, and local features are screened by the super-additivity criterion of cooperative game, so as to achieve the purpose of both improving the generalization of the model and maintaining the accurate description of local traffic.
Wireless communication network traffic prediction is of great significance to operators in network construction, base station wireless resource management and user experience improvement. However, the existing centralized algorithm models face the problems of complexity and timeliness, thus difficult to meet the traffic prediction of the whole city scale. A distributed urban global traffic prediction algorithm Fed-DenseNet is proposed in this paper. Each edge computing server of the algorithm performs collaborative training under the coordination of the central server, and the central server uses KL (Kullback-Leibler) divergence to select regional traffic models with similar traffic distribution and uses the federated average algorithm to fuse the parameters of these regional traffic models. In this way, the urban global traffic prediction can be realized with lower complexity and communication cost. In addition, the traffic in different areas within the city is highly differentiated, so how to improve the accuracy of model prediction is also facing challenges. Based on Fed-Densenet algorithm, a personalized federated learning algorithm p-Fed-DenseNet based on cooperative game is proposed. Each regional data feature in the region is taken as a participant of cooperative game, and local features are screened by the super-additivity criterion of cooperative game, so as to achieve the purpose of both improving the generalization of the model and maintaining the accurate description of local traffic.
2023, 52(1): 74-84.
doi: 10.12178/1001-0548.2022287
Abstract:
The Metaverse can be seen as an extension of the real world, a digital virtual world born out of the real world but also parallels and interacts with it. It connects virtual and reality, enriches people's perception and experience, extends people's creativity, and brings more possibilities for the development of human society. The Metaverse is the coupling between people's imagination and real technical conditions and relies on the system integration and comprehensive application of new technologies such as human-computer interaction, artificial intelligence, blockchain, and the internet of things. In this paper, by combining and analyzing the concepts of the Metaverse and its supporting technologies, a variety of technologies are summarized and unified under a technical framework, the "BIGCHINA" technology system. It includes eight types of technologies breaking through the technology from a single technology to a certain extent, which has specific reference significance for understanding the overall development status and future trends of the current Metaverse technology.
The Metaverse can be seen as an extension of the real world, a digital virtual world born out of the real world but also parallels and interacts with it. It connects virtual and reality, enriches people's perception and experience, extends people's creativity, and brings more possibilities for the development of human society. The Metaverse is the coupling between people's imagination and real technical conditions and relies on the system integration and comprehensive application of new technologies such as human-computer interaction, artificial intelligence, blockchain, and the internet of things. In this paper, by combining and analyzing the concepts of the Metaverse and its supporting technologies, a variety of technologies are summarized and unified under a technical framework, the "BIGCHINA" technology system. It includes eight types of technologies breaking through the technology from a single technology to a certain extent, which has specific reference significance for understanding the overall development status and future trends of the current Metaverse technology.
2023, 52(1): 85-94.
doi: 10.12178/1001-0548.2022009
Abstract:
There are two problems to be solved in multi-sensors wireless sensor networks: low data collection efficiency and the risk of data leakage when a large amount of data is processed in sensor cloud. Owing to these reasons, we devise a safe, energy-saving, and efficient distributed edge collaborative sensor network resource selection architecture firstly. Secondly, to address first problem, an edge analysis node selection (edge collaborative analysis node selection, ECANS) algorithm is proposed. Through the analysis of user requests, the best strategy of sensor network nodes is obtained to reduce the node's delay and energy consumption of data collection. Aiming at the second problem, an edge collaborative sensor network privacy protection data offloading model is constructed to maximize privacy entropy, and the edge resource selection strategy with the largest privacy entropy is gained through intelligent heuristic algorithm. Al last, experimental results show that ECANS algorithm can reduce node delay and energy consumption by 56.71% and 57.66% compared with effective node sensing (ENS) data collection methods. In the edge resource selection stage, the maximum privacy entropy model makes the system privacy entropy increased by 32.07% and 15.36%, compared with genetic algorithm (GA) resource selection scheme and particle swarm optimization (PSO) resource selection scheme. The latency and energy consumption of the sensor network were reduced by 46.92% and 11.26% compared with no-EC.
There are two problems to be solved in multi-sensors wireless sensor networks: low data collection efficiency and the risk of data leakage when a large amount of data is processed in sensor cloud. Owing to these reasons, we devise a safe, energy-saving, and efficient distributed edge collaborative sensor network resource selection architecture firstly. Secondly, to address first problem, an edge analysis node selection (edge collaborative analysis node selection, ECANS) algorithm is proposed. Through the analysis of user requests, the best strategy of sensor network nodes is obtained to reduce the node's delay and energy consumption of data collection. Aiming at the second problem, an edge collaborative sensor network privacy protection data offloading model is constructed to maximize privacy entropy, and the edge resource selection strategy with the largest privacy entropy is gained through intelligent heuristic algorithm. Al last, experimental results show that ECANS algorithm can reduce node delay and energy consumption by 56.71% and 57.66% compared with effective node sensing (ENS) data collection methods. In the edge resource selection stage, the maximum privacy entropy model makes the system privacy entropy increased by 32.07% and 15.36%, compared with genetic algorithm (GA) resource selection scheme and particle swarm optimization (PSO) resource selection scheme. The latency and energy consumption of the sensor network were reduced by 46.92% and 11.26% compared with no-EC.
2023, 52(1): 95-101.
doi: 10.12178/1001-0548.2022176
Abstract:
Edge intelligent computing is widely used in the fields of Internet of things (IoT), industrial control UAV cluster and so on, which has the advantages of high data processing efficiency, strong real-time performance and low network delay. However, there are many problems when edge intelligent device, edge gateways and cloud complete the task unloading, scheduling and coordination. For example, there are problems that are privacy disclosure, limited calculation force. As is known to all, federated learning allows all training devices to complete training in parallel, which greatly improve training efficiency. However, traditional federated learning will expose the edge device’s information of the training set. So, this article propose a privacy protection scheme combining edge intelligent computing and federated learning (PPCEF). First of all, we propose a lightweight privacy protection protocol based on sharing secret and weight mask, which is based on a random mask scheme of secret sharing. It can not only protect gradient privacy without losing model accuracy, but also resist equipment dropping and collusion attacks between devices, which has strong practicability. Secondly, we design an algorithm based on digital signature and hash function, which can not only achieve the integrity and consistency of the message, but also resist replay attacks. Finally, we use MNIST and CIFAR 10 data sets to prove that our scheme is safe in practice.
Edge intelligent computing is widely used in the fields of Internet of things (IoT), industrial control UAV cluster and so on, which has the advantages of high data processing efficiency, strong real-time performance and low network delay. However, there are many problems when edge intelligent device, edge gateways and cloud complete the task unloading, scheduling and coordination. For example, there are problems that are privacy disclosure, limited calculation force. As is known to all, federated learning allows all training devices to complete training in parallel, which greatly improve training efficiency. However, traditional federated learning will expose the edge device’s information of the training set. So, this article propose a privacy protection scheme combining edge intelligent computing and federated learning (PPCEF). First of all, we propose a lightweight privacy protection protocol based on sharing secret and weight mask, which is based on a random mask scheme of secret sharing. It can not only protect gradient privacy without losing model accuracy, but also resist equipment dropping and collusion attacks between devices, which has strong practicability. Secondly, we design an algorithm based on digital signature and hash function, which can not only achieve the integrity and consistency of the message, but also resist replay attacks. Finally, we use MNIST and CIFAR 10 data sets to prove that our scheme is safe in practice.
2023, 52(1): 102-107.
doi: 10.12178/1001-0548.2022064
Abstract:
By studying the operation of the cofactor of Shannon decomposition, six attributes of symmetric variables and independent variables in NP equivalent transformation are found. By making full use of the invariance of the symmetry and independence of variables after NP transformation, the phase uncertainty of independent variables, the unavailability of both independent variables to identify other variables and other variables to identify independent variables in the process of matching, we propose an NPN equivalent matching algorithm based on canonical form. We performed matching experiments on the 7-22 variable Boolean functions of a large number of MCNC benchmark circuits and randomly generated circuits.. The experimental results show that the algorithm of this paper reduces the search space in the matching process by 58.8% and increases matching speed by 45.6% on the two experimental circuit sets compared with the algorithm based on higher order general signature. This indicates that the algorithm of this paper can provide faster and more effective Boolean matching for circuit optimization and circuit mapping.
By studying the operation of the cofactor of Shannon decomposition, six attributes of symmetric variables and independent variables in NP equivalent transformation are found. By making full use of the invariance of the symmetry and independence of variables after NP transformation, the phase uncertainty of independent variables, the unavailability of both independent variables to identify other variables and other variables to identify independent variables in the process of matching, we propose an NPN equivalent matching algorithm based on canonical form. We performed matching experiments on the 7-22 variable Boolean functions of a large number of MCNC benchmark circuits and randomly generated circuits.. The experimental results show that the algorithm of this paper reduces the search space in the matching process by 58.8% and increases matching speed by 45.6% on the two experimental circuit sets compared with the algorithm based on higher order general signature. This indicates that the algorithm of this paper can provide faster and more effective Boolean matching for circuit optimization and circuit mapping.
2023, 52(1): 108-115.
doi: 10.12178/1001-0548.2022090
Abstract:
Indicators of compromise (IOC), as behavioral descriptions of cyber threats, are important credentials for identifying and defending against cyberattacks. The current IOC recognition mainly adopts the deep neural network training model, and its effect depends on a large amount of training data. However, there is currently a lack of recognized datasets in the field of IOC recognition. IOC can only be manually labeled by security experts, the labeling cost is high, and it is difficult to obtain a large amount of labeling data. To solve this problem, we propose a threat intelligence IOC identification method with active learning, called ICAL (IOC identification combined with active learning). The method first selects the initial samples for manual labeling according to the representativeness of the samples; then it pseudo-labels the clustered samples according to the clustering hypothesis; finally, it continues to iteratively label the samples according to the uncertainty of the samples until the termination conditions are satisfied. Using CNNPLUS as the classification model, experiments are performed on the self-built threat intelligence dataset. The results show that ICAL reduces the labeling workload by nearly 58% compared with the traditional IOC automatic identification strategies, and the recognition accuracy rate reaches 94.2%. ICAL reduces the amount of data labeling in IOC identification with strong practicability.
Indicators of compromise (IOC), as behavioral descriptions of cyber threats, are important credentials for identifying and defending against cyberattacks. The current IOC recognition mainly adopts the deep neural network training model, and its effect depends on a large amount of training data. However, there is currently a lack of recognized datasets in the field of IOC recognition. IOC can only be manually labeled by security experts, the labeling cost is high, and it is difficult to obtain a large amount of labeling data. To solve this problem, we propose a threat intelligence IOC identification method with active learning, called ICAL (IOC identification combined with active learning). The method first selects the initial samples for manual labeling according to the representativeness of the samples; then it pseudo-labels the clustered samples according to the clustering hypothesis; finally, it continues to iteratively label the samples according to the uncertainty of the samples until the termination conditions are satisfied. Using CNNPLUS as the classification model, experiments are performed on the self-built threat intelligence dataset. The results show that ICAL reduces the labeling workload by nearly 58% compared with the traditional IOC automatic identification strategies, and the recognition accuracy rate reaches 94.2%. ICAL reduces the amount of data labeling in IOC identification with strong practicability.
2023, 52(1): 116-124.
doi: 10.12178/1001-0548.2022016
Abstract:
Deep neural networks (DNN) in object detecting tasks have witnessed significant progress in the past years. However, it relies on intensive training data with accurate bounding box annotations for a remarkable performance. Once the labelled data are hard to catch, the generalization ability of DNN is far from satisfactory. We propose a few-shot object detecting method based on a multi-stage training strategy within feature redistribution (MSFR). Based on the analyses of the distribution of source domain dataset and target domain dataset in few-shot tasks, a feature redistribution algorithm is proposed to make the feature distribution meet Gaussian distribution or quasi-Gaussian distribution. It solves the inconsistency distribution of the source domain dataset and the target domain dataset. Then, a multi-stage training algorithm is proposed, which improves the efficiency of transferring the source domain knowledge to the target domain task when only a small amount of labeled data for training in each class. Thus, our proposed method significantly improves the detection performance of few-shot target domain categories while maximizing the detection accuracy of the source domain categories. The experimental results on VOC datasets show that the proposed algorithm achieves a precision improvement of up to 9.06% on different tasks, compared with existing few-shot object detection approaches.
Deep neural networks (DNN) in object detecting tasks have witnessed significant progress in the past years. However, it relies on intensive training data with accurate bounding box annotations for a remarkable performance. Once the labelled data are hard to catch, the generalization ability of DNN is far from satisfactory. We propose a few-shot object detecting method based on a multi-stage training strategy within feature redistribution (MSFR). Based on the analyses of the distribution of source domain dataset and target domain dataset in few-shot tasks, a feature redistribution algorithm is proposed to make the feature distribution meet Gaussian distribution or quasi-Gaussian distribution. It solves the inconsistency distribution of the source domain dataset and the target domain dataset. Then, a multi-stage training algorithm is proposed, which improves the efficiency of transferring the source domain knowledge to the target domain task when only a small amount of labeled data for training in each class. Thus, our proposed method significantly improves the detection performance of few-shot target domain categories while maximizing the detection accuracy of the source domain categories. The experimental results on VOC datasets show that the proposed algorithm achieves a precision improvement of up to 9.06% on different tasks, compared with existing few-shot object detection approaches.
2023, 52(1): 125-131.
doi: 10.12178/1001-0548.2022174
Abstract:
In order to measure the relationship between features more accurately, an objective function representation method based on mutual information adaptive estimation is proposed for speaker verification systems. This objective function introduces an adaptive metric learning method, and the optimization objective is maximizing the intra-class similarity and minimizing the inter-class similarity. Meanwhile, the objective function can dynamically adjust the similarity according to the real distribution of deep features. Based on dynamically adjusting, the deep neural networks can be optimized towards the direction of stronger discrimination. In addition, the adaptive metric method is used for feature sampling and update the parameters according to the characteristics of the features. Thus, the feature can be more typical and beneficial to improve the supervised ability of the optimization direction of the deep neural networks. Experimental results show that, compared with other deep neural networks, the relative equal error rate of the proposed method is reduced by up to 28%, and the performance of the speaker verification system is significantly improved.
In order to measure the relationship between features more accurately, an objective function representation method based on mutual information adaptive estimation is proposed for speaker verification systems. This objective function introduces an adaptive metric learning method, and the optimization objective is maximizing the intra-class similarity and minimizing the inter-class similarity. Meanwhile, the objective function can dynamically adjust the similarity according to the real distribution of deep features. Based on dynamically adjusting, the deep neural networks can be optimized towards the direction of stronger discrimination. In addition, the adaptive metric method is used for feature sampling and update the parameters according to the characteristics of the features. Thus, the feature can be more typical and beneficial to improve the supervised ability of the optimization direction of the deep neural networks. Experimental results show that, compared with other deep neural networks, the relative equal error rate of the proposed method is reduced by up to 28%, and the performance of the speaker verification system is significantly improved.
2023, 52(1): 132-139.
doi: 10.12178/1001-0548.2021257
Abstract:
In order to solve the problem of uncertainty in reliability allocation and prediction of industrial robots at the early stage of development in the absence of fault information, a fuzzy comprehensive evaluation method (FCEM) considering multiple influencing factors is proposed for reliability allocation and prediction of industrial robots on the basis of fuzzy mathematical theory. In order to improve the computational efficiency of the FCEM, the best worst method (BWM) is introduced to increase the consistency of the judgment results and reduce the possibility of subjective errors. First, the unit set of the industrial robot and the set of factors affecting its reliability allocation and prediction are determined according to its working principle and structural composition. Then, the BWM is applied to the comparison process of the fuzzy comprehensive evaluation method to determine the weights of factors affecting the reliability allocation and prediction of industrial robots. Finally, the reliability allocation weights and reliability prediction correction factors for each subsystem of the industrial robot are obtained by fuzzy comprehensive operation to allocate and predict its reliability.
In order to solve the problem of uncertainty in reliability allocation and prediction of industrial robots at the early stage of development in the absence of fault information, a fuzzy comprehensive evaluation method (FCEM) considering multiple influencing factors is proposed for reliability allocation and prediction of industrial robots on the basis of fuzzy mathematical theory. In order to improve the computational efficiency of the FCEM, the best worst method (BWM) is introduced to increase the consistency of the judgment results and reduce the possibility of subjective errors. First, the unit set of the industrial robot and the set of factors affecting its reliability allocation and prediction are determined according to its working principle and structural composition. Then, the BWM is applied to the comparison process of the fuzzy comprehensive evaluation method to determine the weights of factors affecting the reliability allocation and prediction of industrial robots. Finally, the reliability allocation weights and reliability prediction correction factors for each subsystem of the industrial robot are obtained by fuzzy comprehensive operation to allocate and predict its reliability.
2023, 52(1): 140-145.
doi: 10.12178/1001-0548.2022153
Abstract:
Heavy-duty CNC machine tools occupy an important position in the field of machining, and improving its reliability and machining accuracy is of great significance to the industrial development of China. Compared with conventional machine tools, heavy-duty CNC machine tools have the characteristics of complex structure, difficulty in fault tracing, few samples, and insufficient data, which make it difficult to conduct reliability research on them. Aiming at this problem, this paper uses the Weibull distribution of two parameters to establish the reliability model of the machine tool, introduces Bayesian theory to estimate its parameters, and calculates the parameter estimation results through the Markov Chain Monte Carlo method (MCMC). In order to improve the accuracy of parameter estimation, the traditional Bayesian method is improved, and the standard root mean square error value and confidence interval are used for evaluation and comparison. The results show that the improved Bayesian method has better parameter estimation accuracy and is more conducive to the establishment of machine tool reliability models.
Heavy-duty CNC machine tools occupy an important position in the field of machining, and improving its reliability and machining accuracy is of great significance to the industrial development of China. Compared with conventional machine tools, heavy-duty CNC machine tools have the characteristics of complex structure, difficulty in fault tracing, few samples, and insufficient data, which make it difficult to conduct reliability research on them. Aiming at this problem, this paper uses the Weibull distribution of two parameters to establish the reliability model of the machine tool, introduces Bayesian theory to estimate its parameters, and calculates the parameter estimation results through the Markov Chain Monte Carlo method (MCMC). In order to improve the accuracy of parameter estimation, the traditional Bayesian method is improved, and the standard root mean square error value and confidence interval are used for evaluation and comparison. The results show that the improved Bayesian method has better parameter estimation accuracy and is more conducive to the establishment of machine tool reliability models.
2023, 52(1): 146-153.
doi: 10.12178/1001-0548.2021284
Abstract:
Heterogeneous graph (HG) embedding method has been proposed as a new learning paradigm that embeds vertices into a low-dimensional dense vector space, by preserving Heterogeneous graph topology structure and vertex attributes information. In order to improve the quality of HG embedding and reduce distortion, a method for embedding HGs into hyperbolic space based on Lorentz model is proposed. This method employ the meta-path guided random walk to capture the structure and semantic relations between nodes. Specifically, the maximum likelihood estimate based on negative sampling is used as the objective function to achieve binary classification: making the target node more similar to its neighbor and farther away from non-neighbor nodes. Then, the Riemann gradient descent, which is different from the Euclidean space, is used to optimize the model parameters. Experiments on PubMed dataset demonstrate that our proposed model not only has superior performance on link prediction tasks than 4 baseline methods but also show its ability of capture graph’s hierarchy structure. Hyperbolic space provides a new idea for analyzing structure of heterogeneous graphs and can provide higher-quality embedding results for downstream tasks of heterogeneous graphs.
Heterogeneous graph (HG) embedding method has been proposed as a new learning paradigm that embeds vertices into a low-dimensional dense vector space, by preserving Heterogeneous graph topology structure and vertex attributes information. In order to improve the quality of HG embedding and reduce distortion, a method for embedding HGs into hyperbolic space based on Lorentz model is proposed. This method employ the meta-path guided random walk to capture the structure and semantic relations between nodes. Specifically, the maximum likelihood estimate based on negative sampling is used as the objective function to achieve binary classification: making the target node more similar to its neighbor and farther away from non-neighbor nodes. Then, the Riemann gradient descent, which is different from the Euclidean space, is used to optimize the model parameters. Experiments on PubMed dataset demonstrate that our proposed model not only has superior performance on link prediction tasks than 4 baseline methods but also show its ability of capture graph’s hierarchy structure. Hyperbolic space provides a new idea for analyzing structure of heterogeneous graphs and can provide higher-quality embedding results for downstream tasks of heterogeneous graphs.
2023, 52(1): 154-160.
doi: 10.12178/1001-0548.2021354
Abstract:
This paper tries to introduce temporal network into the study of disinformation propagation, and proposes a method to explore the propagation mechanism of disinformation by carving the propagation network through Temporal Motif, which integrates the structural characteristics of the propagation network and the temporal properties of information, and also uses several real data sets to test the generalizability of the method in disinformation detection. Based on the results of empirical data, it is shown that the change patterns of true and false information are different at different time scales Temporal Motif, and false information spreads faster and deeper than true information at large time scales, and false information can be detected more accurately using the method based on the temporal modal degree. This study reveals the multi-timescale propagation mechanism of false information, which can be used to prevent the propagation of false information.
This paper tries to introduce temporal network into the study of disinformation propagation, and proposes a method to explore the propagation mechanism of disinformation by carving the propagation network through Temporal Motif, which integrates the structural characteristics of the propagation network and the temporal properties of information, and also uses several real data sets to test the generalizability of the method in disinformation detection. Based on the results of empirical data, it is shown that the change patterns of true and false information are different at different time scales Temporal Motif, and false information spreads faster and deeper than true information at large time scales, and false information can be detected more accurately using the method based on the temporal modal degree. This study reveals the multi-timescale propagation mechanism of false information, which can be used to prevent the propagation of false information.
