Pub Date : 2015-07-12DOI: 10.1109/IJCNN.2015.7280484
B. Mitchell, H. Tosun, John W. Sheppard
Deep learning is a popular field that encompasses a range of multi-layer connectionist techniques. While these techniques have achieved great success on a number of difficult computer vision problems, the representation biases that allow this success have not been thoroughly explored. In this paper, we examine the hypothesis that one strength of many deep learning algorithms is their ability to exploit spatially local statistical information. We present a formal description of how data vectors can be partitioned into sub-vectors that preserve spatially local information. As a test case, we then use statistical models to examine how much of such structure exists in the MNIST dataset. Finally, we present experimental results from training RBMs using partitioned data, and demonstrate the advantages they have over non-partitioned RBMs. Through these results, we show how the performance advantage is reliant on spatially local structure, by demonstrating the performance impact of randomly permuting the input data to destroy local structure. Overall, our results support the hypothesis that a representation bias reliant upon spatially local statistical information can improve performance, so long as this bias is a good match for the data. We also suggest statistical tools for determining a priori whether a dataset is a good match for this bias or not.
{"title":"Deep learning using partitioned data vectors","authors":"B. Mitchell, H. Tosun, John W. Sheppard","doi":"10.1109/IJCNN.2015.7280484","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280484","url":null,"abstract":"Deep learning is a popular field that encompasses a range of multi-layer connectionist techniques. While these techniques have achieved great success on a number of difficult computer vision problems, the representation biases that allow this success have not been thoroughly explored. In this paper, we examine the hypothesis that one strength of many deep learning algorithms is their ability to exploit spatially local statistical information. We present a formal description of how data vectors can be partitioned into sub-vectors that preserve spatially local information. As a test case, we then use statistical models to examine how much of such structure exists in the MNIST dataset. Finally, we present experimental results from training RBMs using partitioned data, and demonstrate the advantages they have over non-partitioned RBMs. Through these results, we show how the performance advantage is reliant on spatially local structure, by demonstrating the performance impact of randomly permuting the input data to destroy local structure. Overall, our results support the hypothesis that a representation bias reliant upon spatially local statistical information can improve performance, so long as this bias is a good match for the data. We also suggest statistical tools for determining a priori whether a dataset is a good match for this bias or not.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"5 1","pages":"1-8"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"85534956","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 : 2015-07-12DOI: 10.1109/IJCNN.2015.7280486
Mototaka Yoshioka, Kenta Murakami, Jun Ozawa
This paper proposes a method to estimate temporally accurate human pulse peaks for noncontact pulse transit time (PTT) measurements. The PTT is considered as a significant diagnostic index for conditions such as blood pressure and arterial stiffness; however, millisecond-order accuracy is required in the determination of each pulse peak. In this study, human pulse waveforms are obtained from wrist and ankle images taken using a webcam at 90 cm distance. In the proposed method, the waveform is smoothed using finite impulse response low-pass filtering that sustains the shape of the pulse waveform, and the phase delay is compensated. Then, features of the first-order derivative of the filtered waveform are used to estimate the pulse peaks. The interbeat intervals obtained from the peaks estimated by the proposed method closely coincided with those obtained from a contact-type photoplethysmogram sensor, yielding less absolute error than that obtained from a comparative method; thus, this confirms the improved temporal accuracy of the proposed method. The PTTs are calculated from the time differences between the estimated pulse peaks of the wrist and those of the ankle images. The benefit of accurate pulse peak estimation is demonstrated by investigating the relation between the PTT and blood pressure. The PTTs are correlated with blood pressure in ten human participants, and a high correlation coefficient of -0.88 was obtained, indicating a direct relation between these two measures.
{"title":"Improved human pulse peak estimation using derivative features for noncontact pulse transit time measurements","authors":"Mototaka Yoshioka, Kenta Murakami, Jun Ozawa","doi":"10.1109/IJCNN.2015.7280486","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280486","url":null,"abstract":"This paper proposes a method to estimate temporally accurate human pulse peaks for noncontact pulse transit time (PTT) measurements. The PTT is considered as a significant diagnostic index for conditions such as blood pressure and arterial stiffness; however, millisecond-order accuracy is required in the determination of each pulse peak. In this study, human pulse waveforms are obtained from wrist and ankle images taken using a webcam at 90 cm distance. In the proposed method, the waveform is smoothed using finite impulse response low-pass filtering that sustains the shape of the pulse waveform, and the phase delay is compensated. Then, features of the first-order derivative of the filtered waveform are used to estimate the pulse peaks. The interbeat intervals obtained from the peaks estimated by the proposed method closely coincided with those obtained from a contact-type photoplethysmogram sensor, yielding less absolute error than that obtained from a comparative method; thus, this confirms the improved temporal accuracy of the proposed method. The PTTs are calculated from the time differences between the estimated pulse peaks of the wrist and those of the ankle images. The benefit of accurate pulse peak estimation is demonstrated by investigating the relation between the PTT and blood pressure. The PTTs are correlated with blood pressure in ten human participants, and a high correlation coefficient of -0.88 was obtained, indicating a direct relation between these two measures.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"10 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81962549","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 : 2015-07-12DOI: 10.1109/IJCNN.2015.7280541
R. Kamimura, Ryozo Kitajima
The present paper proposes a new type of information-theoretic method to enhance the potentiality of input neurons for improving the class structure of the self-organizing maps (SOM). The SOM has received much attention in neural networks, because it can be used to visualize input patterns, in particular, to clarify class structure. However, it has been observed that the good performance of visualization is limited to relatively simple data sets. To visualize more complex data sets, it is needed to develop a method to extract main characteristics of input patterns more explicitly. For this, several information-theoretic methods have been developed with some problems. One of the main problems is that the method needs much heavy computation to obtain the main features, because the computational procedures to obtain information content should be repeated many times. To simplify the procedures, a new measure called “potentiality” of input neurons is proposed. The potentiality is based on the variance of connection weights for input neurons and it can be computed without the complex computation of information content. The method was applied to the artificial and symmetric data set and the biodegradation data from the machine learning database. Experimental results showed that the method could be used to enhance a smaller number of input neurons. Those neurons were effective in intensifying class boundaries for clearer class structures. The present results show the effectiveness of the new measure of the potentiality for improved visualization and class structure.
{"title":"Selective potentiality maximization for input neuron selection in self-organizing maps","authors":"R. Kamimura, Ryozo Kitajima","doi":"10.1109/IJCNN.2015.7280541","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280541","url":null,"abstract":"The present paper proposes a new type of information-theoretic method to enhance the potentiality of input neurons for improving the class structure of the self-organizing maps (SOM). The SOM has received much attention in neural networks, because it can be used to visualize input patterns, in particular, to clarify class structure. However, it has been observed that the good performance of visualization is limited to relatively simple data sets. To visualize more complex data sets, it is needed to develop a method to extract main characteristics of input patterns more explicitly. For this, several information-theoretic methods have been developed with some problems. One of the main problems is that the method needs much heavy computation to obtain the main features, because the computational procedures to obtain information content should be repeated many times. To simplify the procedures, a new measure called “potentiality” of input neurons is proposed. The potentiality is based on the variance of connection weights for input neurons and it can be computed without the complex computation of information content. The method was applied to the artificial and symmetric data set and the biodegradation data from the machine learning database. Experimental results showed that the method could be used to enhance a smaller number of input neurons. Those neurons were effective in intensifying class boundaries for clearer class structures. The present results show the effectiveness of the new measure of the potentiality for improved visualization and class structure.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"4 1","pages":"1-8"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"81997194","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 : 2015-07-12DOI: 10.1109/IJCNN.2015.7280803
M. Saffar, Banafsheh Rekabdar, S. Louis, M. Nicolescu
This paper investigates three approaches to the problem of identity recognition in real-world unconstrained environments. We describe a new and challenging face recognition dataset captured in a laboratory environment with no strong constraints on lighting, motion, or subject pose, orientation, distance, or facial expression. We then evaluate three approaches to identity recognition on this new dataset. We find that a deep neural network with stacked denoising auto-encoders significantly outperforms a standard feedforward neural network and a baseline eigenfaces approach from the OpenCV library. Despite the 66 million plus parameters in the best trained deep network, it significantly outperforms the other two methods even on the relatively small number (relative to the number of deep network parameters) of 8,895 training samples. We believe our work adds to the growing empirical and theoretical evidence that deep networks provide a promising approach to unconstrained recognition problems.
{"title":"Face recognition in unconstrained environments","authors":"M. Saffar, Banafsheh Rekabdar, S. Louis, M. Nicolescu","doi":"10.1109/IJCNN.2015.7280803","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280803","url":null,"abstract":"This paper investigates three approaches to the problem of identity recognition in real-world unconstrained environments. We describe a new and challenging face recognition dataset captured in a laboratory environment with no strong constraints on lighting, motion, or subject pose, orientation, distance, or facial expression. We then evaluate three approaches to identity recognition on this new dataset. We find that a deep neural network with stacked denoising auto-encoders significantly outperforms a standard feedforward neural network and a baseline eigenfaces approach from the OpenCV library. Despite the 66 million plus parameters in the best trained deep network, it significantly outperforms the other two methods even on the relatively small number (relative to the number of deep network parameters) of 8,895 training samples. We believe our work adds to the growing empirical and theoretical evidence that deep networks provide a promising approach to unconstrained recognition problems.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"5 1","pages":"1-7"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"82089766","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 : 2015-07-12DOI: 10.1109/IJCNN.2015.7280677
Chang Liu, F. Sun
HMAX model is a bio-inspired feedforward architecture for object recognition, which is derived from the simple and complex cells model in cortex proposed by Hubel and Wiesel. As a hierarchical bio-based recognition model, HMAX captures the properties of primate cortex with alternated S layers and C layers, corresponding to simple cells and complex cells respectively. Although constrained by biological factors, HMAX shows satisfying performance in different fields when competing with other state-of-the-art computer vision algorithms. Insightful ideas and methods have been developed for this hierarchical model, which advances the progress of HMAX model. This paper reviews the origin of this model, as well as the improvements and modifications based on this model.
{"title":"HMAX model: A survey","authors":"Chang Liu, F. Sun","doi":"10.1109/IJCNN.2015.7280677","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280677","url":null,"abstract":"HMAX model is a bio-inspired feedforward architecture for object recognition, which is derived from the simple and complex cells model in cortex proposed by Hubel and Wiesel. As a hierarchical bio-based recognition model, HMAX captures the properties of primate cortex with alternated S layers and C layers, corresponding to simple cells and complex cells respectively. Although constrained by biological factors, HMAX shows satisfying performance in different fields when competing with other state-of-the-art computer vision algorithms. Insightful ideas and methods have been developed for this hierarchical model, which advances the progress of HMAX model. This paper reviews the origin of this model, as well as the improvements and modifications based on this model.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"346 1","pages":"1-7"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"79660481","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 : 2015-07-12DOI: 10.1109/IJCNN.2015.7280518
S. Pal, Alireza Alaei, U. Pal, M. Blumenstein
The objective of this investigation is to present an interval-symbolic representation based method for offline signature verification. In the feature extraction stage, Connected Components (CC), Enclosed Regions (ER), Basic Features (BF) and Curvelet Feature (CF)-based approaches are used to characterize signatures. Considering the extracted feature vectors, an interval data value is created for each feature extracted from every individual's signatures as an interval-valued symbolic data. This process results in a signature model for each individual that consists of a set of interval values. A similarity measure is proposed as the classifier in this paper. The interval-valued symbolic representation based method has never been used for signature verification considering Indian script signatures. Therefore, to evaluate the proposed method, a Hindi signature database consisting of 2400 (100×24) genuine signatures and 3000 (100×30) skilled forgeries is employed for experimentation. Concerning this large Hindi signature dataset, the highest verification accuracy of 91.83% was obtained on a joint feature set considering all four sets of features, while 2.5%, 13.84% and 8.17% of FAR (False Acceptance Rate), FRR (False Rejection Rate), and AER (Average Error Rate) were achieved, respectively.
{"title":"Interval-valued symbolic representation based method for off-line signature verification","authors":"S. Pal, Alireza Alaei, U. Pal, M. Blumenstein","doi":"10.1109/IJCNN.2015.7280518","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280518","url":null,"abstract":"The objective of this investigation is to present an interval-symbolic representation based method for offline signature verification. In the feature extraction stage, Connected Components (CC), Enclosed Regions (ER), Basic Features (BF) and Curvelet Feature (CF)-based approaches are used to characterize signatures. Considering the extracted feature vectors, an interval data value is created for each feature extracted from every individual's signatures as an interval-valued symbolic data. This process results in a signature model for each individual that consists of a set of interval values. A similarity measure is proposed as the classifier in this paper. The interval-valued symbolic representation based method has never been used for signature verification considering Indian script signatures. Therefore, to evaluate the proposed method, a Hindi signature database consisting of 2400 (100×24) genuine signatures and 3000 (100×30) skilled forgeries is employed for experimentation. Concerning this large Hindi signature dataset, the highest verification accuracy of 91.83% was obtained on a joint feature set considering all four sets of features, while 2.5%, 13.84% and 8.17% of FAR (False Acceptance Rate), FRR (False Rejection Rate), and AER (Average Error Rate) were achieved, respectively.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"29 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84129554","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 : 2015-07-12DOI: 10.1109/IJCNN.2015.7280590
Seyed Amin Khatami, S. Mirghasemi, A. Khosravi, S. Nahavandi
Proposing efficient methods for fire protection is becoming more and more important, because a small flame of fire may cause huge problems in social safety. In this paper, an effective fire flame detection method is investigated. This fire detection method includes four main stages: in the first step, a linear transformation is applied to convert red, green, and blue (RGB) color space through a 3*3 matrix to a new color space. In the next step, fuzzy c-mean clustering method (FCM) is used to distinguish between fire flame and non-fire flame pixels. Particle Swarm Optimization algorithm (PSO) is also utilized in the last step to decrease the error value measured by FCM after conversion. Finally, we apply Otsu threshold method to the new converted images to make a binary picture. Empirical results show the strength, accuracy and fast-response of the proposed algorithm in detecting fire flames in color images.
{"title":"An efficient hybrid algorithm for fire flame detection","authors":"Seyed Amin Khatami, S. Mirghasemi, A. Khosravi, S. Nahavandi","doi":"10.1109/IJCNN.2015.7280590","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280590","url":null,"abstract":"Proposing efficient methods for fire protection is becoming more and more important, because a small flame of fire may cause huge problems in social safety. In this paper, an effective fire flame detection method is investigated. This fire detection method includes four main stages: in the first step, a linear transformation is applied to convert red, green, and blue (RGB) color space through a 3*3 matrix to a new color space. In the next step, fuzzy c-mean clustering method (FCM) is used to distinguish between fire flame and non-fire flame pixels. Particle Swarm Optimization algorithm (PSO) is also utilized in the last step to decrease the error value measured by FCM after conversion. Finally, we apply Otsu threshold method to the new converted images to make a binary picture. Empirical results show the strength, accuracy and fast-response of the proposed algorithm in detecting fire flames in color images.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"29 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84135840","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 : 2015-07-12DOI: 10.1109/IJCNN.2015.7280562
Vigneshwaran Senthilvel, B. S. Mahanand, S. Sundaram, N. Sundararajan
This paper presents an approach for automatic diagnosis of Autism Spectrum Disorder (ASD) among males using functional Magnetic Resonance Imaging (fMRI). fMRI has the capability to identify any abnormal neural interactions that may be responsible for behavioral symptoms observed in ASD patients. In this paper, the regional homogeneity of the voxels in the 116 regions of the automated anatomical labeling (AAL) atlas of the brain are used as features which result in a large set of 54837 features. Chi-square feature selection method is then used to identify the most significant features and only these features are then used for classification with a metacognitive radial basis function classifier. Since genetic studies have indicated that ASD manifests differently in males and females, a large scale study specific to males is highlighted here using the publicly available preprocessed fMRI dataset from the Autism Brain Imaging Data Exchange (ABIDE), unlike existing studies which are either smaller in scale or consider both males and females together. Among the males, it is shown here that the classification performance can be improved (by up to 10%) by considering adults and adolescents separately. By using Chi-square algorithm the number of features was reduced drastically to lower than 200 in contrast to the thousands of features that have been used in recent studies.
{"title":"Using regional homogeneity from functional MRI for diagnosis of ASD among males","authors":"Vigneshwaran Senthilvel, B. S. Mahanand, S. Sundaram, N. Sundararajan","doi":"10.1109/IJCNN.2015.7280562","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280562","url":null,"abstract":"This paper presents an approach for automatic diagnosis of Autism Spectrum Disorder (ASD) among males using functional Magnetic Resonance Imaging (fMRI). fMRI has the capability to identify any abnormal neural interactions that may be responsible for behavioral symptoms observed in ASD patients. In this paper, the regional homogeneity of the voxels in the 116 regions of the automated anatomical labeling (AAL) atlas of the brain are used as features which result in a large set of 54837 features. Chi-square feature selection method is then used to identify the most significant features and only these features are then used for classification with a metacognitive radial basis function classifier. Since genetic studies have indicated that ASD manifests differently in males and females, a large scale study specific to males is highlighted here using the publicly available preprocessed fMRI dataset from the Autism Brain Imaging Data Exchange (ABIDE), unlike existing studies which are either smaller in scale or consider both males and females together. Among the males, it is shown here that the classification performance can be improved (by up to 10%) by considering adults and adolescents separately. By using Chi-square algorithm the number of features was reduced drastically to lower than 200 in contrast to the thousands of features that have been used in recent studies.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"40 1","pages":"1-8"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84629439","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}
Brain waves are classified as gamma, beta, alpha, theta, and delta waves to quantify brain activity and can be approximated as sinusoidal waves of different frequencies. In this work, we use sinusoidal waves at two different frequencies to control chaos in a chaotic neural network (CNN) to explore the effect of multi-frequency sinusoidal waves in chaos control. We propose two methods to control chaos. In one, two sinusoidal wave signals are added to different groups of neurons. In the other, a control signal with a mixture of two sinusoidal waves with different frequencies is added to all neurons. The controlling dynamics differ in these two cases. A stable output sequence of the controlled CNN contains only one type of stored pattern and its reversed pattern, which are related to the initial pattern.
{"title":"Multi-frequency sinusoidal wave control in a chaotic neural network","authors":"Guoguang He, Chongchong Wang, Xiaoping Xie, Ping Zhu","doi":"10.1109/IJCNN.2015.7280380","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280380","url":null,"abstract":"Brain waves are classified as gamma, beta, alpha, theta, and delta waves to quantify brain activity and can be approximated as sinusoidal waves of different frequencies. In this work, we use sinusoidal waves at two different frequencies to control chaos in a chaotic neural network (CNN) to explore the effect of multi-frequency sinusoidal waves in chaos control. We propose two methods to control chaos. In one, two sinusoidal wave signals are added to different groups of neurons. In the other, a control signal with a mixture of two sinusoidal waves with different frequencies is added to all neurons. The controlling dynamics differ in these two cases. A stable output sequence of the controlled CNN contains only one type of stored pattern and its reversed pattern, which are related to the initial pattern.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"476 1","pages":"1-6"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"84691066","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 : 2015-07-12DOI: 10.1109/IJCNN.2015.7280805
Akihisa Kato, Hirohito Kawahara, K. Yamauchi
In this study, a lightweight kernel regression algorithm for embedded systems is proposed. In our previous study, we proposed an online learning method with a limited number of kernels based on a kernel regression model known as a limited general regression neural network (LGRNN). The LGRNN behavior is similar to that of k-nearest neighbors except for its continual interpolation between learned samples. The output of kernel regression to an input is dominant for the closest kernel output. This is in contrast to the output of kernel perceptrons, which is determined by the combination of several nested kernels. This means that the output of a kernel regression model can be lightly weighted by omitting calculations for the other kernels. Therefore, we have to find the closest kernel and its neighbors to the current input vector quickly. To realize this, we introduce a tree-search-based calculation method for LGRNN. In the LGRNN learning method, the kernels are clustered into k groups and organized as tree-structured data for the tree-search algorithm.
{"title":"Incremental learning on a budget and a quick calculation method using a tree-search algorithm","authors":"Akihisa Kato, Hirohito Kawahara, K. Yamauchi","doi":"10.1109/IJCNN.2015.7280805","DOIUrl":"https://doi.org/10.1109/IJCNN.2015.7280805","url":null,"abstract":"In this study, a lightweight kernel regression algorithm for embedded systems is proposed. In our previous study, we proposed an online learning method with a limited number of kernels based on a kernel regression model known as a limited general regression neural network (LGRNN). The LGRNN behavior is similar to that of k-nearest neighbors except for its continual interpolation between learned samples. The output of kernel regression to an input is dominant for the closest kernel output. This is in contrast to the output of kernel perceptrons, which is determined by the combination of several nested kernels. This means that the output of a kernel regression model can be lightly weighted by omitting calculations for the other kernels. Therefore, we have to find the closest kernel and its neighbors to the current input vector quickly. To realize this, we introduce a tree-search-based calculation method for LGRNN. In the LGRNN learning method, the kernels are clustered into k groups and organized as tree-structured data for the tree-search algorithm.","PeriodicalId":6539,"journal":{"name":"2015 International Joint Conference on Neural Networks (IJCNN)","volume":"6 1","pages":"1-7"},"PeriodicalIF":0.0,"publicationDate":"2015-07-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"80711597","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：