The current adversarial attacks against deep learning models have achieved incredible success in the white-box scenario.However,they often exhibit weak transferability in the black-box scenario,especially when attacki...The current adversarial attacks against deep learning models have achieved incredible success in the white-box scenario.However,they often exhibit weak transferability in the black-box scenario,especially when attacking those with defense mechanisms.In this work,we propose a new transfer-based blackbox attack called the channel decomposition attack method(CDAM).It can attack multiple black-box models by enhancing the transferability of the adversarial examples.On the one hand,it tunes the gradient and stabilizes the update direction by decomposing the channels of the input example and calculating the aggregate gradient.On the other hand,it helps to escape from local optima by initializing the data point with random noise.Besides,it could combine with other transfer-based attacks flexibly.Extensive experiments on the standard ImageNet dataset show that our method could significantly improve the transferability of adversarial attacks.Compared with the state-of-the-art method,our approach improves the average success rate from 88.2%to 96.6%when attacking three adversarially trained black-box models,demonstrating the remaining shortcomings of existing deep learning models.展开更多
In the sorting system of the production line,the object movement,fixed angle of view,light intensity and other reasons lead to obscure blurred images.It results in bar code recognition rate being low and real time bei...In the sorting system of the production line,the object movement,fixed angle of view,light intensity and other reasons lead to obscure blurred images.It results in bar code recognition rate being low and real time being poor.Aiming at the above problems,a progressive bar code compressed recognition algorithm is proposed.First,assuming that the source image is not tilted,use the direct recognition method to quickly identify the compressed source image.Failure indicates that the compression ratio is improper or the image is skewed.Then,the source image is enhanced to identify the source image directly.Finally,the inclination of the compressed image is detected by the barcode region recognition method and the source image is corrected to locate the barcode information in the barcode region recognition image.The results of multitype image experiments show that the proposed method is improved by 5+times computational efficiency compared with the former methods,and can recognize fuzzy images better.展开更多
Nowadays,Internet has become an indispensable part of daily life and is used in many fields.Due to the large amount of Internet traffic,computers are subject to various security threats,which may cause serious economi...Nowadays,Internet has become an indispensable part of daily life and is used in many fields.Due to the large amount of Internet traffic,computers are subject to various security threats,which may cause serious economic losses and even endanger national security.It is hoped that an effective security method can systematically classify intrusion data in order to avoid leakage of important data or misuse of data.As machine learning technology matures,deep learning is widely used in various industries.Combining deep learning with network security and intrusion detection is the current trend.In this paper,the problem of data classification in intrusion detection system is studied.We propose an intrusion detection model based on stack bidirectional long short-term memory(LSTM),introduce stack bidirectional LSTM into the field of intrusion detection and apply it to the intrusion detection.In order to determine the appropriate parameters and structure of stack bidirectional LSTM network,we have carried out experiments on various network structures and parameters and analyzed the experimental results.The classic KDD Cup’1999 dataset was selected for experiments so that we can obtain convincing and comparable results.Experimental results derived from the KDD Cup’1999 dataset show that the network with three hidden layers containing 80 LSTM cells is superior to other algorithms in computational cost and detection performance due to stack bidirectional LSTM model’s ability to review time and correlate with connected records continuously.The experiment shows the effectiveness of stack bidirectional LSTM network in intrusion detection.展开更多
Deep learning techniques have outstanding performance in feature extraction and modelfitting.In thefield of aero-engine fault diagnosis,the intro-duction of deep learning technology is of great significance.The aero-engi...Deep learning techniques have outstanding performance in feature extraction and modelfitting.In thefield of aero-engine fault diagnosis,the intro-duction of deep learning technology is of great significance.The aero-engine is the heart of the aircraft,and its stable operation is the primary guarantee of the aircraft.In order to ensure the normal operation of the aircraft,it is necessary to study and diagnose the faults of the aero-engine.Among the many engine fail-ures,the one that occurs more frequently and is more hazardous is the wheeze,which often poses a great threat toflight safety.On the basis of analyzing the mechanism of aero-engine surge,an aero-engine surge fault diagnosis method based on deep learning technology is proposed.In this paper,key sensor data are obtained by analyzing different engine sensor data.An aero-engine surge data-set acquisition algorithm(ASDA)is proposed to sample the fault and normal points to generate the training set,validation set and test set.Based on neural net-work models such as one-dimensional convolutional neural network(1D-CNN),convolutional neural network(RNN),and long-short memory neural network(LSTM),different neural network optimization algorithms are selected to achieve fault diagnosis and classification.The experimental results show that the deep learning technique has good effect in aero-engine surge fault diagnosis.The aero-engine surge fault diagnosis network(ASFDN)proposed in this paper achieves better results.Through training,the network achieves more than 99%classification accuracy for the test set.展开更多
Given the glut of information on the web,it is crucially important to have a system,which will parse the information appropriately and recommend users with relevant information,this class of systems is known as Recomm...Given the glut of information on the web,it is crucially important to have a system,which will parse the information appropriately and recommend users with relevant information,this class of systems is known as Recommendation Systems(RS)-it is one of the most extensively used systems on the web today.Recently,Deep Learning(DL)models are being used to generate recommendations,as it has shown state-of-the-art(SoTA)results in the field of Speech Recognition and Computer Vision in the last decade.However,the RS is a much harder problem,as the central variable in the recommendation system’s environment is the chaotic nature of the human’s purchasing/consuming behaviors and their interest.These user-item interactions cannot be fully represented in the Euclidean-Space,as it will trivialize the interaction and undermine the implicit interactions patterns.So to preserve the implicit as well as explicit interactions of user and items,we propose a new graph based recommendation framework.The fundamental idea behind this framework is not only to generate the recommendations in the unsupervised fashion but to learn the dynamics of the graph and predict the short and long term interest of the users.In this paper,we propose the first step,a heuristic multi-layer high-dimensional graph which preserves the implicit and explicit interactions between users and items using SoTA Deep Learning models such as AutoEncoders.To generate recommendation from this generated graph a new class of neural network architecture-Graph Neural Network-can be used.展开更多
Quantum correlation plays a critical role in the maintenance of quantum information processing and nanometer device design.In the past two decades,several quantitative methods had been proposed to study the quantum co...Quantum correlation plays a critical role in the maintenance of quantum information processing and nanometer device design.In the past two decades,several quantitative methods had been proposed to study the quantum correlation of certain open quantum systems,including the geometry and entropy style discord methods.However,there are differences among these quantification methods,which promote a deep understanding of the quantum correlation.In this paper,a novel time-dependent three environmental open system model is established to study the quantum correlation.This system model interacts with two independent spin-environments(two spin-environments are connected to the other spin-environment)respectively.We have calculated and compared the changing properties of the quantum correlation under three kinds of geometry and two entropy discords,especially for the freezing phenomenon.At the same time,some original and novel changing behaviors of the quantum correlation under different timedependent parameters are studied,which is helpful to achieve the optimal revival of the quantum discord and the similar serrated form of the freezing phenomenon.Finally,it shows the controllability of the freezing correlation and the robustness of these methods by adjusting time-dependent parameters.This work provides a new way to control the quantum correlation and design nanospintronic devices.展开更多
There are two difficult in the existing image restoration methods.One is that the method is difficult to repair the image with a large damaged,the other is the result of image completion is not good and the speed is s...There are two difficult in the existing image restoration methods.One is that the method is difficult to repair the image with a large damaged,the other is the result of image completion is not good and the speed is slow.With the development and application of deep learning,the image repair algorithm based on generative adversarial networks can repair images by simulating the distribution of data.In the process of image completion,the first step is trained the generator to simulate data distribution and generate samples.Then a large number of falsified images are quickly generated using the generative adversarial network and search for the code of the closest damaged image.Finally,the generator generates missing content by using this code.On this basis,this paper combines the semantic loss function and the perceptual loss function.Experimental result show that the method successfully predicts the information of large areas missing in the image,and realizes the photorealism,producing clearer and more consistent results than previous methods.展开更多
Image conversion refers to converting an image from one style to another and ensuring that the content of the image remains unchanged.Using Generative Adversarial Networks(GAN)for image conversion can achieve good res...Image conversion refers to converting an image from one style to another and ensuring that the content of the image remains unchanged.Using Generative Adversarial Networks(GAN)for image conversion can achieve good results.However,if there are enough samples,any image in the target domain can be mapped to the same set of inputs.On this basis,the Cycle Consistency Generative Adversarial Network(CycleGAN)was developed.This article verifies and discusses the advantages and disadvantages of the CycleGAN model in image style conversion.CycleGAN uses two generator networks and two discriminator networks.The purpose is to learn the mapping relationship and inverse mapping relationship between the source domain and the target domain.It can reduce the mapping and improve the quality of the generated image.Through the idea of loop,the loss of information in image style conversion is reduced.When evaluating the results of the experiment,the degree of retention of the input image content will be judged.Through the experimental results,CycleGAN can understand the artist’s overall artistic style and successfully convert real landscape paintings.The advantage is that most of the content of the original picture can be retained,and only the texture line of the picture is changed to a level similar to the artist’s style.展开更多
Deep learning has made a great breakthrough in the field of speech and image recognition.Mature deep learning neural network has completely changed the field of natural language processing(NLP).Due to the enormous amo...Deep learning has made a great breakthrough in the field of speech and image recognition.Mature deep learning neural network has completely changed the field of natural language processing(NLP).Due to the enormous amount of data and opinions being produced,shared and transferred everyday across the Internet and other media,sentiment analysis has become one of the most active research fields in natural language processing.This paper introduces three deep learning networks applied in IMDB movie reviews sentiment analysis.Dataset was divided to 50%positive reviews and 50%negative reviews.Recurrent Neural Network(RNN)and Long Short-Term Memory(LSTM)neural networks are two main types,which are widely used in NLP tasks,while Convolutional Neural Networks(CNN)is often used in image recognition.The results have shown that,CNN network model can achieve good classification effect when applied to sentiment analysis of movie reviews.CNN have reported the accuracy of 88.22%,while RNN and LSTM have reported accuracy of 68.64%and 85.32%respectively.展开更多
基金This work was supported by Sichuan Science and Technology Program[No.2022YFG0315,2022YFG0174]Sichuan Gas Turbine Research Institute stability support project of China Aero Engine Group Co.,Ltd.[GJCZ-2019-71]Key project of Chengdu[No.2019-YF09-00044-CG].
文摘The current adversarial attacks against deep learning models have achieved incredible success in the white-box scenario.However,they often exhibit weak transferability in the black-box scenario,especially when attacking those with defense mechanisms.In this work,we propose a new transfer-based blackbox attack called the channel decomposition attack method(CDAM).It can attack multiple black-box models by enhancing the transferability of the adversarial examples.On the one hand,it tunes the gradient and stabilizes the update direction by decomposing the channels of the input example and calculating the aggregate gradient.On the other hand,it helps to escape from local optima by initializing the data point with random noise.Besides,it could combine with other transfer-based attacks flexibly.Extensive experiments on the standard ImageNet dataset show that our method could significantly improve the transferability of adversarial attacks.Compared with the state-of-the-art method,our approach improves the average success rate from 88.2%to 96.6%when attacking three adversarially trained black-box models,demonstrating the remaining shortcomings of existing deep learning models.
基金This work was supported by Scientific Research Starting Project of SWPU[Zheng,D.,No.0202002131604]Major Science and Technology Project of Sichuan Province[Zheng,D.,No.8ZDZX0143]+1 种基金Ministry of Education Collaborative Education Project of China[Zheng,D.,No.952]Fundamental Research Project[Zheng,D.,Nos.549,550].
文摘In the sorting system of the production line,the object movement,fixed angle of view,light intensity and other reasons lead to obscure blurred images.It results in bar code recognition rate being low and real time being poor.Aiming at the above problems,a progressive bar code compressed recognition algorithm is proposed.First,assuming that the source image is not tilted,use the direct recognition method to quickly identify the compressed source image.Failure indicates that the compression ratio is improper or the image is skewed.Then,the source image is enhanced to identify the source image directly.Finally,the inclination of the compressed image is detected by the barcode region recognition method and the source image is corrected to locate the barcode information in the barcode region recognition image.The results of multitype image experiments show that the proposed method is improved by 5+times computational efficiency compared with the former methods,and can recognize fuzzy images better.
基金This work was supported by Scientific Research Starting Project of SWPU[Zheng,D.,No.0202002131604]Major Science and Technology Project of Sichuan Province[Zheng,D.,No.8ZDZX0143]+1 种基金Ministry of Education Collaborative Education Project of China[Zheng,D.,No.952]Fundamental Research Project[Zheng,D.,Nos.549,550].
文摘Nowadays,Internet has become an indispensable part of daily life and is used in many fields.Due to the large amount of Internet traffic,computers are subject to various security threats,which may cause serious economic losses and even endanger national security.It is hoped that an effective security method can systematically classify intrusion data in order to avoid leakage of important data or misuse of data.As machine learning technology matures,deep learning is widely used in various industries.Combining deep learning with network security and intrusion detection is the current trend.In this paper,the problem of data classification in intrusion detection system is studied.We propose an intrusion detection model based on stack bidirectional long short-term memory(LSTM),introduce stack bidirectional LSTM into the field of intrusion detection and apply it to the intrusion detection.In order to determine the appropriate parameters and structure of stack bidirectional LSTM network,we have carried out experiments on various network structures and parameters and analyzed the experimental results.The classic KDD Cup’1999 dataset was selected for experiments so that we can obtain convincing and comparable results.Experimental results derived from the KDD Cup’1999 dataset show that the network with three hidden layers containing 80 LSTM cells is superior to other algorithms in computational cost and detection performance due to stack bidirectional LSTM model’s ability to review time and correlate with connected records continuously.The experiment shows the effectiveness of stack bidirectional LSTM network in intrusion detection.
基金supported by Scientific Research Starting Project of SWPU[No.0202002131604]Major Science and Technology Project of Sichuan Province[No.8ZDZX0143,2019YFG0424]+2 种基金Ministry of Education Collaborative Education Project of China[No.952]Fundamental Research Project[Nos.549,550]Development of Aero-engine Test and training platform based on Simulation Technology[18ZA0030].
文摘Deep learning techniques have outstanding performance in feature extraction and modelfitting.In thefield of aero-engine fault diagnosis,the intro-duction of deep learning technology is of great significance.The aero-engine is the heart of the aircraft,and its stable operation is the primary guarantee of the aircraft.In order to ensure the normal operation of the aircraft,it is necessary to study and diagnose the faults of the aero-engine.Among the many engine fail-ures,the one that occurs more frequently and is more hazardous is the wheeze,which often poses a great threat toflight safety.On the basis of analyzing the mechanism of aero-engine surge,an aero-engine surge fault diagnosis method based on deep learning technology is proposed.In this paper,key sensor data are obtained by analyzing different engine sensor data.An aero-engine surge data-set acquisition algorithm(ASDA)is proposed to sample the fault and normal points to generate the training set,validation set and test set.Based on neural net-work models such as one-dimensional convolutional neural network(1D-CNN),convolutional neural network(RNN),and long-short memory neural network(LSTM),different neural network optimization algorithms are selected to achieve fault diagnosis and classification.The experimental results show that the deep learning technique has good effect in aero-engine surge fault diagnosis.The aero-engine surge fault diagnosis network(ASFDN)proposed in this paper achieves better results.Through training,the network achieves more than 99%classification accuracy for the test set.
基金This work is supported by The National Natural Science Foundation of China[61502082].
文摘Given the glut of information on the web,it is crucially important to have a system,which will parse the information appropriately and recommend users with relevant information,this class of systems is known as Recommendation Systems(RS)-it is one of the most extensively used systems on the web today.Recently,Deep Learning(DL)models are being used to generate recommendations,as it has shown state-of-the-art(SoTA)results in the field of Speech Recognition and Computer Vision in the last decade.However,the RS is a much harder problem,as the central variable in the recommendation system’s environment is the chaotic nature of the human’s purchasing/consuming behaviors and their interest.These user-item interactions cannot be fully represented in the Euclidean-Space,as it will trivialize the interaction and undermine the implicit interactions patterns.So to preserve the implicit as well as explicit interactions of user and items,we propose a new graph based recommendation framework.The fundamental idea behind this framework is not only to generate the recommendations in the unsupervised fashion but to learn the dynamics of the graph and predict the short and long term interest of the users.In this paper,we propose the first step,a heuristic multi-layer high-dimensional graph which preserves the implicit and explicit interactions between users and items using SoTA Deep Learning models such as AutoEncoders.To generate recommendation from this generated graph a new class of neural network architecture-Graph Neural Network-can be used.
基金Scientific Research Starting Project of SWPU[Zheng,D.,No.0202002131604]Major Science and Technology Project of Sichuan Province[Zheng,D.,No.8ZDZX0143]+1 种基金Ministry of Education Collaborative Education Project of China[Zheng,D.,No.952]Fundamental Research Project[Zheng,D.,Nos.549,550].
文摘Quantum correlation plays a critical role in the maintenance of quantum information processing and nanometer device design.In the past two decades,several quantitative methods had been proposed to study the quantum correlation of certain open quantum systems,including the geometry and entropy style discord methods.However,there are differences among these quantification methods,which promote a deep understanding of the quantum correlation.In this paper,a novel time-dependent three environmental open system model is established to study the quantum correlation.This system model interacts with two independent spin-environments(two spin-environments are connected to the other spin-environment)respectively.We have calculated and compared the changing properties of the quantum correlation under three kinds of geometry and two entropy discords,especially for the freezing phenomenon.At the same time,some original and novel changing behaviors of the quantum correlation under different timedependent parameters are studied,which is helpful to achieve the optimal revival of the quantum discord and the similar serrated form of the freezing phenomenon.Finally,it shows the controllability of the freezing correlation and the robustness of these methods by adjusting time-dependent parameters.This work provides a new way to control the quantum correlation and design nanospintronic devices.
基金supported by Scientific Research Starting Project of SWPU(No.0202002131604)Major Science and Technology Project of Sichuan Province(No.8ZDZX0143)+1 种基金Ministry of Education Collaborative Education Project of China(No.952)Fundamental Research Project(Nos.549,550).
文摘There are two difficult in the existing image restoration methods.One is that the method is difficult to repair the image with a large damaged,the other is the result of image completion is not good and the speed is slow.With the development and application of deep learning,the image repair algorithm based on generative adversarial networks can repair images by simulating the distribution of data.In the process of image completion,the first step is trained the generator to simulate data distribution and generate samples.Then a large number of falsified images are quickly generated using the generative adversarial network and search for the code of the closest damaged image.Finally,the generator generates missing content by using this code.On this basis,this paper combines the semantic loss function and the perceptual loss function.Experimental result show that the method successfully predicts the information of large areas missing in the image,and realizes the photorealism,producing clearer and more consistent results than previous methods.
基金supported by Scientific Research Starting Project of SWPU[No.0202002131604]Major Science and Technology Project of Sichuan Province[Nos.8ZDZX0143,2019YFG0424]+1 种基金Ministry of Education Collaborative Education Project of China[No.952]Fundamental Research Project[Nos.549,550].
文摘Image conversion refers to converting an image from one style to another and ensuring that the content of the image remains unchanged.Using Generative Adversarial Networks(GAN)for image conversion can achieve good results.However,if there are enough samples,any image in the target domain can be mapped to the same set of inputs.On this basis,the Cycle Consistency Generative Adversarial Network(CycleGAN)was developed.This article verifies and discusses the advantages and disadvantages of the CycleGAN model in image style conversion.CycleGAN uses two generator networks and two discriminator networks.The purpose is to learn the mapping relationship and inverse mapping relationship between the source domain and the target domain.It can reduce the mapping and improve the quality of the generated image.Through the idea of loop,the loss of information in image style conversion is reduced.When evaluating the results of the experiment,the degree of retention of the input image content will be judged.Through the experimental results,CycleGAN can understand the artist’s overall artistic style and successfully convert real landscape paintings.The advantage is that most of the content of the original picture can be retained,and only the texture line of the picture is changed to a level similar to the artist’s style.
基金This work was supported by Scientific Research Starting Project of SWPU[No.0202002131604]Major Science and Technology Project of Sichuan Province[No.8ZDZX0143]Ministry of education integration project[No.952],and The Project[Nos.549,550].
文摘Deep learning has made a great breakthrough in the field of speech and image recognition.Mature deep learning neural network has completely changed the field of natural language processing(NLP).Due to the enormous amount of data and opinions being produced,shared and transferred everyday across the Internet and other media,sentiment analysis has become one of the most active research fields in natural language processing.This paper introduces three deep learning networks applied in IMDB movie reviews sentiment analysis.Dataset was divided to 50%positive reviews and 50%negative reviews.Recurrent Neural Network(RNN)and Long Short-Term Memory(LSTM)neural networks are two main types,which are widely used in NLP tasks,while Convolutional Neural Networks(CNN)is often used in image recognition.The results have shown that,CNN network model can achieve good classification effect when applied to sentiment analysis of movie reviews.CNN have reported the accuracy of 88.22%,while RNN and LSTM have reported accuracy of 68.64%and 85.32%respectively.
