2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)最新文献
Pub Date : 2017-06-01DOI: 10.1109/SNPD.2017.8022746
Taisuke Ono, H. M. Waidyasooriya, M. Hariyama, Tsukasa Ishigaki
Latent Dirichlet allocation (LDA) based topic inference is a data classification method, that is used efficiently for extremely large data sets. However, the processing time is very large due to the serial computational behavior of the Markov Chain Monte Carlo method used for the topic inference. We propose a pipelined hardware architecture and memory allocation scheme to accelerate LDA using parallel processing. The proposed architecture is implemented on a reconfigurable hardware called FPGA (field programmable gate array), using OpenCL design environment. According to the experimental results, we achieved maximum speed-up of 2.38 times, while maintaining the same quality compared to the conventional CPU-based implementation.
{"title":"Architecture of an FPGA accelerator for LDA-based inference","authors":"Taisuke Ono, H. M. Waidyasooriya, M. Hariyama, Tsukasa Ishigaki","doi":"10.1109/SNPD.2017.8022746","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022746","url":null,"abstract":"Latent Dirichlet allocation (LDA) based topic inference is a data classification method, that is used efficiently for extremely large data sets. However, the processing time is very large due to the serial computational behavior of the Markov Chain Monte Carlo method used for the topic inference. We propose a pipelined hardware architecture and memory allocation scheme to accelerate LDA using parallel processing. The proposed architecture is implemented on a reconfigurable hardware called FPGA (field programmable gate array), using OpenCL design environment. According to the experimental results, we achieved maximum speed-up of 2.38 times, while maintaining the same quality compared to the conventional CPU-based implementation.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114611648","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022727
M. Phankokkruad, Phichaya Jaturawat
Face recognition is a personal identification method using biometrics that is gaining the attention in this research field. The face recognition process can be done without the human and devices interaction, so it can be applied in several applications. In additions, the face recognition systems are typically implemented at different places in unconstrained environments. Hence, the study of the factors that impact the face recognition accuracy is an interesting and challenging topic. In the class attendance checking system using face recognition, there are variations of three factors that possibly affect the accuracy of the system; facial expressions, and face viewpoints. This study intends to compare facial recognition accuracy of three well-known algorithms namely Eigenfaces, Fisherfaces, and LBPH. The experiments conducted in the respects of the variation of facial expressions, and face viewpoints in the actual classroom. The results of the experiment demonstrated that LBPH is the most precise algorithm which achieves 81.67% of accuracy in still-image-based testing. The facial expression that has the most impact on accuracy is the grin, and face viewpoints that affect accuracy are looking down and tilting left, and right respectively. Therefore, LBPH is the most suitable algorithm to apply in a class attendance checking system after considering the accuracy.
{"title":"Influence of facial expression and viewpoint variations on face recognition accuracy by different face recognition algorithms","authors":"M. Phankokkruad, Phichaya Jaturawat","doi":"10.1109/SNPD.2017.8022727","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022727","url":null,"abstract":"Face recognition is a personal identification method using biometrics that is gaining the attention in this research field. The face recognition process can be done without the human and devices interaction, so it can be applied in several applications. In additions, the face recognition systems are typically implemented at different places in unconstrained environments. Hence, the study of the factors that impact the face recognition accuracy is an interesting and challenging topic. In the class attendance checking system using face recognition, there are variations of three factors that possibly affect the accuracy of the system; facial expressions, and face viewpoints. This study intends to compare facial recognition accuracy of three well-known algorithms namely Eigenfaces, Fisherfaces, and LBPH. The experiments conducted in the respects of the variation of facial expressions, and face viewpoints in the actual classroom. The results of the experiment demonstrated that LBPH is the most precise algorithm which achieves 81.67% of accuracy in still-image-based testing. The facial expression that has the most impact on accuracy is the grin, and face viewpoints that affect accuracy are looking down and tilting left, and right respectively. Therefore, LBPH is the most suitable algorithm to apply in a class attendance checking system after considering the accuracy.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123843442","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022714
L. Lazli, M. Boukadoum, O. Mohamed
The effect of partial volume related to anatomical MRI and functional images limit the diagnostic potential of brain imaging. To remedy for this problem, we propose a fuzzy-genetic brain segmentation scheme for the assessment of white matter, gray matter and cerebrospinal fluid volumes, from brain images of Alzheimer patients from a real database. This clustering process based on Possibilistic C-Means (PCM) algorithm, which allows modeling the degree of relationship between each voxels and a given tissue; and based on fuzzy genetic initialization for the centers of clusters by a Fuzzy C-Means (FCM) algorithm, and for which the result is optimized by genetic process. The visual results show a concordance between the ground truth segmentation and the hybrid algorithm results, which allows efficient tissue classification. The superiority was also proved with the quantitative results of the proposed method in comparison with the both conventional FCM and PCM algorithms.
{"title":"Hybrid possibilistic-genetic technique for assessment of brain tissues volume: Case study for Alzheimer patients images clustering","authors":"L. Lazli, M. Boukadoum, O. Mohamed","doi":"10.1109/SNPD.2017.8022714","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022714","url":null,"abstract":"The effect of partial volume related to anatomical MRI and functional images limit the diagnostic potential of brain imaging. To remedy for this problem, we propose a fuzzy-genetic brain segmentation scheme for the assessment of white matter, gray matter and cerebrospinal fluid volumes, from brain images of Alzheimer patients from a real database. This clustering process based on Possibilistic C-Means (PCM) algorithm, which allows modeling the degree of relationship between each voxels and a given tissue; and based on fuzzy genetic initialization for the centers of clusters by a Fuzzy C-Means (FCM) algorithm, and for which the result is optimized by genetic process. The visual results show a concordance between the ground truth segmentation and the hybrid algorithm results, which allows efficient tissue classification. The superiority was also proved with the quantitative results of the proposed method in comparison with the both conventional FCM and PCM algorithms.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"29 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122844718","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022695
Daranat Tansui, A. Thammano
In this article, a new optimization algorithm that is inspired by the biology of hydrozoa (HA) is proposed. Our aim was to develop an algorithm that is based on the regeneration and transplantation processes of hydrozoa for finding the best solutions for continuous optimization problems. Basically, HA follows the same general processes of evolutionary algorithm; however, its distinctive processes mimic the life cycle of 3 basic forms of hydrozoa: motile planula, polyps, and medusa. In particular, the growth of strong buds from the polyp stage depends on levels of morphogens: activators and inhibitors. These 3 forms develop or evolve into the best solution. HA was performance tested with 20 standard benchmark functions and compared with genetic algorithm and Particle Swarm Optimization (PSO). The test results have confirmed that the proposed algorithm is computationally more efficient than both GA and PSO. It works very well on most benchmark functions.
{"title":"Nature-inspired optimization method: Hydrozoan algorithm for solving continuous problems","authors":"Daranat Tansui, A. Thammano","doi":"10.1109/SNPD.2017.8022695","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022695","url":null,"abstract":"In this article, a new optimization algorithm that is inspired by the biology of hydrozoa (HA) is proposed. Our aim was to develop an algorithm that is based on the regeneration and transplantation processes of hydrozoa for finding the best solutions for continuous optimization problems. Basically, HA follows the same general processes of evolutionary algorithm; however, its distinctive processes mimic the life cycle of 3 basic forms of hydrozoa: motile planula, polyps, and medusa. In particular, the growth of strong buds from the polyp stage depends on levels of morphogens: activators and inhibitors. These 3 forms develop or evolve into the best solution. HA was performance tested with 20 standard benchmark functions and compared with genetic algorithm and Particle Swarm Optimization (PSO). The test results have confirmed that the proposed algorithm is computationally more efficient than both GA and PSO. It works very well on most benchmark functions.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"199 1-6","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114048302","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022776
Kotoko Yamaguchi, T. Hochin, Hiroki Nomiya, Yukiko Nishizaki
This paper tried to find the cause of fear by horror games. We prepared two types of horror games whose reality of pictures are changed, and observed the change of autonomic nervous function and the amount of perspiration. As a result, the physiological response to the horror game which has real pictures was scarier. Also, people who are familiar with horror games are scared at several horror scenes in the horror game, whereas the others are not. Additionally, men are afraid of the horror games of real pictures, while women feel scared of horror games of simple pictures.
{"title":"Effect of the reality of pictures in a horror game on feelings of fear","authors":"Kotoko Yamaguchi, T. Hochin, Hiroki Nomiya, Yukiko Nishizaki","doi":"10.1109/SNPD.2017.8022776","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022776","url":null,"abstract":"This paper tried to find the cause of fear by horror games. We prepared two types of horror games whose reality of pictures are changed, and observed the change of autonomic nervous function and the amount of perspiration. As a result, the physiological response to the horror game which has real pictures was scarier. Also, people who are familiar with horror games are scared at several horror scenes in the horror game, whereas the others are not. Additionally, men are afraid of the horror games of real pictures, while women feel scared of horror games of simple pictures.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116693217","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022732
Noraset Noradachanon, T. Senivongse
Identity management (IDM) refers to a security discipline that employs technologies to manage information about the identity of users and control their access to organization resources. Various IDM products are in the market to offer corporate customers productivity and security while lowering costs of identity management. System integrators who implement the IDM solutions for their corporate customers often face problems when choosing the right IDM products for integrating with the customers' enterprise systems. This paper proposes a decision model for IDM product selection which is used in evaluating and ranking IDM products based on the proposed customer requirements questionnaire. The model is comprehensive in that its decision criteria comprise not only the technical specifications of the products but also other important aspects including price and accountability of system integrators and product vendors. Based on the decision model, product selection is conducted using a multi-criteria decision making technique called fuzzy analytic hierarchy process (fuzzy AHP). In an evaluation, a leading system integrator company in Thailand applies the proposed approach to IDM product selection for its customer and the result is quite satisfactory.
{"title":"Decision model for identity management product selection using fuzzy AHP","authors":"Noraset Noradachanon, T. Senivongse","doi":"10.1109/SNPD.2017.8022732","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022732","url":null,"abstract":"Identity management (IDM) refers to a security discipline that employs technologies to manage information about the identity of users and control their access to organization resources. Various IDM products are in the market to offer corporate customers productivity and security while lowering costs of identity management. System integrators who implement the IDM solutions for their corporate customers often face problems when choosing the right IDM products for integrating with the customers' enterprise systems. This paper proposes a decision model for IDM product selection which is used in evaluating and ranking IDM products based on the proposed customer requirements questionnaire. The model is comprehensive in that its decision criteria comprise not only the technical specifications of the products but also other important aspects including price and accountability of system integrators and product vendors. Based on the decision model, product selection is conducted using a multi-criteria decision making technique called fuzzy analytic hierarchy process (fuzzy AHP). In an evaluation, a leading system integrator company in Thailand applies the proposed approach to IDM product selection for its customer and the result is quite satisfactory.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"34 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126285932","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022744
P. Galanakou, T. Leventouri, A. Georgakilas, G. Kalantzis
Intensity modulated radiation therapy (IMRT) exhibits the ability to deliver the prescribed dose to the planning target volume (PTV), while minimizing the delivered dose to the organs at risk (OARs). Metaheuristic algorithms, among them the simulating annealing algorithm (SAA), have been proposed in the past for optimization of IMRT. Despite the advantage of the SAA to be a global optimizer, IMRT optimization is an extensive computational task due to the large scale of the optimization variables. Therefore stochastic algorithms, such as the SAA, require significant computational resources. In an effort to elucidate the performance improvement of the SAA in highly dimensional optimization tasks, such as the IMRT optimization, we introduce for the first time to our best knowledge a parallel graphic processing unit (GPU)-based SAA developed in MATLAB platform and compliant with the computational environment for radiotherapy research (CERR) for IMRT treatment planning. Our strategy was firstly to identify the major “bottlenecks” of our code and secondly to parallelize those on the GPU accordingly. Performance tests were conducted on four different GPU cards in comparison to a serial version of the algorithm executed on a CPU. Our studies have shown a gradual increase of the speedup factor as a function of the number of beamlets for all four GPUs. Particularly, a maximum speedup factor of ∼33 was achieved when the K40m card was utilized.
{"title":"A parallelized GPU-based simulating annealing algorithm for intensity modulated radiation therapy optimization","authors":"P. Galanakou, T. Leventouri, A. Georgakilas, G. Kalantzis","doi":"10.1109/SNPD.2017.8022744","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022744","url":null,"abstract":"Intensity modulated radiation therapy (IMRT) exhibits the ability to deliver the prescribed dose to the planning target volume (PTV), while minimizing the delivered dose to the organs at risk (OARs). Metaheuristic algorithms, among them the simulating annealing algorithm (SAA), have been proposed in the past for optimization of IMRT. Despite the advantage of the SAA to be a global optimizer, IMRT optimization is an extensive computational task due to the large scale of the optimization variables. Therefore stochastic algorithms, such as the SAA, require significant computational resources. In an effort to elucidate the performance improvement of the SAA in highly dimensional optimization tasks, such as the IMRT optimization, we introduce for the first time to our best knowledge a parallel graphic processing unit (GPU)-based SAA developed in MATLAB platform and compliant with the computational environment for radiotherapy research (CERR) for IMRT treatment planning. Our strategy was firstly to identify the major “bottlenecks” of our code and secondly to parallelize those on the GPU accordingly. Performance tests were conducted on four different GPU cards in comparison to a serial version of the algorithm executed on a CPU. Our studies have shown a gradual increase of the speedup factor as a function of the number of beamlets for all four GPUs. Particularly, a maximum speedup factor of ∼33 was achieved when the K40m card was utilized.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"148 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116540111","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022761
S. Shimano, Atsushi Nunome, Y. Yokoi, Kiyoshi Shibayama, Hiroaki Hirata
We have proposed a distributed storage system which dynamically makes storage tiers and optimizes location of data blocks autonomously. This aims to enhance the I/O performance of the storage system without remarkable network overhead. Our system dynamically organizes storage tiers considering device characteristics. And the data blocks will be placed in a suitable storage tier according to their access pattern. In this paper, we propose a method to select the destination storage node for migration using an access characteristic of a data block to be migrated. This method ranks the storage nodes dynamically, and each storage node configures storage tiers autonomously. Simulation results show that our scheme can short the execution time of a program with file I/O by 49% at maximum, in comparison with the static migration without considering the access characteristics of the migration data.
{"title":"An autonomous configuration scheme of storage tiers for distributed file system","authors":"S. Shimano, Atsushi Nunome, Y. Yokoi, Kiyoshi Shibayama, Hiroaki Hirata","doi":"10.1109/SNPD.2017.8022761","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022761","url":null,"abstract":"We have proposed a distributed storage system which dynamically makes storage tiers and optimizes location of data blocks autonomously. This aims to enhance the I/O performance of the storage system without remarkable network overhead. Our system dynamically organizes storage tiers considering device characteristics. And the data blocks will be placed in a suitable storage tier according to their access pattern. In this paper, we propose a method to select the destination storage node for migration using an access characteristic of a data block to be migrated. This method ranks the storage nodes dynamically, and each storage node configures storage tiers autonomously. Simulation results show that our scheme can short the execution time of a program with file I/O by 49% at maximum, in comparison with the static migration without considering the access characteristics of the migration data.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130073898","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022772
H. Kumagai, T. Hochin, Hiroki Nomiya
This paper tries to clarify the retrieval and playback method considering temporal harmony. In this study, change points of the direction of object movement and the cycle of the motion are used as the accent points of a video clip. Beats are used as synchronization points of a music clip. The music measure is extracted in order to synchronize with the beat at the beginning of the measure. The synchronized playback method to make retrieved data impressive is also proposed.
{"title":"Retrieval and synchronized playback methods considering temporal harmony of music and video clips","authors":"H. Kumagai, T. Hochin, Hiroki Nomiya","doi":"10.1109/SNPD.2017.8022772","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022772","url":null,"abstract":"This paper tries to clarify the retrieval and playback method considering temporal harmony. In this study, change points of the direction of object movement and the cycle of the motion are used as the accent points of a video clip. Beats are used as synchronization points of a music clip. The music measure is extracted in order to synchronize with the beat at the beginning of the measure. The synchronized playback method to make retrieved data impressive is also proposed.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133843607","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 : 2017-06-01DOI: 10.1109/SNPD.2017.8022700
Wenjun Bai, Changqin Quan, Zhiwei Luo
In order to defend adversarial attacks in computer vision models, the conventional approach arises on actively incorporate such samples into the training datasets. Nonetheless, the manual production of adversarial samples is painful and labor intensive. Here we propose a novel generative model: Convolutional Autoencoder Model to add unsupervised adversarial training, i.e., the production of adversarial images from the encoded feature representation, on conventional supervised convolutional neural network training. To accomplish such objective, we first provide a novel statistical understanding of convolutional neural network to translate convolution and pooling computations equivalently as a hierarchy of encoders, and sampling tricks, respectively. Then, we derive our proposed Convolutional Autoencoder Model with the ‘adversarial decoders’ to automate the generation of adversarial samples. We validated our proposed Convolutional Autoencoder Model on MNIST dataset, and achieved the clear-cut performance improvement over the normal Convolutional Neural Network.
{"title":"Alleviating adversarial attacks via convolutional autoencoder","authors":"Wenjun Bai, Changqin Quan, Zhiwei Luo","doi":"10.1109/SNPD.2017.8022700","DOIUrl":"https://doi.org/10.1109/SNPD.2017.8022700","url":null,"abstract":"In order to defend adversarial attacks in computer vision models, the conventional approach arises on actively incorporate such samples into the training datasets. Nonetheless, the manual production of adversarial samples is painful and labor intensive. Here we propose a novel generative model: Convolutional Autoencoder Model to add unsupervised adversarial training, i.e., the production of adversarial images from the encoded feature representation, on conventional supervised convolutional neural network training. To accomplish such objective, we first provide a novel statistical understanding of convolutional neural network to translate convolution and pooling computations equivalently as a hierarchy of encoders, and sampling tricks, respectively. Then, we derive our proposed Convolutional Autoencoder Model with the ‘adversarial decoders’ to automate the generation of adversarial samples. We validated our proposed Convolutional Autoencoder Model on MNIST dataset, and achieved the clear-cut performance improvement over the normal Convolutional Neural Network.","PeriodicalId":186094,"journal":{"name":"2017 18th IEEE/ACIS International Conference on Software Engineering, Artificial Intelligence, Networking and Parallel/Distributed Computing (SNPD)","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2017-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123842083","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：