Pub Date : 2019-12-01DOI: 10.1109/SSCI44817.2019.9002958
S. Saha, Thiago Rios, Leandro L. Minku, X. Yao, Zhao Xu, B. Sendhoff, S. Menzel
Shape morphing methods are a key representation in human user-centered design as well as computational optimization of engineering applications in the automotive domain.3D digital objects are modified using deformation algorithms to alter the shape for optimal product performance or design aesthetics. We imagine a system which can learn from historic user deformation sequences and support the user in present design tasks by predicting potential design variations based on currently observed design changes carried out by the user. Towards a practical realization, a large amount of human user deformation sequence data is required which is practically not available. To overcome this limitation, we propose to use a computational target shape matching optimization whose hyper-parameters are tuned to exemplary human user sequence data and that allows us to afterwards generate large data-sets of human-like shape modification data in an automated fashion. In addition, we classified the user sequences to experience levels based on their variance. These user experience-tuned evolutionary optimizers allow us in future to mimic different user behavior and generate a large number of potential design variations in an automated fashion.
{"title":"Optimal Evolutionary Optimization Hyper-parameters to Mimic Human User Behavior","authors":"S. Saha, Thiago Rios, Leandro L. Minku, X. Yao, Zhao Xu, B. Sendhoff, S. Menzel","doi":"10.1109/SSCI44817.2019.9002958","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9002958","url":null,"abstract":"Shape morphing methods are a key representation in human user-centered design as well as computational optimization of engineering applications in the automotive domain.3D digital objects are modified using deformation algorithms to alter the shape for optimal product performance or design aesthetics. We imagine a system which can learn from historic user deformation sequences and support the user in present design tasks by predicting potential design variations based on currently observed design changes carried out by the user. Towards a practical realization, a large amount of human user deformation sequence data is required which is practically not available. To overcome this limitation, we propose to use a computational target shape matching optimization whose hyper-parameters are tuned to exemplary human user sequence data and that allows us to afterwards generate large data-sets of human-like shape modification data in an automated fashion. In addition, we classified the user sequences to experience levels based on their variance. These user experience-tuned evolutionary optimizers allow us in future to mimic different user behavior and generate a large number of potential design variations in an automated fashion.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"12 1","pages":"858-866"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75120040","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 : 2019-12-01DOI: 10.1109/ssci44817.2019.9002702
Guijin Han, Qian Li, You Zhou, Yue He
In view of the problems of high missed detection rate and inaccurate position of small targets in the cigarette detection algorithm based on Faster Regions Convolutional Neural Networks(Faster R-CNN), a cigarette detection algorithm based on Feature pyramid networks (FPN) and Faster R-CNN is proposed. The feature map with high-level semantic information and low-resolution of the last layer is adopted by the Faster R-CNN as the input of Region Proposal Network (RPN), resulting in low recognition rate of small targets. The improved Faster R-CNN framework combined with FPN algorithm continuously fuses the high-level feature maps with the feature maps of the front layer through up-sampling, and constructs the feature pyramid model of different scales as the input of RPN network, which improves the detection effect of cigarette effectively.
{"title":"Cigarette Detection Algorithm Based on Improved Faster R-CNN","authors":"Guijin Han, Qian Li, You Zhou, Yue He","doi":"10.1109/ssci44817.2019.9002702","DOIUrl":"https://doi.org/10.1109/ssci44817.2019.9002702","url":null,"abstract":"In view of the problems of high missed detection rate and inaccurate position of small targets in the cigarette detection algorithm based on Faster Regions Convolutional Neural Networks(Faster R-CNN), a cigarette detection algorithm based on Feature pyramid networks (FPN) and Faster R-CNN is proposed. The feature map with high-level semantic information and low-resolution of the last layer is adopted by the Faster R-CNN as the input of Region Proposal Network (RPN), resulting in low recognition rate of small targets. The improved Faster R-CNN framework combined with FPN algorithm continuously fuses the high-level feature maps with the feature maps of the front layer through up-sampling, and constructs the feature pyramid model of different scales as the input of RPN network, which improves the detection effect of cigarette effectively.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"73 1","pages":"2766-2770"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74135166","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 : 2019-12-01DOI: 10.1109/SSCI44817.2019.9002700
Cheng-Lung Lee, Yuan Fang, Yi-Hsin Huang, Szu-Hao Lee, W. Yeh
In 2009, the United States National Academy of Sciences (NAS) proposed a list of suggestions to forensic-related issues, with emphasis on the correct use of modern technology to improve on-the-scene investigation. To ensure that the scene is fully detailed, a team of personnel including but not limited to photographer, videographer, evidence collector, and those who estimate and mark the scene are involved. However, traditional way of investigation not only utilizes too much manpower, but the destruction of crime scene is also an unavoidable problem we faced. Hence, the urgent tasks to be improved now are to protect, record and efficiently investigate the crime scene, to instantly deliver evidences for data matching, as well to provide an option to setup a video conference with experts in Forensics fields to assist officers on scene.The research integrates the concepts of "Wearable Devices" and "Forensic Cloud Computing", to showcase the benefits of Forensic database. Through cloud sharing, it is now possible to use portable devices for all on-the-scene records, video taking, and on-the-spot graphic detailing. Forensic Cloud is used with wireless and 3G/4G to deliver evidences and information. The user friendly design and mobile computing can be integrated together with instant communication and on-line forensic database. The research not only utilizes the latest technology to protect the integrity of crime scene, but will make a breakthrough in Crime Scene Investigation (CSI)—cost reduction in manpower, as well as enhancing the efficiency and effectiveness of investigation. Finally, with this research, we hope to strengthen scientific evidence, to push forward judiciary reformation, and to reduce the possibility of erroneous conviction.
{"title":"Application of Wearable Devices in Crime Scene Investigation and Virtual Reality","authors":"Cheng-Lung Lee, Yuan Fang, Yi-Hsin Huang, Szu-Hao Lee, W. Yeh","doi":"10.1109/SSCI44817.2019.9002700","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9002700","url":null,"abstract":"In 2009, the United States National Academy of Sciences (NAS) proposed a list of suggestions to forensic-related issues, with emphasis on the correct use of modern technology to improve on-the-scene investigation. To ensure that the scene is fully detailed, a team of personnel including but not limited to photographer, videographer, evidence collector, and those who estimate and mark the scene are involved. However, traditional way of investigation not only utilizes too much manpower, but the destruction of crime scene is also an unavoidable problem we faced. Hence, the urgent tasks to be improved now are to protect, record and efficiently investigate the crime scene, to instantly deliver evidences for data matching, as well to provide an option to setup a video conference with experts in Forensics fields to assist officers on scene.The research integrates the concepts of \"Wearable Devices\" and \"Forensic Cloud Computing\", to showcase the benefits of Forensic database. Through cloud sharing, it is now possible to use portable devices for all on-the-scene records, video taking, and on-the-spot graphic detailing. Forensic Cloud is used with wireless and 3G/4G to deliver evidences and information. The user friendly design and mobile computing can be integrated together with instant communication and on-line forensic database. The research not only utilizes the latest technology to protect the integrity of crime scene, but will make a breakthrough in Crime Scene Investigation (CSI)—cost reduction in manpower, as well as enhancing the efficiency and effectiveness of investigation. Finally, with this research, we hope to strengthen scientific evidence, to push forward judiciary reformation, and to reduce the possibility of erroneous conviction.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"65 1","pages":"206-210"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74408817","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 : 2019-12-01DOI: 10.1109/SSCI44817.2019.9002848
Zaixiang Zhang, Yunhao Zhu, A. Fujiwara, K. Ohnishi
We develop two types of simple population-based search algorithms that model two types of scale-free behaviors of individuals. The scale-free behavior is a particular behavior of individuals of species that search for food not cooperatively but independently. One type of the scale-free behaviors is that a moving distance of an individual from the present food source follows a power low distribution, which is called the spatial scale-free behavior. The other is that a staying duration of an individual at the preset food source follows a power low distribution, which is called the temporal scale-free behavior. We assume static and dynamic problems in which a position of the best food source (the global optimum) is not changed and changed, respectively. In addition, we assume a special event that individuals near the best food source are probabilistically eliminated. We compare the two search algorithms and show that they are complementary with respect to suitable problems. Therefore, we develop a search algorithm that initially includes both types of individuals in a population and evolutionarily adaptively increases an appropriate type of individuals in it. The algorithm is shown to be not the best but work quite well for any problems used.
{"title":"Population-based Search Relying on Spatial and/or Temporal Scale-free Behaviors of Individuals","authors":"Zaixiang Zhang, Yunhao Zhu, A. Fujiwara, K. Ohnishi","doi":"10.1109/SSCI44817.2019.9002848","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9002848","url":null,"abstract":"We develop two types of simple population-based search algorithms that model two types of scale-free behaviors of individuals. The scale-free behavior is a particular behavior of individuals of species that search for food not cooperatively but independently. One type of the scale-free behaviors is that a moving distance of an individual from the present food source follows a power low distribution, which is called the spatial scale-free behavior. The other is that a staying duration of an individual at the preset food source follows a power low distribution, which is called the temporal scale-free behavior. We assume static and dynamic problems in which a position of the best food source (the global optimum) is not changed and changed, respectively. In addition, we assume a special event that individuals near the best food source are probabilistically eliminated. We compare the two search algorithms and show that they are complementary with respect to suitable problems. Therefore, we develop a search algorithm that initially includes both types of individuals in a population and evolutionarily adaptively increases an appropriate type of individuals in it. The algorithm is shown to be not the best but work quite well for any problems used.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"106 1","pages":"2327-2334"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74813457","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 : 2019-12-01DOI: 10.1109/SSCI44817.2019.9003021
Yapei Wu, Xingguang Peng, Demin Xu
Identifying the interaction of search variables of black-box optimization problem and exploiting the learned interaction structure back to optimization process is a very meaningful research topic. Evaluating the interaction between variables based on information theory is a popular and effective method. However, very little research pay attention to what kind of data can help identify interactions between variables. In this paper, we propose a method to identify the interaction between variables by using the local optima solutions of the objective function. First, a multimodal optimization algorithm is used to search for multiple local optima of the optimization problem. Then, hierarchical clustering is used to cluster and discretize local optima. Finally, the interaction between variables is quantified using the mutual information of local optima. Experimental results show that the proposed method can use the information of local optima to identify the interaction of search variables.
{"title":"Identifying Variables Interaction for Black-box Continuous Optimization with Mutual Information of Multiple Local Optima","authors":"Yapei Wu, Xingguang Peng, Demin Xu","doi":"10.1109/SSCI44817.2019.9003021","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9003021","url":null,"abstract":"Identifying the interaction of search variables of black-box optimization problem and exploiting the learned interaction structure back to optimization process is a very meaningful research topic. Evaluating the interaction between variables based on information theory is a popular and effective method. However, very little research pay attention to what kind of data can help identify interactions between variables. In this paper, we propose a method to identify the interaction between variables by using the local optima solutions of the objective function. First, a multimodal optimization algorithm is used to search for multiple local optima of the optimization problem. Then, hierarchical clustering is used to cluster and discretize local optima. Finally, the interaction between variables is quantified using the mutual information of local optima. Experimental results show that the proposed method can use the information of local optima to identify the interaction of search variables.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"32 1","pages":"2683-2689"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74112505","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 : 2019-12-01DOI: 10.1109/SSCI44817.2019.9003090
R. Sifa, Maren Pielka, Rajkumar Ramamurthy, Anna Ladi, L. Hillebrand, C. Bauckhage
With the recent advancements in Machine Learning based Natural Language Processing (NLP), language dependency has always been a limiting factor for a majority of NLP applications. Typically, models are trained for the English language due to the availability of very large labeled and unlabeled datasets, which also allow to fine tune models for that language. Contradiction Detection is one such problem that has found many practical applications in NLP and up to this point has only been studied in the context of English language. The scope of this paper is to examine a set of baseline methods for the Contradiction Detection task on German text. For this purpose, the well-known Stanford Natural Language Inference (SNLI) data set (110,000 sentence pairs) is machine-translated from English to German. We train and evaluate four classifiers on both the original and the translated data, using state-of-the-art textual data representations. Our main contribution is the first large-scale assessment for this problem in German, and a validation of machine translation as a data generation method. We also present a novel approach to learn sentence embeddings by exploiting the hidden states of an encoder-decoder Sequence-To-Sequence RNN trained for autoencoding or translation.
{"title":"Towards Contradiction Detection in German: a Translation-Driven Approach","authors":"R. Sifa, Maren Pielka, Rajkumar Ramamurthy, Anna Ladi, L. Hillebrand, C. Bauckhage","doi":"10.1109/SSCI44817.2019.9003090","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9003090","url":null,"abstract":"With the recent advancements in Machine Learning based Natural Language Processing (NLP), language dependency has always been a limiting factor for a majority of NLP applications. Typically, models are trained for the English language due to the availability of very large labeled and unlabeled datasets, which also allow to fine tune models for that language. Contradiction Detection is one such problem that has found many practical applications in NLP and up to this point has only been studied in the context of English language. The scope of this paper is to examine a set of baseline methods for the Contradiction Detection task on German text. For this purpose, the well-known Stanford Natural Language Inference (SNLI) data set (110,000 sentence pairs) is machine-translated from English to German. We train and evaluate four classifiers on both the original and the translated data, using state-of-the-art textual data representations. Our main contribution is the first large-scale assessment for this problem in German, and a validation of machine translation as a data generation method. We also present a novel approach to learn sentence embeddings by exploiting the hidden states of an encoder-decoder Sequence-To-Sequence RNN trained for autoencoding or translation.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"46 1","pages":"2497-2505"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74130441","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 : 2019-12-01DOI: 10.1109/SSCI44817.2019.9003098
Naoki Masuyama, Narito Amako, Y. Nojima, Yiping Liu, C. Loo, H. Ishibuchi
Adaptive Resonance Theory (ART)-based growing self-organizing clustering is one of the most promising approaches for unsupervised topological clustering. In our previous study, we proposed a Topological Correntropy induced metric based ART (TCA) and shown its superior performance. However, TCA suffers from a data-dependent parameter and a complicated network creation process which lead to inefficient learning. This paper aims to solve problems of TCA by implementing an automatic parameter specification mechanism and simplifying a learning algorithm. Experimental results show that the proposed algorithm in this paper successfully solved the above problems.
{"title":"Fast Topological Adaptive Resonance Theory Based on Correntropy Induced Metric","authors":"Naoki Masuyama, Narito Amako, Y. Nojima, Yiping Liu, C. Loo, H. Ishibuchi","doi":"10.1109/SSCI44817.2019.9003098","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9003098","url":null,"abstract":"Adaptive Resonance Theory (ART)-based growing self-organizing clustering is one of the most promising approaches for unsupervised topological clustering. In our previous study, we proposed a Topological Correntropy induced metric based ART (TCA) and shown its superior performance. However, TCA suffers from a data-dependent parameter and a complicated network creation process which lead to inefficient learning. This paper aims to solve problems of TCA by implementing an automatic parameter specification mechanism and simplifying a learning algorithm. Experimental results show that the proposed algorithm in this paper successfully solved the above problems.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"19 1","pages":"2215-2221"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74415874","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 : 2019-12-01DOI: 10.1109/SSCI44817.2019.9002838
Chengxin Xian, Yu Zhao
Distributed average tracking (DAT) problems are investigated for general linear dynamical systems under undirected connected topology in this paper. A kind of distributed event-triggered DAT algorithms with static gain is designed by using model-based local sampled state information. The control objective of the considered DAT problem is achieved by using the proposed event-triggered DAT algorithms. Meanwhile, the Zeno behavior is excluded. Finally, a simulation example is presented to validate the proposed control laws.
{"title":"Distributed Average Tracking with Event-Triggered Algorithms of Multi-Agent Systems","authors":"Chengxin Xian, Yu Zhao","doi":"10.1109/SSCI44817.2019.9002838","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9002838","url":null,"abstract":"Distributed average tracking (DAT) problems are investigated for general linear dynamical systems under undirected connected topology in this paper. A kind of distributed event-triggered DAT algorithms with static gain is designed by using model-based local sampled state information. The control objective of the considered DAT problem is achieved by using the proposed event-triggered DAT algorithms. Meanwhile, the Zeno behavior is excluded. Finally, a simulation example is presented to validate the proposed control laws.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"13 1","pages":"1982-1987"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74480101","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 : 2019-12-01DOI: 10.1109/SSCI44817.2019.9003087
Qiuyue Wei, Yufan Liu, Guangyuan Zhao, Bo Shang, Yun Yan, Yue He
Aiming at the requirement of experimental and exhibition, a six-degree-of-freedom control system with ATmega328P as the main controller, steering gear driving module as the motion unit of the manipulator is designed based on the analysis of the experimental application of the traditional manipulator control system. The system has two control modes: "handle control" and "automatic gesture action". According to the structural characteristics of the manipulator, the Denavit-Hartenberg(D-H) coordinate was obtained and its forward kinematic analysis was done. The experimental results show that the control system can quickly and accurately adjust the trajectory of the robot arm to complete the corresponding action, which can be used for experiments and exhibitions.
{"title":"Design of the Controlling System of a Six-DoF Manipulator","authors":"Qiuyue Wei, Yufan Liu, Guangyuan Zhao, Bo Shang, Yun Yan, Yue He","doi":"10.1109/SSCI44817.2019.9003087","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9003087","url":null,"abstract":"Aiming at the requirement of experimental and exhibition, a six-degree-of-freedom control system with ATmega328P as the main controller, steering gear driving module as the motion unit of the manipulator is designed based on the analysis of the experimental application of the traditional manipulator control system. The system has two control modes: \"handle control\" and \"automatic gesture action\". According to the structural characteristics of the manipulator, the Denavit-Hartenberg(D-H) coordinate was obtained and its forward kinematic analysis was done. The experimental results show that the control system can quickly and accurately adjust the trajectory of the robot arm to complete the corresponding action, which can be used for experiments and exhibitions.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"15 1","pages":"2874-2878"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"75269527","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 : 2019-12-01DOI: 10.1109/SSCI44817.2019.9002808
Zhiwei Liu, Yuanlong Yu, Zhenzhen Sun
Feature selection is an important data preprocessing for machine learning. It can improve the performance of machine learning algorithms by removing redundant and noisy features. Among all the methods, those based on l1-norms or l2,1-norms have received considerable attention due to their good performance. However, these methods cannot produce exact row sparsity to the weight matrix, so the number of selected features cannot be determined automatically without using a threshold. To this end, this paper proposes a feature selection method incorporating the l2,0-norm, which can guarantee exact row sparsity of weight matrix. A method based on iterative hard thresholding (IHT) algorithm is also proposed to solve the l2,0- norm regularized least square problem. For fully using the role of row-sparsity induced by the l2,0-norm, this method acts as network pruning for single-hidden-layer neural networks. This method is conducted on the hidden features and it can achieve node-level pruning rather than the connection-level pruning. The experimental results in several public data sets and three image recognition data sets have shown that this method can not only effectively prune the useless hidden nodes, but also obtain better performance.
{"title":"A Hidden Feature Selection Method based on l2,0-Norm Regularization for Training Single-hidden-layer Neural Networks","authors":"Zhiwei Liu, Yuanlong Yu, Zhenzhen Sun","doi":"10.1109/SSCI44817.2019.9002808","DOIUrl":"https://doi.org/10.1109/SSCI44817.2019.9002808","url":null,"abstract":"Feature selection is an important data preprocessing for machine learning. It can improve the performance of machine learning algorithms by removing redundant and noisy features. Among all the methods, those based on l1-norms or l2,1-norms have received considerable attention due to their good performance. However, these methods cannot produce exact row sparsity to the weight matrix, so the number of selected features cannot be determined automatically without using a threshold. To this end, this paper proposes a feature selection method incorporating the l2,0-norm, which can guarantee exact row sparsity of weight matrix. A method based on iterative hard thresholding (IHT) algorithm is also proposed to solve the l2,0- norm regularized least square problem. For fully using the role of row-sparsity induced by the l2,0-norm, this method acts as network pruning for single-hidden-layer neural networks. This method is conducted on the hidden features and it can achieve node-level pruning rather than the connection-level pruning. The experimental results in several public data sets and three image recognition data sets have shown that this method can not only effectively prune the useless hidden nodes, but also obtain better performance.","PeriodicalId":6729,"journal":{"name":"2019 IEEE Symposium Series on Computational Intelligence (SSCI)","volume":"50 1","pages":"1810-1817"},"PeriodicalIF":0.0,"publicationDate":"2019-12-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76112972","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：