Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00105
Yucheng He, Xinyan Li, Rui Li
For large area robot skin design, the distribution of rigid components and wires in traditional array sensors leads to the decrease of the flexibility and extensibility. The flexible sensors based on non-invasive electrical impedance tomography (EIT) can avoid these shortcomings. However, the number, position and driving pattern of the central electrode will have a great impact on the performance of the sensor. Multi-walled carbon nanotubes (MWCNTs) are used as filler materials to prepare flexible materials. Different numbers of central electrodes are introduced into the traditional 16-electrode flexible sensor. The corresponding driving pattern is designed and the position error is taken as the evaluation index to carry out comparative experiments. The simulation and experimental results show that the 18-electrode flexible sensor with two central electrodes has the best detection performance, and the detection position error can be reduced by 57.6% at the optimal position of the central electrodes at 0.24 from the center of the circle. The optimal design of the central electrode can effectively improve the performance of the flexible sensor, which has a certain guiding significance for the design of large area robot skin.
{"title":"Design of an EIT-based flexible tactile sensor with center electrodes","authors":"Yucheng He, Xinyan Li, Rui Li","doi":"10.1109/ICCEA53728.2021.00105","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00105","url":null,"abstract":"For large area robot skin design, the distribution of rigid components and wires in traditional array sensors leads to the decrease of the flexibility and extensibility. The flexible sensors based on non-invasive electrical impedance tomography (EIT) can avoid these shortcomings. However, the number, position and driving pattern of the central electrode will have a great impact on the performance of the sensor. Multi-walled carbon nanotubes (MWCNTs) are used as filler materials to prepare flexible materials. Different numbers of central electrodes are introduced into the traditional 16-electrode flexible sensor. The corresponding driving pattern is designed and the position error is taken as the evaluation index to carry out comparative experiments. The simulation and experimental results show that the 18-electrode flexible sensor with two central electrodes has the best detection performance, and the detection position error can be reduced by 57.6% at the optimal position of the central electrodes at 0.24 from the center of the circle. The optimal design of the central electrode can effectively improve the performance of the flexible sensor, which has a certain guiding significance for the design of large area robot skin.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123911172","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00024
Haiyuan Guo, Chengying Chi, Xuegang Zhan
For the Chinese text sentiment classification task, the preprocessing based on deep learning models cannot retain the information and polysemy of the word in the sentence well. So this paper adopts the newly developed ERNIE [1–2] (Knowledge Enhanced Semantic Representation) pre-training model from Baidu, which is based on word feature input modeling, not only enhances the semantic representation of the word, but also preserves the contextual information of the word and the polysemy of the word. After pre-training by ERNIE model, the output word vector is used as the input of BiLSTM (bidirectional long and short-term memory network) model for training and obtaining sentiment classification results. The accuracy rate of Ernie bilstm model is 92.35% after verification on nlpcc2014 microblog sentiment analysis sample data set, which proves that the model has good performance in Chinese text sentiment classification task.
{"title":"ERNIE-BiLSTM Based Chinese Text Sentiment Classification Method","authors":"Haiyuan Guo, Chengying Chi, Xuegang Zhan","doi":"10.1109/ICCEA53728.2021.00024","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00024","url":null,"abstract":"For the Chinese text sentiment classification task, the preprocessing based on deep learning models cannot retain the information and polysemy of the word in the sentence well. So this paper adopts the newly developed ERNIE [1–2] (Knowledge Enhanced Semantic Representation) pre-training model from Baidu, which is based on word feature input modeling, not only enhances the semantic representation of the word, but also preserves the contextual information of the word and the polysemy of the word. After pre-training by ERNIE model, the output word vector is used as the input of BiLSTM (bidirectional long and short-term memory network) model for training and obtaining sentiment classification results. The accuracy rate of Ernie bilstm model is 92.35% after verification on nlpcc2014 microblog sentiment analysis sample data set, which proves that the model has good performance in Chinese text sentiment classification task.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"60 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124030107","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00097
Hongli Chen, Di Chen, Luyao Wang
Brain tumors are usually diagnosed manually by the doctors from the Magnetic Resonance Images, which decreases the efficiency of the diagnosis process. Facing the situation that diagnosis of brain tumors from Magnetic Resonance Images needs effective methods to increase the speed and enhance the accuracy, we proposed algorithms using Convolutional Neural Network, the MobileNet, and AlexNet models to help classify the tumor while also developed an interface system to connect the algorithm directly to hospital system. We utilized grouped dataset and developed the algorithm to classify whether there is brain tumor occurred in the Magnetic Resonance images. The patients can employ the interface system developed through Tkinter by simply typing the information and automatically get the final results appears on the screen. From our result, compared with other models such as MobileNet and AlexNet, the proposed Convolutional Neural Network algorithm reaches the highest accuracy and lowest loss. Our interface system enables the patients of the hospital to directly and conveniently access the diagnosis of our algorithm.
{"title":"CNN-based MRI Brain Tumor Detection Application","authors":"Hongli Chen, Di Chen, Luyao Wang","doi":"10.1109/ICCEA53728.2021.00097","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00097","url":null,"abstract":"Brain tumors are usually diagnosed manually by the doctors from the Magnetic Resonance Images, which decreases the efficiency of the diagnosis process. Facing the situation that diagnosis of brain tumors from Magnetic Resonance Images needs effective methods to increase the speed and enhance the accuracy, we proposed algorithms using Convolutional Neural Network, the MobileNet, and AlexNet models to help classify the tumor while also developed an interface system to connect the algorithm directly to hospital system. We utilized grouped dataset and developed the algorithm to classify whether there is brain tumor occurred in the Magnetic Resonance images. The patients can employ the interface system developed through Tkinter by simply typing the information and automatically get the final results appears on the screen. From our result, compared with other models such as MobileNet and AlexNet, the proposed Convolutional Neural Network algorithm reaches the highest accuracy and lowest loss. Our interface system enables the patients of the hospital to directly and conveniently access the diagnosis of our algorithm.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"62 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121144719","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00019
Yu Chen, Junjie Liu, Yuanzhuo Li
Few-shot image classification aims at recognizing image categories with only a few labeled examples. The metric-based model is commonly used in few-shot learning. But restricted by needing a large amount of memory in training process, existing highly efficient backbone network cannot be used and light weight Residual Network performs not well. So we construct a new light weight network based on the idea of multi-scale analyzation as the feature extractor. We test it on several public datasets and it can run effectively under existing public equipment and provides better efficiency compared with ResNet with the same number of layers.
{"title":"Few-shot Image Classification based on LMRNet","authors":"Yu Chen, Junjie Liu, Yuanzhuo Li","doi":"10.1109/ICCEA53728.2021.00019","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00019","url":null,"abstract":"Few-shot image classification aims at recognizing image categories with only a few labeled examples. The metric-based model is commonly used in few-shot learning. But restricted by needing a large amount of memory in training process, existing highly efficient backbone network cannot be used and light weight Residual Network performs not well. So we construct a new light weight network based on the idea of multi-scale analyzation as the feature extractor. We test it on several public datasets and it can run effectively under existing public equipment and provides better efficiency compared with ResNet with the same number of layers.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"94 2","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114111619","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00046
Donghua Zheng, Weirong Xiu, Lizhu Ye
In order to improve the data acquisition capability of wireless sensor networks, a data compression sampling technology based on unbalanced data collaborative filtering is proposed. Establishing a data compression sampling state feature analysis model, designing a linear kernel function, a probability density feature kernel function and a Gaussian kernel function for wireless sensor network communication transmission data compression sampling, realizing wireless sensor network data compression and feature separation by an unbalanced data collaborative filtering method, constructing a boundary solution vector function for data compression sampling by adopting a support vector machine model, and realizing classification processing after data feature compression by adopting a fuzzy c-means clustering analysis method. Combined with threshold judgment method, the filtering analysis and subspace noise reduction of wireless sensor network data compression are realized, and the unbalanced data collaborative filtering detection model is constructed. According to the data feature detection results, the wireless sensor network data compression sampling is realized. The simulation results show that the feature clustering of wireless sensor network data compression sampling is better and the data detection accuracy is higher.
{"title":"Wireless sensor network data compression sampling technology based on unbalanced data collaborative filtering","authors":"Donghua Zheng, Weirong Xiu, Lizhu Ye","doi":"10.1109/ICCEA53728.2021.00046","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00046","url":null,"abstract":"In order to improve the data acquisition capability of wireless sensor networks, a data compression sampling technology based on unbalanced data collaborative filtering is proposed. Establishing a data compression sampling state feature analysis model, designing a linear kernel function, a probability density feature kernel function and a Gaussian kernel function for wireless sensor network communication transmission data compression sampling, realizing wireless sensor network data compression and feature separation by an unbalanced data collaborative filtering method, constructing a boundary solution vector function for data compression sampling by adopting a support vector machine model, and realizing classification processing after data feature compression by adopting a fuzzy c-means clustering analysis method. Combined with threshold judgment method, the filtering analysis and subspace noise reduction of wireless sensor network data compression are realized, and the unbalanced data collaborative filtering detection model is constructed. According to the data feature detection results, the wireless sensor network data compression sampling is realized. The simulation results show that the feature clustering of wireless sensor network data compression sampling is better and the data detection accuracy is higher.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130168170","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00054
Shao-yu Lv, Mu-yuan Jiang, Yuzhuo Chen, Yunsheng Wang
Picking the best velocity from the velocity spectrum is one of the keys to process seismic data. Aiming at the problems of lower efficiency of manual picking and poor precision of general automatic picking, a solution space constraint method to pick the best velocity automatically was proposed. Firstly, according to the signal similarity coefficient criterion, the original velocity solution space P is constrained to obtain the space P’; Secondly, using the signal in-phase criterion perform the peak match based on kd-Tree’s nearest neighbor search, the space P’ is changed into the space P” by the matching results; Finally, in accordance with the objective function, the automatic picking of the optimal velocity is achieved by the improved particle swarm model in constraint space P”. Experimental results show that the calculation speed of this algorithm is faster, and the error between the automatic picking result and the real reflected signal value is smaller, which meets the needs of actual engineering.
{"title":"Using the Solution Space Constraint to Pick the Best Velocity Automatically","authors":"Shao-yu Lv, Mu-yuan Jiang, Yuzhuo Chen, Yunsheng Wang","doi":"10.1109/ICCEA53728.2021.00054","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00054","url":null,"abstract":"Picking the best velocity from the velocity spectrum is one of the keys to process seismic data. Aiming at the problems of lower efficiency of manual picking and poor precision of general automatic picking, a solution space constraint method to pick the best velocity automatically was proposed. Firstly, according to the signal similarity coefficient criterion, the original velocity solution space P is constrained to obtain the space P’; Secondly, using the signal in-phase criterion perform the peak match based on kd-Tree’s nearest neighbor search, the space P’ is changed into the space P” by the matching results; Finally, in accordance with the objective function, the automatic picking of the optimal velocity is achieved by the improved particle swarm model in constraint space P”. Experimental results show that the calculation speed of this algorithm is faster, and the error between the automatic picking result and the real reflected signal value is smaller, which meets the needs of actual engineering.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"59 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133962153","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00033
W. Yuan, S. Jianwei
In order to cater to the advanced user positioning function of the Beidou satellite navigation system, the simulated domestic positioning point data is used to realize the visualization of user distribution in the form of a heat map through the Webside 3D visualization technology. In the case of a large amount of rendering data, by optimizing the kernel density estimation algorithm and optimizing the calculation method of the data points in the raster, the effect of real-time rendering of the heat map in the optimal display form is realized with the change of the viewpoint position, and the rendering Performance has been significantly improved.
{"title":"Research on Visualization Method of Large-scale User Location Distribution Based on CesiumJS","authors":"W. Yuan, S. Jianwei","doi":"10.1109/ICCEA53728.2021.00033","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00033","url":null,"abstract":"In order to cater to the advanced user positioning function of the Beidou satellite navigation system, the simulated domestic positioning point data is used to realize the visualization of user distribution in the form of a heat map through the Webside 3D visualization technology. In the case of a large amount of rendering data, by optimizing the kernel density estimation algorithm and optimizing the calculation method of the data points in the raster, the effect of real-time rendering of the heat map in the optimal display form is realized with the change of the viewpoint position, and the rendering Performance has been significantly improved.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"164 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134282882","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00017
Limei Wu, Heng Yue, Honghong Hu
Aiming at the problem of measuring the degree of supply chain resource coordination, based on the synergetics theory, an artificial intelligence based supply chain multi resource inventory automatic coordination model is constructed. By constructing five index systems of supply chain logistics, capital flow, information flow, market flow and management flow of artificial intelligence, this paper makes an empirical study on the application of the model. The results show that the measurement model of the degree of supply chain resource coordination can reflect the degree and trend of the coordinated development of the supply chain system, and the specific reasons for the low degree of supply chain coordination can be found through the analysis, which is conducive to the targeted and fundamental improvement of enterprises. The supply chain multi resource inventory automatic coordination model based on artificial intelligence not only makes up for the lack of qualitative research in the existing coordination degree model, but also has great significance for guiding the actual operation of the supply chain.
{"title":"Multi resource inventory automatic coordination model of Supply Chain Based on Artificial Intelligence","authors":"Limei Wu, Heng Yue, Honghong Hu","doi":"10.1109/ICCEA53728.2021.00017","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00017","url":null,"abstract":"Aiming at the problem of measuring the degree of supply chain resource coordination, based on the synergetics theory, an artificial intelligence based supply chain multi resource inventory automatic coordination model is constructed. By constructing five index systems of supply chain logistics, capital flow, information flow, market flow and management flow of artificial intelligence, this paper makes an empirical study on the application of the model. The results show that the measurement model of the degree of supply chain resource coordination can reflect the degree and trend of the coordinated development of the supply chain system, and the specific reasons for the low degree of supply chain coordination can be found through the analysis, which is conducive to the targeted and fundamental improvement of enterprises. The supply chain multi resource inventory automatic coordination model based on artificial intelligence not only makes up for the lack of qualitative research in the existing coordination degree model, but also has great significance for guiding the actual operation of the supply chain.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132263986","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00038
Yu Wang, Mingjie Song, Xinyu Tian
General breast cancer detection contains two steps, the breast abnormality classification, and the diagnostic classification. The determination of the abnormality contributes further to the following steps, and computational technologies can aid in the process. A lot of machine learning methods have been applied to automate the detection. However, most of them focus on the diagnostic classification and the breast abnormality classification only attracts little attention. The insufficient size of public mammogram datasets also limits the performance of many machine learning algorithms. Considering the importance of breast abnormality classification and the shortage of public large-scale medical datasets, we proposed a meta-learning-based breast abnormality classification method. Our model referred to the latest work of meta-learning-based image classifier and modified it. Specifically, we applied the idea of meta-learning to retrain a pretrained embedding neural network in order to adapt its feature extraction ability to the CBIS-DDSM dataset [1]. The dataset contains two types of abnormal breast mammograms, mass and calcification, and each type is made of two categories of medical images, full mammograms, and ROI [2]. The application of the data augmentation techniques and the idea of meta-learning helped to deal with the insufficient training sample problem and showed a final accuracy of 76%, which beat the 71% accuracy reached by a neural network baseline model.
{"title":"Meta-learning Based Breast Abnormality Classification on Screening Mammograms","authors":"Yu Wang, Mingjie Song, Xinyu Tian","doi":"10.1109/ICCEA53728.2021.00038","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00038","url":null,"abstract":"General breast cancer detection contains two steps, the breast abnormality classification, and the diagnostic classification. The determination of the abnormality contributes further to the following steps, and computational technologies can aid in the process. A lot of machine learning methods have been applied to automate the detection. However, most of them focus on the diagnostic classification and the breast abnormality classification only attracts little attention. The insufficient size of public mammogram datasets also limits the performance of many machine learning algorithms. Considering the importance of breast abnormality classification and the shortage of public large-scale medical datasets, we proposed a meta-learning-based breast abnormality classification method. Our model referred to the latest work of meta-learning-based image classifier and modified it. Specifically, we applied the idea of meta-learning to retrain a pretrained embedding neural network in order to adapt its feature extraction ability to the CBIS-DDSM dataset [1]. The dataset contains two types of abnormal breast mammograms, mass and calcification, and each type is made of two categories of medical images, full mammograms, and ROI [2]. The application of the data augmentation techniques and the idea of meta-learning helped to deal with the insufficient training sample problem and showed a final accuracy of 76%, which beat the 71% accuracy reached by a neural network baseline model.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132441273","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2021-06-01DOI: 10.1109/ICCEA53728.2021.00071
Jinping Su, L. Jing
Adversarial examples expose the vulnerability of deep neural networks that perform well in various fields. However, adversarial perturbations crafted by the existing attack methods are often aimed at the whole image. They are usually random, and the human eye can even easily perceive some of them. This paper proposes an adaptive method to compress the adversarial perturbation. Under the premise of ensuring the success of attacks, generating perturbations as small as possible to change the decision of classifiers. First, the authors find the minimum point of loss function by the optimization method, to expand the spanning space of adversarial examples. Calculating and selecting the smaller perturbation between this point and the original input. Then, in order to retain the useful perturbation and remove redundancy, the authors look for important regions in the input data that determine the network predict results, and construct an importance mask for the smaller perturbation of the previous stage. Extensive experiments on the ImageNet dataset and multiple network classifiers show that our method is effective. Compared with advanced attack methods, the $mathbf{L}_{2}$ distance of adversarial perturbation obtained by our method is smaller and more practical, and the generated adversarial examples have strong transferability.
{"title":"Robust Attack with Adaptive Compress Adversarial Perturbations","authors":"Jinping Su, L. Jing","doi":"10.1109/ICCEA53728.2021.00071","DOIUrl":"https://doi.org/10.1109/ICCEA53728.2021.00071","url":null,"abstract":"Adversarial examples expose the vulnerability of deep neural networks that perform well in various fields. However, adversarial perturbations crafted by the existing attack methods are often aimed at the whole image. They are usually random, and the human eye can even easily perceive some of them. This paper proposes an adaptive method to compress the adversarial perturbation. Under the premise of ensuring the success of attacks, generating perturbations as small as possible to change the decision of classifiers. First, the authors find the minimum point of loss function by the optimization method, to expand the spanning space of adversarial examples. Calculating and selecting the smaller perturbation between this point and the original input. Then, in order to retain the useful perturbation and remove redundancy, the authors look for important regions in the input data that determine the network predict results, and construct an importance mask for the smaller perturbation of the previous stage. Extensive experiments on the ImageNet dataset and multiple network classifiers show that our method is effective. Compared with advanced attack methods, the $mathbf{L}_{2}$ distance of adversarial perturbation obtained by our method is smaller and more practical, and the generated adversarial examples have strong transferability.","PeriodicalId":325790,"journal":{"name":"2021 International Conference on Computer Engineering and Application (ICCEA)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2021-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115695852","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：