2022/07/02 更新

写真a

ムラタ ノボル
村田 昇
所属
理工学術院 先進理工学部
職名
教授

兼担

  • 理工学術院   大学院先進理工学研究科

学内研究所等

  • 2021年
    -
    2022年

    データ科学センター   兼任センター員

  • 2020年
    -
    2022年

    リサーチイノベ オープンイノベーション推進セクション   兼任センター員

  • 2020年
    -
    2022年

    理工学術院総合研究所   兼任研究員

学位

  • 東京大学   博士(工学)

所属学協会

  •  
     
     

    日本鉄鋼協会

  •  
     
     

    計測自動制御学会

  •  
     
     

    電子情報通信学会

 

研究キーワード

  • 数理工学

論文

  • Detecting cell assemblies by NMF-based clustering from calcium imaging data

    Mizuo Nagayama, Toshimitsu Aritake, Hideitsu Hino, Takeshi Kanda, Takehiro Miyazaki, Masashi Yanagisawa, Shotaro Akaho, Noboru Murata

    Neural Networks    2022年02月

    DOI

  • Fast and robust multiplane single-molecule localization microscopy using a deep neural network

    Toshimitsu Aritake, Hideitsu Hino, Shigeyuki Namiki, Daisuke Asanuma, Kenzo Hirose, Noboru Murata

    Neurocomputing   451   279 - 289  2021年09月

    DOI

  • Designing Comprehensive Cyber Threat Analysis Platform: Can We Orchestrate Analysis Engines?

    Takeshi Takahashi 0001, Yuki Umemura, Chansu Han, Tao Ban, Keisuke Furumoto, Ohnori Nakamura, Katsunari Yoshioka, Jun'ichi Takeuchi, Noboru Murata, Yoshiaki Shiraishi

    19th IEEE International Conference on Pervasive Computing and Communications Workshops and other Affiliated Events     376 - 379  2021年

    DOI

  • Single-molecule localization by voxel-wise regression using convolutional neural network

    Toshimitsu Aritake, Hideitsu Hino, Shigeyuki Namiki, Daisuke Asanuma, Kenzo Hirose, Noboru Murata

    Results in Optics   1   100019 - 100019  2020年11月

    DOI

  • Automation of Vulnerability Classification from its Description using Machine Learning.

    Masaki Aota, Hideaki Kanehara, Masaki Kubo, Noboru Murata, Bo Sun, Takeshi Takahashi 0001

    IEEE Symposium on Computers and Communications(ISCC)     1 - 7  2020年

    DOI

  • Information Geometry of Modal Linear Regression

    Keishi Sando, Shotaro Akaho, Noboru Murata, Hideitsu Hino

    Information Geometry    2019年07月  [査読有り]

    DOI

  • 非負値行列因子分解を用いたカルシウムイメージングデータからの睡眠状態解析

    永山瑞生, 有竹俊光, 日野英逸, 上田壮志, 宮崎峻弘, 柳沢正史, 赤穂昭太郎, 村田昇

    情報論的学習理論と機械学習研究会 (IBISML)    2019年06月

  • Real-time botnet detection using nonnegative tucker decomposition.

    Hideaki Kanehara, Yuma Murakami, Jumpei Shimamura, Takeshi Takahashi 0001, Daisuke Inoue, Noboru Murata

    Proceedings of the 34th ACM/SIGAPP Symposium on Applied Computing(SAC)     1337 - 1344  2019年

    DOI

  • Transport Analysis of Infinitely Deep Neural Network

    園田 翔, 村田 昇

    Journal of Machine Learning Research   20 ( 2 ) 1 - 52  2019年  [査読有り]

  • Sleep State Analysis using Calcium Imaging Data by Non-negative Matrix Factorization.

    Nagayama, Mizuo, Aritake, Toshimitsu, Hino, Hideitsu, Kanda, Takeshi, Miyazaki, Takehiro, Yanagisawa, Masashi, Akaho, Shotaro, Murata, Noboru

    Artificial Neural Networks and Machine Learning – ICANN 2019: Theoretical Neural Computation.   11727   102 - 113  2019年  [査読有り]

    DOI

  • EEG dipole source localization with information criteria for multiple particle filters.

    Sho Sonoda, Keita Nakamura, Yuki Kaneda, Hideitsu Hino, Shotaro Akaho, Noboru Murata, Eri Miyauchi, Masahiro Kawasaki

    Neural networks : the official journal of the International Neural Network Society   108   68 - 82  2018年12月  [査読有り]  [国際誌]

     概要を見る

    Electroencephalography (EEG) is a non-invasive brain imaging technique that describes neural electrical activation with good temporal resolution. Source localization is required for clinical and functional interpretations of EEG signals, and most commonly is achieved via the dipole model; however, the number of dipoles in the brain should be determined for a reasonably accurate interpretation. In this paper, we propose a dipole source localization (DSL) method that adaptively estimates the dipole number by using a novel information criterion. Since the particle filtering process is nonparametric, it is not clear whether conventional information criteria such as Akaike's information criterion (AIC) and Bayesian information criterion (BIC) can be applied. In the proposed method, multiple particle filters run in parallel, each of which respectively estimates the dipole locations and moments, with the assumption that the dipole number is known and fixed; at every time step, the most predictive particle filter is selected by using an information criterion tailored for particle filters. We tested the proposed information criterion first through experiments on artificial datasets; these experiments supported the hypothesis that the proposed information criterion would outperform both AIC and BIC. We then analyzed real human EEG datasets collected during an auditory short-term memory task using the proposed method. We found that the alpha-band dipoles were localized to the right and left auditory areas during the auditory short-term memory task, which is consistent with previous physiological findings. These analyses suggest the proposed information criterion can work well in both model and real-world situations.

    DOI PubMed

  • Estimationofneuralconnectionsfrompartiallyobservedneuralspikes

    Taishi Iwasaki, Hideitsu Hino, Masami Tatsuno, Shotaro Akaho, Noboru Murata

    Neural Networks   108   172 - 191  2018年08月  [査読有り]

  • Distributed Energy Management for Comprehensive Utilization of Residential Photovoltaic Outputs

    Yu Fujimoto, Hiroshi Kikusato, Shinya Yoshizawa, Shunsuke Kawano, Akira Yoshida, Shinji Wakao, Noboru Murata, Yoshiharu Amano, Shin-ichi Tanabe, Yasuhiro Hayashi

    IEEE TRANSACTIONS ON SMART GRID   9 ( 2 ) 1216 - 1227  2018年03月  [査読有り]

     概要を見る

    The introduction of photovoltaic power systems is being significantly promoted. This paper proposes the implementation of a distributed energy management framework linking demand-side management systems and supply-side management system under the given time-of-use pricing program for efficient utilization of photovoltaic power outputs; each system implements a consistent management flow composed of forecasting, operation planning, and control steps. In our framework, demand-side systems distributed in the electric distribution network manage individual energy consumption to reduce the residential operating cost by utilizing the residential photovoltaic power system and controllable energy appliances so as not to inconvenience residents. On the other hand, the supply-side system utilizes photovoltaic power maximally while maintaining the quality of electric power. The effectiveness of the proposed framework is evaluated on the basis of an actual Japanese distribution network simulation model from both the supply-side and demand-side viewpoints.

    DOI

  • Estimation of neural connections from partially observed neural spikes

    Iwasaki T, Hino H, Tatsuno M, Akaho S, Murata N

    Neural Networks   108   172 - 191  2018年  [査読有り]

    DOI

  • Integral representation of the global minimizer.

    Sho Sonoda, Isao Ishikawa, Masahiro Ikeda, Kei Hagihara, Yoshihiro Sawano, Takuo Matsubara, Noboru Murata

    CoRR   abs/1805.07517  2018年  [査読有り]

  • Neural network with unbounded activation functions is universal approximator

    Sho Sonoda, Noboru Murata

    APPLIED AND COMPUTATIONAL HARMONIC ANALYSIS   43 ( 2 ) 233 - 268  2017年09月  [査読有り]

     概要を見る

    This paper presents an investigation of the approximation property of neural networks with unbounded activation functions, such as the rectified linear unit (ReLU), which is the new de-facto standard of deep learning. The ReLU network can be analyzed by the ridgelet transform with respect to Lizorkin distributions. By showing three reconstruction formulas by using the Fourier slice theorem, the Radon transform, and Parseval's relation, it is shown that a neural network with unbounded activation functions still satisfies the universal approximation property. As an additional consequence, the ridgelet transform, or the backprojection filter in the Radon domain, is what the network learns after backpropagation. Subject to a constructive admissibility condition, the trained network can be obtained by simply discretizing the ridgelet transform, without backpropagation. Numerical examples not only support the consistency of the admissibility condition but also imply that some non-admissible cases result in low-pass filtering. (C) 2015 Elsevier Inc. All rights reserved.

    DOI

  • Local Intrinsic Dimension Estimation by Generalized Linear Modeling

    Hino, Hideitsu, Fujiki, Jun, Akaho, Shotaro, Murata, Noboru

    Neural computation   29 ( 7 ) 1838 - 1878  2017年07月  [査読有り]

     概要を見る

    We propose a method for intrinsic dimension estimation. By fitting the power of distance from an inspection point and the number of samples included inside a ball with a radius equal to the distance, to a regression model, we estimate the goodness of fit. Then, by using the maximum likelihood method, we estimate the local intrinsic dimension around the inspection point. The proposed method is shown to be comparable to conventional methods in global intrinsic dimension estimation experiments. Furthermore, we experimentally show that the proposed method outperforms a conventional local dimension estimation method.

    DOI

  • Local Intrinsic Dimension Estimation by Generalized Linear Modeling

    Hideitsu Hino, Jun Fujiki, Shotaro Akaho, Noboru Murata

    NEURAL COMPUTATION   29 ( 7 ) 1838 - 1878  2017年07月  [査読有り]

     概要を見る

    We propose a method for intrinsic dimension estimation. By fitting the power of distance from an inspection point and the number of samples included inside a ball with a radius equal to the distance, to a regression model, we estimate the goodness of fit. Then, by using the maximum likelihood method, we estimate the local intrinsic dimension around the inspection point. The proposed method is shown to be comparable to conventional methods in global intrinsic dimension estimation experiments. Furthermore, we experimentally show that the proposed method outperforms a conventional local dimension estimation method.

    DOI

  • Double sparsity for multi-frame super resolution

    Toshiyuki Kato, Hideitsu Hino, Noboru Murata

    NEUROCOMPUTING   240   115 - 126  2017年05月  [査読有り]

     概要を見る

    A number of image super resolution algorithms based on the sparse coding have successfully implemented multi-frame super resolution in recent years. In order to utilize multiple low-resolution observations, both accurate image registration and sparse coding are required. Previous study on multi-frame super resolution based on sparse coding firstly apply block matching for image registration, followed by sparse coding to enhance the image resolution. In this paper, these two problems are solved by optimizing a single objective function. The proposed formulation not only has a mathematically interesting structure, called the double sparsity, but also yields comparable or improved numerical performance to conventional methods. (C) 2017 Elsevier B.V. All rights reserved.

    DOI

  • Time-Varying Transition Probability Matrix Estimation and Its Application to Brand Share Analysis

    Tomoaki Chiba, Hideitsu Hino, Shotaro Akaho, Noboru Murata

    PLOS ONE   12 ( 1 )  2017年01月  [査読有り]

     概要を見る

    In a product market or stock market, different products or stocks compete for the same consumers or purchasers. We propose a method to estimate the time-varying transition matrix of the product share using a multivariate time series of the product share. The method is based on the assumption that each of the observed time series of shares is a stationary distribution of the underlying Markov processes characterized by transition probability matrices. We estimate transition probability matrices for every observation under natural assumptions. We demonstrate, on a real-world dataset of the share of automobiles, that the proposed method can find intrinsic transition of shares. The resulting transition matrices reveal interesting phenomena, for example, the change in flows between TOYOTA group and GM group for the fiscal year where TOYOTA group's sales beat GM's sales, which is a reasonable scenario.

    DOI

  • Nonparametric e-Mixture Estimation

    Ken Takano, Hideitsu Hino, Shotaro Akaho, Noboru Murata

    NEURAL COMPUTATION   28 ( 12 ) 2687 - 2725  2016年12月  [査読有り]

     概要を見る

    This study considers the common situation in data analysis when there are few observations of the distribution of interest or the target distribution, while abundant observations are available from auxiliary distributions. In this situation, it is natural to compensate for the lack of data from the target distribution by using data sets from these auxiliary distributionsin other words, approximating the target distribution in a subspace spanned by a set of auxiliary distributions. Mixture modeling is one of the simplest ways to integrate information from the target and auxiliary distributions in order to express the target distribution as accurately as possible. There are two typical mixtures in the context of information geometry: the The is applied in a variety of research fields because of the presence of the well-known expectation-maximazation algorithm for parameter estimation, whereas the e-mixture is rarely used because of its difficulty of estimation, particularly for nonparametric models. The e-mixture, however, is a well-tempered distribution that satisfies the principle of maximum entropy. To model a target distribution with scarce observations accurately, this letter proposes a novel framework for a nonparametric modeling of the e-mixture and a geometrically inspired estimation algorithm. As numerical examples of the proposed framework, a transfer learning setup is considered. The experimental results show that this framework works well for three types of synthetic data sets, as well as an EEG real-world data set.

    DOI

  • Optical detection of neuron connectivity by random access two-photon microscopy

    Nasrin Shafeghat, Morteza Heidarinejad, Noboru Murata, Hideki Nakamura, Takafumi Inoue

    JOURNAL OF NEUROSCIENCE METHODS   263   48 - 56  2016年04月  [査読有り]

     概要を見る

    Background: Knowledge about the distribution, strength, and direction of synaptic connections within neuronal networks are crucial for understanding brain function. Electrophysiology using multiple electrodes provides a very high temporal resolution, but does not yield sufficient spatial information for resolving neuronal connection topology. Optical recording techniques using single-cell resolution have provided promise for providing spatial information. Although calcium imaging from hundreds of neurons has provided a novel view of the neural connections within the network, the kinetics of calcium responses are not fast enough to resolve each action potential event with high fidelity. Therefore, it is not possible to detect the direction of neuronal connections.
    New method: We took advantage of the fast kinetics and large dynamic range of the DiO/DPA combination of voltage sensitive dye and the fast scan speed of a custom-made random-access two-photon microscope to resolve each action potential event from multiple neurons in culture.
    Results: Long-duration recording up to 100 min from cultured hippocampal neurons yielded sufficient numbers of spike events for analyzing synaptic connections. Cross-correlation analysis of neuron pairs clearly distinguished synaptically connected neuron pairs with the connection direction.
    Comparison with existing method: The long duration recording of action potentials with voltage-sensitive dye utilized in the present study is much longer than in previous studies. Simultaneous optical voltage and calcium measurements revealed that voltage-sensitive dye is able to detect firing events more reliably than calcium indicators.
    Conclusions: This novel method reveals a new view of the functional structure of neuronal networks. (C) 2016 Elsevier B.V. All rights reserved.

    DOI

  • Reproducing Statistical Property of Short-term Fluctuation in Wind Power Profiles

    Seigo Furuya, Yu Fujimoto, Noboru Murata, Yasuhiro Hayashi

    10TH INTERNATIONAL RENEWABLE ENERGY STORAGE CONFERENCE, IRES 2016   99   130 - 136  2016年  [査読有り]

     概要を見る

    Unexpected fluctuation of wind power output will become serious problems from the viewpoint of stable supply for an electricity grid. Operating a battery system installed in the grid for mitigating the short-term fluctuation is one of the new approaches for grid stabilization. In this paper, we propose a method of generating synthetic wind power profiles with high temporal resolution for power flow simulation which aims to estimate the impact of wind power fluctuation and specify the required battery system. We numerically show the plausibility of the synthetic wind power profiles from the viewpoints of statistical properties. (C) 2016 The Authors. Published by Elsevier Ltd.

    DOI

  • Non-parametric e-mixture of Density Functions

    Hideitsu Hino, Ken Takano, Shotaro Akaho, Noboru Murata

    NEURAL INFORMATION PROCESSING, ICONIP 2016, PT II   9948   3 - 10  2016年  [査読有り]

     概要を見る

    Mixture modeling is one of the simplest ways to represent complicated probability density functions, and to integrate information from different sources. There are two typical mixtures in the context of information geometry, the m- and e-mixtures. This paper proposes a novel framework of non-parametric e-mixture modeling by using a simple estimation algorithm based on geometrical insights into the characteristics of the e-mixture. An experimental result supports the proposed framework.

    DOI

  • MDL Criterion for NMF with Application to Botnet Detection

    Shoma Tanaka, Yuki Kawamura, Masanori Kawakita, Noboru Murata, Jun'ichi Takeuchi

    NEURAL INFORMATION PROCESSING, ICONIP 2016, PT I   9947   570 - 578  2016年  [査読有り]

     概要を見る

    A method for botnet detection from traffic data of the Internet by the Non-negative Matrix Factorization (NMF) was proposed by (Yamauchi et al. 2012). This method assumes that traffic data is composed by several types of communications, and estimates the number of types in the data by the minimum description length (MDL) criterion. However, consideration on the MDL criterion was not sufficient and validity has not been guaranteed. In this paper, we refine the MDL criterion for NMF and report results of experiments for the new MDL criterion on synthetic and real data.

    DOI

  • Graph structure modeling for multi-neuronal spike data

    Shotaro Akaho, Sho Higuchi, Taishi Iwasaki, Hideitsu Hino, Masami Tatsuno, Noboru Murata

    INTERNATIONAL MEETING ON HIGH-DIMENSIONAL DATA-DRIVEN SCIENCE (HD3-2015)   699 ( 1 )  2016年  [査読有り]

     概要を見る

    We propose a method to extract connectivity between neurons for extracellularly recorded multiple spike trains. The method removes pseudo-correlation caused by propagation of information along an indirect pathway, and is also robust against the influence from unobserved neurons. The estimation algorithm consists of iterations of a simple matrix inversion, which is scalable to large data sets. The performance is examined by synthetic spike data.

    DOI

  • DOUBLY SPARSE STRUCTURE IN IMAGE SUPER RESOLUTION

    Toshiyuki Kato, Hideitsu Hino, Noboru Murata

    2016 IEEE 26TH INTERNATIONAL WORKSHOP ON MACHINE LEARNING FOR SIGNAL PROCESSING (MLSP)   2016-November  2016年  [査読有り]

     概要を見る

    There are a large number of image super resolution algorithms based on the sparse coding, and some algorithms realize multi-frame super resolution. For utilizing multiple low resolution observations, both accurate image registration and sparse coding are required. Previous study on multi-frame super resolution based on sparse coding firstly apply block matching for image registration, followed by sparse coding to enhance the image resolution. In this paper, these two problems are solved by optimizing a single objective function. The proposed formulation not only has a mathematically interesting structure called the double sparsity, but also offers improved numerical performance.

    DOI

  • Change-Point Detection in a Sequence of Bags-of-Data

    Kensuke Koshijima, Hideitsu Hino, Noboru Murata

    2016 32ND IEEE INTERNATIONAL CONFERENCE ON DATA ENGINEERING (ICDE)     1560 - 1561  2016年  [査読有り]

     概要を見る

    In this paper, the limitation that is prominent in most existing works of change-point detection methods is addressed by proposing a nonparametric, computationally efficient method. The limitation is that most works assume that each data point observed at each time step is a single multi-dimensional vector. However, there are many situations where this does not hold. Therefore, a setting where each observation is a collection of random variables, which we call a bag of data, is considered.

    DOI

  • An Entropy Estimator Based on Polynomial Regression with Poisson Error Structure

    Hideitsu Hino, Shotaro Akaho, Noboru Murata

    NEURAL INFORMATION PROCESSING, ICONIP 2016, PT II   9948   11 - 19  2016年  [査読有り]

     概要を見る

    A method for estimating Shannon differential entropy is proposed based on the second order expansion of the probability mass around the inspection point with respect to the distance from the point. Polynomial regression with Poisson error structure is utilized to estimate the values of density function. The density estimates at every given data points are averaged to obtain entropy estimators. The proposed estimator is shown to perform well through numerical experiments for various probability distributions.

    DOI

  • mmpp: A Package for Calculating Similarity and Distance Metrics for Simple and Marked Temporal Point Processes

    Hideitsu Hino, Ken Takano, Noboru Murata

    R JOURNAL   7 ( 2 ) 237 - 248  2015年12月  [査読有り]

     概要を見る

    A simple temporal point process (SPP) is an important class of time series, where the sample realization of the process is solely composed of the times at which events occur. Particular examples of point process data are neuronal spike patterns or spike trains, and a large number of distance and similarity metrics for those data have been proposed. A marked point process (MPP) is an extension of a simple temporal point process, in which a certain vector valued mark is associated with each of the temporal points in the SPP. Analyses of MPPs are of practical importance because instances of MPPs include recordings of natural disasters such as earthquakes and tornadoes. In this paper, we introduce the R package mmpp, which implements a number of distance and similarity metrics for SPPs, and also extends those metrics for dealing with MPPs.

  • Change-Point Detection in a Sequence of Bags-of-Data

    Kensuke Koshijima, Hideitsu Hino, Noboru Murata

    IEEE TRANSACTIONS ON KNOWLEDGE AND DATA ENGINEERING   27 ( 10 ) 2632 - 2644  2015年10月  [査読有り]

     概要を見る

    In this paper, the limitation that is prominent in most existing works of change-point detection methods is addressed by proposing a nonparametric, computationally efficient method. The limitation is that most works assume that each data point observed at each time step is a single multi-dimensional vector. However, there are many situations where this does not hold. Therefore, a setting where each observation is a collection of random variables, which we call a bag of data, is considered. After estimating the underlying distribution behind each bag of data and embedding those distributions in a metric space, the change-point score is derived by evaluating how the sequence of distributions is fluctuating in the metric space using a distance-based information estimator. Also, a procedure that adaptively determines when to raise alerts is incorporated by calculating the confidence interval of the change-point score at each time step. This avoids raising false alarms in highly noisy situations and enables detecting changes of various magnitudes. A number of experimental studies and numerical examples are provided to demonstrate the generality and the effectiveness of our approach with both synthetic and real datasets.

    DOI

  • Non-parametric entropy estimators based on simple linear regression

    Hideitsu Hino, Kensuke Koshijima, Noboru Murata

    COMPUTATIONAL STATISTICS & DATA ANALYSIS   89   72 - 84  2015年09月  [査読有り]

     概要を見る

    Estimators for differential entropy are proposed. The estimators are based on the second order expansion of the probability mass around the inspection point with respect to the distance from the point. Simple linear regression is utilized to estimate the values of density function and its second derivative at a point. After estimating the values of the probability density function at each of the given sample points, by taking the empirical average of the negative logarithm of the density estimates, two entropy estimators are derived. Other entropy estimators which directly estimate entropy by linear regression, are also proposed. The proposed four estimators are shown to perform well through numerical experiments for various probability distributions. (C) 2015 Elsevier B.V. All rights reserved.

    DOI

  • Multi-frame image super resolution based on sparse coding

    Toshiyuki Kato, Hideitsu Hino, Noboru Murata

    NEURAL NETWORKS   66   64 - 78  2015年06月  [査読有り]

     概要を見る

    An image super-resolution method from multiple observation of low-resolution images is proposed. The method is based on sub-pixel accuracy block matching for estimating relative displacements of observed images, and sparse signal representation for estimating the corresponding high-resolution image, where correspondence between high- and low-resolution images are modeled by a certain degradation process. Relative displacements of small patches of observed low-resolution images are accurately estimated by a computationally efficient block matching method. The matching scores of the block matching are used to select a subset of low-resolution patches for reconstructing a high-resolution patch, that is, an adaptive selection of informative low-resolution images is realized. The proposed method is shown to perform comparable or superior to conventional super-resolution methods through experiments using various images. (C) 2015 Elsevier Ltd. All rights reserved.

    DOI

  • Patchworking Multiple Pairwise Distances for Learning with Distance Matrices

    Ken Takano, Hideitsu Hino, Yuki Yoshikawa, Noboru Murata

    LATENT VARIABLE ANALYSIS AND SIGNAL SEPARATION, LVA/ICA 2015   9237   287 - 294  2015年  [査読有り]

     概要を見る

    A classification framework using only a set of distance matrices is proposed. The proposed algorithm can learn a classifier only from a set of distance matrices or similarity matrices, hence applicable to structured data, which do not have natural vector representation such as time series and graphs. Random forest is used to explore ideal feature representation based on the distance between points defined by a set of given distance matrices. The effectiveness of the proposed method is evaluated through experiments with point process data and graph structured data.

    DOI

  • Effects of a MIP start for solving weekly operational planning problem of a residential energy system

    Yoshida A, Yoshikawa J, Murata N, Amano Y

    ICOPE 2015 - International Conference on Power Engineering    2015年  [査読有り]

    DOI

  • Economic evaluations of residential energy systems based on the prediction-operational planning-control method using time-of-use prices

    Ogata R, Yoshida A, Fujimoto Y, Murata N, Wakao S, Tanabe S.-I, Amano Y

    ECOS 2015 - 28th International Conference on Efficiency, Cost, Optimization, Simulation and Environmental Impact of Energy Systems    2015年  [査読有り]

  • Analytical Estimation of the Convergence Point of Populations

    Noboru Murata, Ryuei Nishii, Hideyuki Takagi, Yan Pei

    2015 IEEE CONGRESS ON EVOLUTIONARY COMPUTATION (CEC)     2619 - 2624  2015年  [査読有り]

     概要を見る

    We propose methods of estimating the convergence point for the moving vectors of individuals between generations or evolution paths and show that the estimated convergence point can be useful information for accelerating evolutionary computation (EC). As the first stage of this new approach, we do not combine the proposed methods with EC search in this paper, but rather evaluate how power an individual the the estimated convergence point is by comparing fitness values. Through experimental evaluations, we show that the estimated point can be a powerful elite for unimodal fitness landscapes and that clustering moving vectors according to the aimed points is the next research target for multimodal fitness landscape.

    DOI

  • Advantage of a home energy management systems for PV utilization connected to grid

    Yoshida A, Yoshizawa S, Fujimoto Y, Murata N, Wakao S, Tanabe S.-I, Hayashi Y, Amano Y

    ECOS 2015 - 28th International Conference on Efficiency, Cost, Optimization, Simulation and Environmental Impact of Energy Systems    2015年  [査読有り]

  • Advantage of a home energy management system for PV utilization connected to grid

    Yoshida A, Yoshizawa S, Fujimoto Y, Murata N, Wakao S, Tanabe S.-I, Hayashi Y, Amano Y

    ECOS 2015 - 28th International Conference on Efficiency, Cost, Optimization, Simulation and Environmental Impact of Energy Systems    2015年  [査読有り]

  • Prediction of Molten Steel Temperature in Steel Making Process with Uncertainty by Integrating Gray-Box Model and Bootstrap Filter

    Iftikhar Ahmad, Manabu Kano, Shinji Hasebe, Hiroshi Kitada, Noboru Murata

    JOURNAL OF CHEMICAL ENGINEERING OF JAPAN   47 ( 11 ) 827 - 834  2014年11月  [査読有り]

     概要を見る

    Stable operation of a continuous casting process requires precise control of molten steel temperature in a tundish (TD temp), which is a container used to feed molten steel into an ingot mold. Since TD temp is implicitly controlled by adjusting molten steel temperature in the preceding secondary refining process (RH temp), a model relating TD temp with RH temp is required. This research proposes a procedure to predict the probability distribution of TD temp by integrating a gray-box model and a bootstrap filter to cope with uncertainties of the process. The derived probability distribution is used not only to predict TD temp but also to evaluate the reliability of prediction. The proposed method was validated through its application to real operation data at a steel work, and it was confirmed that the developed model satisfied the requirements for its industrial application.

    DOI

  • A Nonparametric Clustering Algorithm with a Quantile-Based Likelihood Estimator

    Hideitsu Hino, Noboru Murata

    NEURAL COMPUTATION   26 ( 9 ) 2074 - 2101  2014年09月  [査読有り]

     概要を見る

    Clustering is a representative of unsupervised learning and one of the important approaches in exploratory data analysis. By its very nature, clustering without strong assumption on data distribution is desirable. Information-theoretic clustering is a class of clustering methods that optimize information-theoretic quantities such as entropy and mutual information. These quantities can be estimated in a nonparametric manner, and information-theoretic clustering algorithms are capable of capturing various intrinsic data structures. It is also possible to estimate information-theoretic quantities using a data set with sampling weight for each datum. Assuming the data set is sampled from a certain cluster and assigning different sampling weights depending on the clusters, the cluster-conditional information-theoretic quantities are estimated. In this letter, a simple iterative clustering algorithm is proposed based on a nonparametric estimator of the log likelihood for weighted data sets. The clustering algorithm is also derived from the principle of conditional entropy minimization with maximum entropy regularization. The proposed algorithm does not contain a tuning parameter. The algorithm is experimentally shown to be comparable to or outperform conventional nonparametric clustering methods.

    DOI

  • Intrinsic Graph Structure Estimation Using Graph Laplacian

    Atsushi Noda, Hideitsu Hino, Masami Tatsuno, Shotaro Akaho, Noboru Murata

    NEURAL COMPUTATION   26 ( 7 ) 1455 - 1483  2014年07月  [査読有り]

     概要を見る

    A graph is a mathematical representation of a set of variables where some pairs of the variables are connected by edges. Common examples of graphs are railroads, the Internet, and neural networks. It is both theoretically and practically important to estimate the intensity of direct connections between variables. In this study, a problem of estimating the intrinsic graph structure from observed data is considered. The observed data in this study are a matrix with elements representing dependency between nodes in the graph. The dependency represents more than direct connections because it includes influences of various paths. For example, each element of the observed matrix represents a co-occurrence of events at two nodes or a correlation of variables corresponding to two nodes. In this setting, spurious correlations make the estimation of direct connection difficult. To alleviate this difficulty, a digraph Laplacian is used for characterizing a graph. A generative model of this observed matrix is proposed, and a parameter estimation algorithm for the model is also introduced. The notable advantage of the proposed method is its ability to deal with directed graphs, while conventional graph structure estimation methods such as covariance selections are applicable only to undirected graphs. The algorithm is experimentally shown to be able to identify the intrinsic graph structure.

    DOI

  • Gray-box modeling for prediction and control of molten steel temperature in tundish

    Iftikhar Ahmad, Manabu Kano, Shinji Hasebe, Hiroshi Kitada, Noboru Murata

    JOURNAL OF PROCESS CONTROL   24 ( 4 ) 375 - 382  2014年04月  [査読有り]

     概要を見る

    To realize stable production in the steel industry, it is important to control molten steel temperature in a continuous casting process. The present work aims to provide a general framework of gray-box modeling and to develop a gray-box model that predicts and controls molten steel temperature in a tundish (TD temp) with high accuracy. Since the adopted first-principle model (physical model) cannot accurately describe uncertainties such as degradation of ladles, their overall heat transfer coefficient, which is a parameter in the first-principle model, is optimized for each past batch separately, then the parameter is modeled as a function of process variables through a statistical modeling method, random forests. Such a model is termed as a serial gray-box model. Prediction errors of the first-principle model or the serial gray-box model can be compensated by using another statistical model; this approach derives a parallel gray-box model or a combined gray-box model. In addition, the developed gray-box models are used to determine the optimal molten steel temperature in the Ruhrstahl-Heraeus degassing process from the target TD temp, since the continuous casting process has no manipulated variable to directly control TD temp. The proposed modeling and control strategy is validated through its application to real operation data at a steel work. The results show that the combined gray-box model achieves the best performance in prediction and control of TD temp and satisfies the requirement for its industrial application. (C) 2014 Elsevier Ltd. All rights reserved.

    DOI

  • K-Nearest Neighbor Approach for Forecasting Energy Demands Based on Metric Learning

    Yu Fujimoto, Taiki Sugiura, Noboru Murata

    INTERNATIONAL WORK-CONFERENCE ON TIME SERIES (ITISE 2014)     1127 - 1137  2014年  [査読有り]

     概要を見る

    Forecast of the energy demand is an important topic for the realization of effective energy management. In this paper, the authors focus on the K-nearest neighbor approach for forecast of the energy demand pattern, introduce an idea of distance metric learning to select appropriate K-nearest neighbors, and propose some learning frameworks for forecasting. The proposed frameworks are evaluated based on the real-world datasets from the viewpoint of selection accuracy of the true neighbors.

  • Sampling hidden parameters from oracle distribution

    Sho Sonoda, Noboru Murata

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   8681   539 - 546  2014年  [査読有り]

     概要を見る

    A new sampling learning method for neural networks is proposed. Derived from an integral representation of neural networks, an oracle probability distribution of hidden parameters is introduced. In general rigorous sampling from the oracle distribution holds numerical difficulty, a linear-time sampling algorithm is also developed. Numerical experiments showed that when hidden parameters were initialized by the oracle distribution, following backpropagation converged faster to better parameters than when parameters were initialized by a normal distribution. © 2014 Springer International Publishing Switzerland.

    DOI

  • Sampling hidden parameters from oracle distribution

    Sonoda S, Murata N

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   8681 LNCS   539 - 546  2014年  [査読有り]

    DOI

  • An algorithm for directed graph estimation

    Hideitsu Hino, Atsushi Noda, Masami Tatsuno, Shotaro Akaho, Noboru Murata

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   8681   145 - 152  2014年  [査読有り]

     概要を見る

    A problem of estimating the intrinsic graph structure from observed data is considered. The observed data in this study is a matrix with elements representing dependency between nodes in the graph. Each element of the observed matrix represents, for example, co-occurrence of events at two nodes, or correlation of variables corresponding to two nodes. The dependency does not represent direct connections and includes influences of various paths, and spurious correlations make the estimation of direct connection difficult. To alleviate this difficulty, digraph Laplacian is used for characterizing a graph. A generative model of an observed matrix is proposed, and a parameter estimation algorithm for the model is also proposed. The proposed method is capable of dealing with directed graphs, while conventional graph structure estimation methods from an observed matrix are only applicable to undirected graphs. Experimental result shows that the proposed algorithm is able to identify the intrinsic graph structure. © 2014 Springer International Publishing Switzerland.

    DOI

  • An algorithm for directed graph estimation

    Hideitsu Hino, Atsushi Noda, Masami Tatsuno, Shotaro Akaho, Noboru Murata

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   8681   145 - 152  2014年  [査読有り]

     概要を見る

    A problem of estimating the intrinsic graph structure from observed data is considered. The observed data in this study is a matrix with elements representing dependency between nodes in the graph. Each element of the observed matrix represents, for example, co-occurrence of events at two nodes, or correlation of variables corresponding to two nodes. The dependency does not represent direct connections and includes influences of various paths, and spurious correlations make the estimation of direct connection difficult. To alleviate this difficulty, digraph Laplacian is used for characterizing a graph. A generative model of an observed matrix is proposed, and a parameter estimation algorithm for the model is also proposed. The proposed method is capable of dealing with directed graphs, while conventional graph structure estimation methods from an observed matrix are only applicable to undirected graphs. Experimental result shows that the proposed algorithm is able to identify the intrinsic graph structure. © 2014 Springer International Publishing Switzerland.

    DOI

  • A non-parametric maximum entropy clustering

    Hideitsu Hino, Noboru Murata

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   8681   113 - 120  2014年  [査読有り]

     概要を見る

    Clustering is a fundamental tool for exploratory data analysis. Information theoretic clustering is based on the optimization of information theoretic quantities such as entropy and mutual information. Recently, since these quantities can be estimated in non-parametric manner, non-parametric information theoretic clustering gains much attention. Assuming the dataset is sampled from a certain cluster, and assigning different sampling weights depending on the clusters, the cluster conditional information theoretic quantities are estimated. In this paper, a simple clustering algorithm is proposed based on the principle of maximum entropy. The algorithm is experimentally shown to be comparable to or outperform conventional non-parametric clustering methods. © 2014 Springer International Publishing Switzerland.

    DOI

  • A non-parametric maximum entropy clustering

    Hideitsu Hino, Noboru Murata

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   8681   113 - 120  2014年  [査読有り]

     概要を見る

    Clustering is a fundamental tool for exploratory data analysis. Information theoretic clustering is based on the optimization of information theoretic quantities such as entropy and mutual information. Recently, since these quantities can be estimated in non-parametric manner, non-parametric information theoretic clustering gains much attention. Assuming the dataset is sampled from a certain cluster, and assigning different sampling weights depending on the clusters, the cluster conditional information theoretic quantities are estimated. In this paper, a simple clustering algorithm is proposed based on the principle of maximum entropy. The algorithm is experimentally shown to be comparable to or outperform conventional non-parametric clustering methods. © 2014 Springer International Publishing Switzerland.

    DOI

  • Information estimators for weighted observations

    Hideitsu Hino, Noboru Murata

    NEURAL NETWORKS   46   260 - 275  2013年10月  [査読有り]

     概要を見る

    The Shannon information content is a valuable numerical characteristic of probability distributions. The problem of estimating the information content from an observed dataset is very important in the fields of statistics, information theory, and machine learning. The contribution of the present paper is in proposing information estimators, and showing some of their applications. When the given data are associated with weights, each datum contributes differently to the empirical average of statistics. The proposed estimators can deal with this kind of weighted data. Similar to other conventional methods, the proposed information estimator contains a parameter to be tuned, and is computationally expensive. To overcome these problems, the proposed estimator is further modified so that it is more computationally efficient and has no tuning parameter. The proposed methods are also extended so as to estimate the cross-entropy, entropy, and Kullback-Leibler divergence. Simple numerical experiments show that the information estimators work properly. Then, the estimators are applied to two specific problems, distribution-preserving data compression, and weight optimization for ensemble regression. (C) 2013 Elsevier Ltd. All rights reserved.

    DOI

  • Learning Ancestral Atom via Sparse Coding

    Toshimitsu Aritake, Hideitsu Hino, Noboru Murata

    IEEE JOURNAL OF SELECTED TOPICS IN SIGNAL PROCESSING   7 ( 4 ) 586 - 594  2013年08月  [査読有り]

     概要を見る

    Sparse signal models have been the focus of recent research. In sparse coding, signals are represented with a linear combination of a small number of elementary signals called atoms, and the collection of atoms is called a dictionary. Design of the dictionary has strong influence on the signal approximation performance. Recently, to put prior information into dictionary learning, several methods imposing a certain kind of structure on the dictionary are proposed. In this paper, like wavelet analysis, a dictionary for sparse signal representation is assumed to be generated from an ancestral atom, and a method for learning the ancestral atom is proposed. The proposed algorithm updates the ancestral atom by iterating dictionary update in unstructured dictionary space and projection of the updated dictionary onto the structured dictionary space. The algorithm allows a simple differential geometric interpretation. Numerical experiments are performed to show the characteristics and advantages of the proposed algorithm.

    DOI

  • データ対の直線性に基づく画像上の類似度の定義 : 歪曲画像からの直線検出への応用 (パターン認識・メディア理解)

    日野 英逸, 藤木 淳, 赤穂 昭太郎, 望月 義彦, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   113 ( 75 ) 29 - 34  2013年06月

     概要を見る

    データのクラスタリングは情報処理における最も重要な要素技術の一つであり, 画像処理においても応用が多い.クラスタリングの結果は,データ間に定義される距離あるいは類似度によって大きく左右される. 本稿では,魚眼レンズを用いたカメラや全方位カメラといったデバイスで撮影された歪曲画像からの直線成分の検出のための,データ間類似度の計算手法を提案する.提案手法と既存のクラスタリング手法及び直線・曲線検出手法を,人工データ及び実データを用いて比較する.

    CiNii

  • Regions of Unusually High Flexibility Occur Frequently in Human Genomic DNA

    Hajime Kimura, Dai Kageyama, Mika Furuya, Shigeru Sugiyama, Noboru Murata, Takashi Ohyama

    BIOSCIENCE BIOTECHNOLOGY AND BIOCHEMISTRY   77 ( 3 ) 612 - 617  2013年03月  [査読有り]

     概要を見る

    Remarkable progress has been made in genome science during the past decade, but understanding of genomes of eukaryotes is far from complete. We have created DNA flexibility maps of the human, mouse, fruit fly, and nematode chromosomes. The maps revealed that all of these chromosomes have markedly flexible DNA regions (We named them SPIKEs). SPIKEs occur more frequently in the human chromosomes than in the mouse, fruit fly, and nematode chromosomes. Markedly rigid DNA regions (rSPIKEs) are also present in these chromosomes. The ratio of the number of SPIKEs to the total number of SPIKEs and rSPIKEs correlated positively with evolutionary stage among the organisms. Repetitive DNA sequences with flexible and rigid properties contribute to the formation of SPIKEs and rSPIKEs respectively. However, non-repetitive flexible and rigid sequences appear to play a major role in SPIKE and rSPIKE formation respectively. They might be involved in the genome-folding mechanism of eukaryotes.

    DOI

  • Pairwise similarity for line extraction from distorted images

    Hideitsu Hino, Jun Fujiki, Shotaro Akaho, Yoshihiko Mochizuki, Noboru Murata

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   8048 ( 2 ) 250 - 257  2013年  [査読有り]

     概要を見る

    Clustering a given set of data is crucial in many fields including image processing. It plays important roles in image segmentation and object detection for example. This paper proposes a framework of building a similarity matrix for a given dataset, which is then used for clustering the dataset. The similarity between two points are defined based on how other points distribute around the line connecting the two points. It can capture the degree of how the two points are placed on the same line. The similarity matrix is considered as a kernel matrix of the given dataset, and based on it, the spectral clustering is performed. Clustering with the proposed similarity matrix is shown to perform well through experiments using an artificially designed problem and a real-world problem of detecting lines from a distorted image. © 2013 Springer-Verlag.

    DOI

  • High-Performance Prediction of Molten Steel Temperature in Tundish through Gray-Box Model

    Toshinori Okura, Iftikhar Ahmad, Manabu Kano, Shinji Hasebe, Hiroshi Kitada, Noboru Murata

    ISIJ INTERNATIONAL   53 ( 1 ) 76 - 80  2013年  [査読有り]

     概要を見る

    A novel gray-box model is proposed to estimate molten steel temperature in a continuous casting process at a steel making plant by combining a first-principle model and a statistical model. The first-principle model was developed on the basis of computational fluid dynamics (CFD) simulations to simplify the model and to improve estimation accuracy. Since the derived first-principle model was not able to estimate the molten steel temperature in the tundish with sufficient accuracy, statistical models were developed to estimate the estimation errors of the first-principle model through partial least squares (PLS) and random forest (RF). As a result of comparing the three models, i.e., the first-principle model, the PLS-based gray-box model, and the RF-based gray-box model, the RF-based gray-box model achieved the best estimation performance. Thus, the molten steel temperature in the tundish can be estimated with accuracy by adding estimates of the first-principle model and those of the statistical RF model. The proposed gray-box model was applied to the real process data and the results demonstrated its advantage over other models.

    DOI

  • Entropy-based sliced inverse regression

    Hideitsu Hino, Keigo Wakayama, Noboru Murata

    Computational Statistics and Data Analysis   67   105 - 114  2013年  [査読有り]

     概要を見る

    Abstract The importance of dimension reduction has been increasing according to the growth of the size of available data in many fields. An appropriate dimension reduction method of raw data helps to reduce computational time and to expose the intrinsic structure of complex data. Sliced inverse regression is a well-known dimension reduction method for regression, which assumes an elliptical distribution for the explanatory variable, and ingeniously reduces the problem of dimension reduction to a simple eigenvalue problem. Sliced inverse regression is based on the strong assumptions on the data distribution and the form of regression function, and there are a number of methods to relax or remove these assumptions to extend the applicability of the inverse regression method. However, each method is known to have its drawbacks either theoretically or empirically. To alleviate drawbacks in the existing methods, a dimension reduction method for regression based on the notion of conditional entropy minimization is proposed. Using entropy as a measure of dispersion of data, a low dimensional subspace is estimated without assuming any specific distribution nor any regression function. The proposed method is shown to perform comparable or superior to the conventional methods through experiments using artificial and real-world datasets. © 2013 Elsevier B.V. All rights reserved.

    DOI

  • Design of inner and outer gray-box models to predict molten steel temperature in Tundish

    Iftikhar Ahmad, Manabu Kano, Shinji Hasebe, Hiroshi Kitada, Noboru Murata

    IFAC Proceedings Volumes (IFAC-PapersOnline)   10 ( PART 1 ) 744 - 749  2013年  [査読有り]

     概要を見る

    In order to realize stable production in the steel industry, it is important to control molten steel temperature in a continuous casting process. The present work aims to develop a gray-box model that predicts the molten steel temperature in the tundish (TD temp). In the proposed approach, two parameters in the first-principle model, i.e., overall heat transfer coefficients of ladle and tundish, are optimized for each past batch separately, then the relationship between the two parameters and measured process variables is modeled through random forests (RF). In this inner gray-box model, the statistical models update the physical parameters according to the operating condition. To enhance the accuracy of TD temp estimation, another RF model is developed which compensates errors of the inner gray-box. The proposed approach was validated through its application to real operation data at a steel work. © IFAC.

    DOI

  • A versatile clustering method for electricity consumption pattern analysis in households

    Hideitsu Hino, Haoyang Shen, Noboru Murata, Shinji Wakao, Yasuhiro Hayashi

    IEEE Transactions on Smart Grid   4 ( 2 ) 1048 - 1057  2013年  [査読有り]

     概要を見る

    Analysis and modeling of electric energy demand is indispensable for power planning, operation, facility investment, and urban planning. Because of recent development of renewable energy generation systems oriented for households, there is also a great demand for analysing the electricity usage and optimizing the way to install electricity generation systems for each household. In this study, employing statistical techniques, a method to model daily consumption patterns in households and a method to extract a small number of their typical patterns are presented. The electricity consumption patterns in a household is modeled by a mixture of Gaussian distributions. Then, using the symmetrized generalized Kullback-Leibler divergence as a distance measure of the distributions, typical patterns of the consumption are extracted by means of hierarchical clustering. The statistical modeling of daily consumption patterns allows us to capture essential similarities of the patterns. By experiments using a large-scale dataset including about 500 houses' consumption records in a suburban area in Japan, it is shown that the proposed method is able to extract typical consumption patterns. © 2010-2012 IEEE.

    DOI

  • A comparison of optimal operation of a residential fuel cell co-generation system using clustered demand patterns based on Kullback-Leibler divergence

    Akira Yoshida, Yoshiharu Amano, Noboru Murata, Koichi Ito, Takumi Hasizume

    Energies   6 ( 1 ) 374 - 399  2013年  [査読有り]

     概要を見る

    When evaluating residential energy systems like co-generation systems, hot water and electricity demand profiles are critical. In this paper, the authors aim to extract basic time-series demand patterns from two kinds of measured demand (electricity and domestic hot water), and also aim to reveal effective demand patterns for primary energy saving. Time-series demand data are categorized with a hierarchical clustering method using a statistical pseudo-distance, which is represented by the generalized Kullback-Leibler divergence of two Gaussian mixture distributions. The classified demand patterns are built using hierarchical clustering and then a comparison is made between the optimal operation of a polymer electrolyte membrane fuel cell co-generation system and the operation of a reference system (a conventional combination of a condensing gas boiler and electricity purchased from the grid) using the appropriately built demand profiles. Our results show that basic demand patterns are extracted by the proposed method, and the heat-to-power ratio of demand, the amount of daily demand, and demand patterns affect the primary energy saving of the co-generation system. © 2013 by the authors
    licensee MDPI, Basel, Switzerland.

    DOI

  • Multiple Kernel Learning with Gaussianity Measures

    Hideitsu Hino, Nima Reyhani, Noboru Murata

    NEURAL COMPUTATION   24 ( 7 ) 1853 - 1881  2012年07月  [査読有り]

     概要を見る

    Kernel methods are known to be effective for nonlinear multivariate analysis. One of the main issues in the practical use of kernel methods is the selection of kernel. There have been a lot of studies on kernel selection and kernel learning. Multiple kernel learning (MKL) is one of the promising kernel optimization approaches. Kernel methods are applied to various classifiers including Fisher discriminant analysis (FDA). FDA gives the Bayes optimal classification axis if the data distribution of each class in the feature space is a gaussian with a shared covariance structure. Based on this fact, an MKL framework based on the notion of gaussianity is proposed. As a concrete implementation, an empirical characteristic function is adopted to measure gaussianity in the feature space associated with a convex combination of kernel functions, and two MKL algorithms are derived. From experimental results on some data sets, we show that the proposed kernel learning followed by FDA offers strong classification power.

    DOI

  • Sliced Inverse Regression with Conditional Entropy Minimization

    Hideitsu Hino, Keigo Wakayama, Noboru Murata

    2012 21ST INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR 2012)     1185 - 1188  2012年  [査読有り]

     概要を見る

    An appropriate dimension reduction of raw data helps to reduce computational time and to reveal the intrinsic structure of complex data. In this paper, a dimension reduction method for regression is proposed. The method is based on the well-known sliced inverse regression and conditional entropy minimization. Using entropy as a measure of dispersion of data distribution, dimension reduction subspace is estimated without assuming regression function form nor data distribution, unlike conventional sliced inverse regression. The proposed method is shown to perform well compared to some conventional methods through experiments using both artificial and real-world data sets.

  • Sensitivity analysis for controlling molten steel temperature in Tundish

    Noboru Murata, Sho Sonoda, Hideitsu Hino, Hiroshi Kitada, Manabu Kano

    IFAC Proceedings Volumes (IFAC-PapersOnline)   45 ( 23 ) 270 - 271  2012年  [査読有り]

     概要を見る

    Controlling temperature of molten steel is crucial for product quality in continuous casting. In this paper, sensitivity analysis is carried out on a statistical model for predicting temperature in tundish, and influential operations for controlling temperature are identified. © 2012 IFAC.

    DOI

  • Robust Hypersurface Fitting Based on Random Sampling Approximations

    Jun Fujiki, Shotaro Akaho, Hideitsu Hino, Noboru Murata

    NEURAL INFORMATION PROCESSING, ICONIP 2012, PT III   7665 ( PART 3 ) 520 - 527  2012年  [査読有り]

     概要を見る

    This paper considers N -1-dimensional hypersurface fitting based on L-2 distance in N-dimensional input space. The problem is usually reduced to hyperplane fitting in higher dimension. However, because feature mapping is generally a nonlinear mapping, it does not preserve the order of lengthes, and this derives an unacceptable fitting result. To avoid it, JNLPCA is introduced. JNLPCA defines the L-2 distance in the feature space as a weighted L-2 distance to reflect the metric in the input space. In the fitting, random sampling approximation of least k-th power deviation, and least alpha-percentile of squares are introduced to make estimation robust. The proposed hypersurface fitting method is evaluated by quadratic curve fitting and quadratic curve segments extraction from artificial data and a real image.

    DOI

  • Nonnegative matrix factorization via generalized product rule and its application for classification

    Yu Fujimoto, Noboru Murata

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   7191   263 - 271  2012年  [査読有り]

     概要を見る

    Nonnegative Matrix Factorization (NMF) is broadly used as a mathematical tool for processing tasks of tabulated data. In this paper, an extension of NMF based on a generalized product rule, defined with a nonlinear one-parameter function and its inverse, is proposed. From a viewpoint of subspace methods, the extended NMF constructs flexible subspaces which plays an important role in classification tasks. Experimental results on benchmark datasets show that the proposed extension improves classification accuracies. © 2012 Springer-Verlag.

    DOI

  • Gray-box model to control molten steel temperature in Tundish

    S. Sakashita, T. Okura, I. Ahmad, M. Kano, H. Kitada, N. Murata

    IFAC Proceedings Volumes (IFAC-PapersOnline)   45 ( 23 ) 268 - 269  2012年  [査読有り]

     概要を見る

    Controlling molten steel temperature in a tundish is crucial for stable and efficient production of steel products. In this research, a gray-box model is developed by combining a first-principle model and a statistical model. The first-principle model predicts the molten steel temperature in the tundish, and the statistical model predicts the prediction error of the first-principle model. The derived model was then used to derive the molten steel temperature in the Ruhrstahl-Heraeus process at the end of its operation from the target molten steel temperature in the tundish and other process variables. The result of applying the proposed model to real operation data has demonstrated its practicability. © 2012 IFAC.

    DOI

  • Automatic Extraction of Basic Electricity Consumption Patterns in Household

    Haoyang Shen, Hideitsu Hino, Noboru Murata, Shinji Wakao, Yasuhiro Hayashi

    INTERNATIONAL CONFERENCE ON RENEWABLE ENERGY RESEARCH AND APPLICATIONS (ICRERA)    2012年  [査読有り]

     概要を見る

    Electricity consumption in households varies dependent on a lot of possible reasons such as lifestyle, family configuration, and weather. It is of great importance to optimize the electricity generation system to install for each household. In our previous work, we proposed a clustering approach for extracting a small number of basic electricity consumption patterns in a household. In this study, we apply the method to a larger dataset with many households. In the previous work, we determined the number of basic patterns in a heuristic manner. In this work, we use gap statistics to automatically determine an appropriate number of basic patterns, and we obtained a reasonable result on a large-scale data.

    DOI

  • A tree search approach to sparse coding

    Rui Rei, João P. Pedroso, Hideitsu Hino, Noboru Murata

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   7219   472 - 477  2012年  [査読有り]

     概要を見る

    Sparse coding is an important optimization problem with numerous applications. In this paper, we describe the problem and the commonly used pursuit methods, and propose a best-first tree search algorithm employing multiple queues for unexplored tree nodes. We assess the effectiveness of our method in an extensive computational experiment, showing its superiority over other methods even for modest computational time. © 2012 Springer-Verlag.

    DOI

  • A Statistical Model for Predicting the Liquid Steel Temperature in Ladle and Tundish by Bootstrap Filter

    Sho Sonoda, Noboru Murata, Hideitsu Hino, Hiroshi Kitada, Manabu Kano

    ISIJ INTERNATIONAL   52 ( 6 ) 1086 - 1091  2012年  [査読有り]

     概要を見る

    A statistical model for predicting the liquid steel temperature in the ladle and in the tundish is developed. Given a large data set in a steelmaking process, the proposed model predicts the temperature in a seconds with a good accuracy. The data are divided into four phases at the mediation of five temperature measurements: before tapping from the converter (CV), after throwing ferroalloys into the ladle, before and after the Ruhrstahl-Heraeus (RH) processing, and after casting into the tundish in the continuous casting (CC) machine. Based on the general state space modeling, the bootstrap filter predicts the temperature phase by phase. The particle approximation technique enables to compute general-shaped probability distributions. The proposed model gives a prediction not as a point but as a probability distribution, or a predictive distribution. It evaluates both uncertainty of the prediction and ununiformity of the temperature. It is applicable to sensitivity analysis, process scheduling and temperature control.

    DOI

  • A generalisation of independence in statistical models for categorical distribution

    Yu Fujimoto, Noboru Murata

    International Journal of Data Mining, Modelling and Management   4 ( 2 ) 172 - 187  2012年  [査読有り]

     概要を見る

    In this paper, generalised statistical independence in statistical models for categorical distributions is proposed from the viewpoint of generalised multiplication characterised by a monotonically increasing function and its inverse function, and it is implemented in naive Bayes models. This paper also proposes an idea of their estimation method which directly uses empirical marginal distributions to retain simplicity of calculation. This method is interpreted as an optimisation of a rough approximation of the Bregman divergence so that it is expected to have a kind of robust property. Effectiveness of proposed models is shown by numerical experiments on some benchmark datasets. Copyright © 2012 Inderscience Enterprises Ltd.

    DOI

  • A comparison of optimal operation of residential energy systems using clustered demand patterns based on kullback-leibler divergence

    Yoshida A, Amano Y, Murata N, Ito K, Hashizume T

    Proceedings of the 25th International Conference on Efficiency, Cost, Optimization and Simulation of Energy Conversion Systems and Processes, ECOS 2012   3   1 - 16  2012年  [査読有り]

  • An Estimation of Generalized Bradley-Terry Models Based on the em Algorithm

    Yu Fujimoto, Hideitsu Hino, Noboru Murata

    NEURAL COMPUTATION   23 ( 6 ) 1623 - 1659  2011年06月  [査読有り]

     概要を見る

    The Bradley-Terry model is a statistical representation for one's preference or ranking data by using pairwise comparison results of items. For estimation of the model, several methods based on the sum of weighted Kullback-Leibler divergences have been proposed from various contexts. The purpose of this letter is to interpret an estimation mechanism of the Bradley-Terry model from the viewpoint of flatness, a fundamental notion used in information geometry. Based on this point of view, a new estimation method is proposed on a framework of the em algorithm. The proposed method is different in its objective function from that of conventional methods, especially in treating unobserved comparisons, and it is consistently interpreted in a probability simplex. An estimation method with weight adaptation is also proposed from a viewpoint of the sensitivity. Experimental results show that the proposed method works appropriately, and weight adaptation improves accuracy of the estimate.

    DOI

  • Speaker verification robust to talking style variation using multiple kernel learning based on conditional entropy minimization

    Tetsuji Ogawa, Hideitsu Hino, Noboru Murata, Tetsunori Kobayashi

    12TH ANNUAL CONFERENCE OF THE INTERNATIONAL SPEECH COMMUNICATION ASSOCIATION 2011 (INTERSPEECH 2011), VOLS 1-5     2752 - +  2011年  [査読有り]

     概要を見る

    We developed a new speaker verification system that is robust to intra-speaker variation. There is a strong likelihood that intra-speaker variations will occur due to changes in talking styles, the periods when an individual speaks, and so on. It is well known that such variation generally degrades the performance of speaker verification systems. To solve this problem, we applied multiple kernel learning (MKL) based on conditional entropy minimization, which impose the data to be compactly aggregated for each speaker class and ensure that the different speaker classes were far apart from each other. Experimental results showed that the proposed speaker verification system achieved a robust performance to intra-speaker variation derived from changes in the talking styles compared to the conventional maximum margin-based system.

  • mn SPEAKER RECOGNITION USING MULTIPLE KERNEL LEARNING BA SED ON CONDITIONA L ENTROPY MINIMIZATION

    Tetsuji Ogawa, Hideitsu Hino, Nima Reyhani, Noboru Murata, Tetsunori Kobayashi

    2011 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING     2204 - 2207  2011年  [査読有り]

     概要を見る

    We applied a multiple kernel learning (MKL) method based on information-theoretic optimization to speaker recognition. Most of the kernel methods applied to speaker recognition systems require a suitable kernel function and its parameters to be determined for a given data set. In contrast, MKL eliminates the need for strict determination of the kernel function and parameters by using a convex combination of element kernels. In the present paper, we describe an MKL algorithm based on conditional entropy minimization (MCEM). We experimentally verified the effectiveness of MCEM for speaker classification; this method reduced the speaker error rate as compared to conventional methods.

    DOI

  • Robust hyperplane fitting based on k-th power deviation and α-quantile

    Fujiki J, Akaho S, Hino H, Murata N

    Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)   6854 ( PART 1 ) 278 - 285  2011年  [査読有り]

    DOI

  • Improved methods for dewarping images in convex mirrors in fine art: Applications to van Eyck and Parmigianino

    Yumi Usami, David G. Stork, Jun Fujiki, Hideitsu Hino, Shotaro Akaho, Noboru Murata

    COMPUTER VISION AND IMAGE ANALYSIS OF ART II   7869  2011年  [査読有り]

     概要を見る

    We derive and demonstrate new methods for dewarping images depicted in convex mirrors in artwork and for estimating the three-dimensional shapes of the mirrors themselves. Previous methods were based on the assumption that mirrors were spherical or paraboloidal, an assumption unlikely to hold for hand-blown glass spheres used in early Renaissance art, such as Johannes van Eyck's Portrait of Giovanni (?) Arnolfini and his wife (1434) and Robert Campin's Portrait of St. John the Baptist and Heinrich von Werl (1438). Our methods are more general than such previous methods in that we assume merely that the mirror is radially symmetric and that there are straight lines (or colinear points) in the actual source scene. We express the mirror's shape as a mathematical series and pose the image dewarping task as that of estimating the coefficients in the series expansion. Central to our method is the plumbline principle: that the optimal coefficients are those that dewarp the mirror image so as to straighten lines that correspond to straight lines in the source scene. We solve for these coefficients algebraically through principal component analysis, PCA. Our method relies on a global figure of merit to balance warping errors throughout the image and it thereby reduces a reliance on the somewhat subjective criterion used in earlier methods. Our estimation can be applied to separate image annuli, which is appropriate if the mirror shape is irregular. Once we have found the optimal image dewarping, we compute the mirror shape by solving a differential equation based on the estimated dewarping function. We demonstrate our methods on the Arnolfini mirror and reveal a dewarped image superior to those found in prior work-an image noticeably more rectilinear throughout and having a more coherent geometrical perspective and vanishing points. Moreover, we find the mirror deviated from spherical and paraboloidal shape; this implies that it would have been useless as a concave projection mirror, as has been claimed. Our dewarped image can be compared to the geometry in the full Arnolfini painting; the geometrical agreement strongly suggests that van Eyck worked from an actual room, not, as has been suggested by some art historians, a "fictive" room of his imagination. We apply our method to other mirrors depicted in art, such as Parmigianino's Self-portrait in a convex mirror and compare our results to those from earlier computer graphics simulations.

    DOI

  • Extraction of basic patterns of household energy consumption

    Haoyang Shen, Hideitsu Hino, Noboru Murata, Shinji Wakao

    Proceedings - 10th International Conference on Machine Learning and Applications, ICMLA 2011   2   275 - 280  2011年  [査読有り]

     概要を見る

    Solar power, wind power, and co-generation (combined heat and power) systems are possible candidate for household power generation. These systems have their advantages and disadvantages. To propose the optimal combination of the power generation systems, the extraction of basic patterns of energy consumption of the house is required. In this study, energy consumption patterns are modeled by mixtures of Gaussian distributions. Then, using the symmetrized Kullback-Leibler divergence as a distance measure of the distributions, the basic pattern of energy consumption is extracted by means of hierarchical clustering. By an experiment using the Annex 42 dataset, it is shown that the proposed method is able to extract typical energy consumption patterns. © 2011 IEEE.

    DOI

  • Calibration of radially symmetric distortion based on linearity in the calibrated image

    Jun Fujiki, Hideitsu Hino, Shotaro Akaho, Noboru Murata

    2011 IEEE INTERNATIONAL CONFERENCE ON COMPUTER VISION WORKSHOPS (ICCV WORKSHOPS)     288 - 295  2011年  [査読有り]

     概要を見る

    For calibration of general radially symmetric distortion of omnidirectional cameras such as fish-eye lenses, calibration parameters are usually estimated so that curved lines, which are supposed to be straight in the real-world, are mapped to straight lines in the calibrated image, which is called plumbline principle. Under the principle, the camera with radially symmetric distortion can be calibrated by at least one distorted line in a image, theoretically, and the calibrated image is equivalent to the image taken by an ideal pin-hole camera. In this paper, the method to optimize the calibration parameters by maximizing the sum of the straightness, which is invariant under translation, rotation and magnification (scaling), of distorted lines on calibrated image is proposed. The performance of the proposed method is evaluated by artificial data and a real image.

    DOI

  • A measure of credibility of solar power prediction

    Haoyang Shen, Hideitsu Hino, Noboru Murata, Shinji Wakao

    Proceedings - 10th International Conference on Machine Learning and Applications, ICMLA 2011   2   286 - 291  2011年  [査読有り]

     概要を見る

    Recently, remarkable developments of new energy technologies have been achieved against various energy problems. Photovoltaic (PV) system, one of such technologies, has an advantage of utilizing infinite and clean energy. On the contrary, it also has a disadvantage of unreliable power supply mainly caused by unstable weather. The fluctuation of the power supply of PV systems are considerably large because of rapid insulation changes and rapid weather changes, and in some cases, it seems impossible to realize high-accuracy prediction even with sophisticated prediction models. In this paper, using recently proposed estimator for the Shannon information content, a method to output a measure of credibility for prediction is proposed. With the proposed method, it is possible to judge whether the energy supply at a certain future time is unpredictably fluctuate compared to the current value or not, and it is possible to take measures against the rapid change of solar energy generation in advance. From an experimental result using solar energy supply data, we see that the proposed measure of credibility reflects the difficulty of predicting solar energy supply. © 2011 IEEE.

    DOI

  • A Computationally Efficient Information Estimator for Weighted Data

    Hideitsu Hino, Noboru Murata

    ARTIFICIAL NEURAL NETWORKS AND MACHINE LEARNING - ICANN 2011, PT II   6792 ( PART 2 ) 301 - 308  2011年  [査読有り]

     概要を見る

    The Shannon information content is a fundamental quantity and it is of great importance to estimate it from observed dataset in the field of statistics, information theory, and machine learning. In this study, an estimator for the information content using a given set of weighted data is proposed. The empirical data distribution varies depending on the weight. The notable features of the proposed estimator are its computational efficiency and its ability to deal with weighted data. The proposed estimator is extended in order to estimate cross entropy, entropy and KL divergence with weighted data. Then, the estimators are applied to classification with one-class samples, and distribution preserving data compression problems.

    DOI

  • Stochastic simulation of biological reactions, and its applications for studying actin polymerization

    Kazuhisa Ichikawa, Takashi Suzuki, Noboru Murata

    PHYSICAL BIOLOGY   7 ( 4 )  2010年12月  [査読有り]

     概要を見る

    Molecular events in biological cells occur in local subregions, where the molecules tend to be small in number. The cytoskeleton, which is important for both the structural changes of cells and their functions, is also a countable entity because of its long fibrous shape. To simulate the local environment using a computer, stochastic simulations should be run. We herein report a new method of stochastic simulation based on random walk and reaction by the collision of all molecules. The microscopic reaction rate P-r is calculated from the macroscopic rate constant k. The formula involves only local parameters embedded for each molecule. The results of the stochastic simulations of simple second-order, polymerization, Michaelis-Menten-type and other reactions agreed quite well with those of deterministic simulations when the number of molecules was sufficiently large. An analysis of the theory indicated a relationship between variance and the number of molecules in the system, and results of multiple stochastic simulation runs confirmed this relationship. We simulated Ca2+ dynamics in a cell by inward flow from a point on the cell surface and the polymerization of G-actin forming F-actin. Our results showed that this theory and method can be used to simulate spatially inhomogeneous events.

    DOI

  • A Conditional Entropy Minimization Criterion for Dimensionality Reduction and Multiple Kernel Learning

    Hideitsu Hino, Noboru Murata

    NEURAL COMPUTATION   22 ( 11 ) 2887 - 2923  2010年11月  [査読有り]

     概要を見る

    Reducing the dimensionality of high-dimensional data without losing its essential information is an important task in information processing. When class labels of training data are available, Fisher discriminant analysis (FDA) has been widely used. However, the optimality of FDA is guaranteed only in a very restricted ideal circumstance, and it is often observed that FDA does not provide a good classification surface for many real problems. This letter treats the problem of supervised dimensionality reduction from the viewpoint of information theory and proposes a framework of dimensionality reduction based on class-conditional entropy minimization. The proposed linear dimensionality-reduction technique is validated both theoretically and experimentally. Then, through kernel Fisher discriminant analysis (KFDA), the multiple kernel learning problem is treated in the proposed framework, and a novel algorithm, which iteratively optimizes the parameters of the classification function and kernel combination coefficients, is proposed. The algorithm is experimentally shown to be comparable to or outperforms KFDA for large-scale benchmark data sets, and comparable to other multiple kernel learning techniques on the yeast protein function annotation task.

    DOI

  • A Grouped Ranking Model for Item Preference Parameter

    Hideitsu Hino, Yu Fujimoto, Noboru Murata

    NEURAL COMPUTATION   22 ( 9 ) 2417 - 2451  2010年09月  [査読有り]

     概要を見る

    Given a set of rating data for a set of items, determining preference levels of items is a matter of importance. Various probability models have been proposed to solve this task. One such model is the Plackett-Luce model, which parameterizes the preference level of each item by a real value. In this letter, the Plackett-Luce model is generalized to cope with grouped ranking observations such as movie or restaurant ratings. Since it is difficult to maximize the likelihood of the proposed model directly, a feasible approximation is derived, and the em algorithm is adopted to find the model parameter by maximizing the approximate likelihood which is easily evaluated. The proposed model is extended to a mixture model, and two applications are proposed. To show the effectiveness of the proposed model, numerical experiments with real-world data are carried out.

    DOI

  • Self-calibration of radially symmetric distortion by model selection

    Jun Fujiki, Hideitsu Hino, Yumi Usami, Shotaro Akaho, Noboru Murata

    Proceedings - International Conference on Pattern Recognition     1812 - 1815  2010年  [査読有り]

     概要を見る

    For self-calibration of general radially symmetric distortion (RSD) of omnidirectional cameras such as fish-eye lenses, calibration parameters are usually estimated so that curved lines, which are supposed to be straight in the real-world, are mapped to straight lines in the calibrated image, which is assumed to be taken by an ideal pin-hole camera. In this paper, a method of calibrating RSD is introduced base on the notion of principal component analysis (PCA). In the proposed method, the distortion function, which maps a distorted image to an ideal pin-hole camera image, is assumed to be a linear combination of a certain class of basis functions, and an algorithm for solving its coefficients by using line patterns is given. Then a method of selecting good basis functions is proposed, which aims to realize appropriate calibration in practice. Experimental results for synthetic data and real images are presented to demonstrate the performance of our calibration method. © 2010 IEEE.

    DOI

  • Multiple kernel learning by conditional entropy minimization

    Hideitsu Hino, Nima Reyhani, Noboru Murata

    Proceedings - 9th International Conference on Machine Learning and Applications, ICMLA 2010     223 - 228  2010年  [査読有り]

     概要を見る

    Kernel methods have been successfully used in many practical machine learning problems. Choosing a suitable kernel is left to the practitioner. A common way to an automatic selection of optimal kernels is to learn a linear combination of element kernels. In this paper, a novel framework of multiple kernel learning is proposed based on conditional entropy minimization criterion. For the proposed framework, three multiple kernel learning algorithms are derived. The algorithms are experimentally shown to be comparable to or outperform kernel Fisher discriminant analysis and other multiple kernel learning algorithms on benchmark data sets. © 2010 IEEE.

    DOI

  • Estimation of a rotationally symmetric mirror shape from a frontal image of the mirror

    J. Fujiki, Y. Usami, H. Hino, S. Akaho, N. Murata

    International Conference Image and Vision Computing New Zealand    2010年  [査読有り]

     概要を見る

    In structure from motion problems, mirror shape reconstruction from mirror image is very interesting. In this paper, we propose a method to estimate the shape of a rotationally symmetric mirror by calibrating a mirror image which has radially symmetric distortion. In the proposed method, the shape of mirror is represented by the set of solution of a differential equation, which is derived from distortion function computed by the calibration of radially symmetric distortion of an image on a rotationally symmetric mirror. The differential equation and the boundary condition are simplified under the assumption that the image on mirror is taken by orthographic projection camera. The experimental result for a real image is presented to demonstrate the performance of our reconstruction method, and the method is applied for reconstruction of the shape of the mirror painted on the Jan van Eyck's Arnolfini Wedding which had drawn in the 15th century. © 2010 IEEE.

    DOI

  • A Generalization of Independence in Naive Bayes Model

    Yu Fujimoto, Noboru Murata

    INTELLIGENT DATA ENGINEERING AND AUTOMATED LEARNING - IDEAL 2010   6283   153 - +  2010年  [査読有り]

     概要を見る

    In this paper, generalized statistical independence is proposed from the viewpoint of generalized multiplication characterized by a monotonically increasing function and its inverse function, and it is implemented in naive Bayes models. This paper also proposes an idea of their estimation method which directly uses empirical marginal distributions to retain simplicity of calculation. Our method is interpreted as an optimization of a rough approximation of the Bregman divergence so that it is expected to have a kind of robust property. Effectiveness of our proposed models is shown by numerical experiments on some benchmark data sets.

    DOI

  • Bregman divergence and density integration

    Noboru Murata, Yu Fujimoto

    Journal of Math-for-Industry   JMI2009B-3   97 - 104  2009年10月

  • Model selection and information criterion

    Noboru Murata, Hyeyoung Park

    Information Theory and Statistical Learning     333 - 354  2009年  [査読有り]

     概要を見る

    In this chapter, a problem of estimating model parameters from observed data is considered such as regression and function approximation, and a method of evaluating the goodness of model is introduced. Starting from so-called leave-one-out cross-validation, and investigating asymptotic statistical properties of estimated parameters, a generalized Akaike's information criterion (AIC) is derived for selecting an appropriate model from several candidates. In addition to model selection, the concept of information criteria provides an assessment of the goodness of model in various situations. Finally, an optimization method using regularization is presented as an example. © 2009 Springer US.

    DOI

  • ITEM-USER PREFERENCE MAPPING WITH MIXTURE MODELS Data Visualization for Item Preference

    Yu Fujimoto, Hideitsu Hino, Noboru Murata

    KDIR 2009: PROCEEDINGS OF THE INTERNATIONAL CONFERENCE ON KNOWLEDGE DISCOVERY AND INFORMATION RETRIEVAL     105 - +  2009年  [査読有り]

     概要を見る

    In this paper, we propose a visualization technique of a statistical relation of users and preference of items based on a mixture model. In our visualization, items are given as points in a few dimensional preference space, and user specific preferences are given as lines in the same space. The relationship between items and user preferences are intuitively interpreted via projections from points onto lines. As a primitive implementation, we introduce a mixture of the Bradley-Terry models, and visualize the relation between items and user preferences with benchmark data sets.

  • Item Preference Parameters from Grouped Ranking Observations

    Hideitsu Hino, Yu Fujimoto, Noboru Murata

    ADVANCES IN KNOWLEDGE DISCOVERY AND DATA MINING, PROCEEDINGS   5476   875 - 882  2009年  [査読有り]

     概要を見る

    Given a set of rating data for a set of items, determining the values of items is a matter of importance. Various probability models have been proposed to solve this task. The Plackett-Luce model is one of such models, which parametrizes the value of each item by a real valued preference parameter. In this paper, the Plackett-Luce model is generalized to cope with the grouped ranking observations such as movies or restaurants ratings. Since the maximization of the likelihood of the proposed model is computationally intractable, the lower bound of the likelihood which is easy to evaluate is derived, and the ern, algorithm is adopted to the maximization of the lower bound.

    DOI

  • Calibration of Radially Symmetric Distortion by Fitting Principal Component

    Hideitsu Hino, Yumi Usami, Jun Fujiki, Shotaro Akaho, Noboru Murata

    COMPUTER ANALYSIS OF IMAGES AND PATTERNS, PROCEEDINGS   5702   149 - +  2009年  [査読有り]

     概要を見る

    To calibrate radially symmetric distortion of omnidirectional cameras such as fish-eye lenses, calibration parameters are usually estimated so drat lines, which are supposed to be straight, in the 3D real scene; are mapped to straight lines in the calibrated image. In this paper, this problem is treated as a fitting problem of the principal component in uncalibrated images, and an estimation procedure of calibration parameters is proposed based oil the principal component analysis. Experimental results for synthetic data and real images are presented to demonstrate the performance of our calibration method.

    DOI

  • An Information Theoretic Perspective of the Sparse Coding

    Hideitsu Hino, Noboru Murata

    ADVANCES IN NEURAL NETWORKS - ISNN 2009, PT 1, PROCEEDINGS   5551 ( PART 1 ) 84 - 93  2009年  [査読有り]

     概要を見る

    The sparse coding method is formulated as an information theoretic optimization problem. The rate distortion theory leads to all objective functional which can be interpreted as an information theoretic formulation of the sparse coding. Viewing as an entropy minimization problem, the rate distortion theory and consequently the sparse coding are extended to discriminative variants. As a concrete example of this information theoretic sparse coding, a discriminative non-linear sparse coding algorithm with neural networks is proposed. Experimental results of gender classification by face images show that the discriminative sparse coding is more robust to noise, compared to the conventional method which directly uses images as inputs to a linear support vector machine.

    DOI

  • Robust boosting algorithm against mislabeling in multiclass problems

    Takashi Takenouchi, Shinto Eguchi, Noboru Murata, Takafumi Kanamori

    NEURAL COMPUTATION   20 ( 6 ) 1596 - 1630  2008年06月  [査読有り]

     概要を見る

    We discuss robustness against mislabeling in multiclass labels for classification problems and propose two algorithms of boosting, the normalized Eta-Boost.M and Eta-Boost.M, based on the Eta-divergence. Those two boosting algorithms are closely related to models of mislabeling in which the label is erroneously exchanged for others. For the two boosting algorithms, theoretical aspects supporting the robustness for mislabeling are explored. We apply the proposed two boosting methods for synthetic and real data sets to investigate the performance of these methods, focusing on robustness, and confirm the validity of the proposed methods.

    DOI

  • 球面上の点列に対する連接小円回帰を用いたカメラ運動の平滑化

    野田 容士, 藤木 淳, 村田 昇

    電子情報通信学会論文誌   J91-D ( 5 ) 1336 - 1348  2008年05月

  • Neuromagnetic responses to chords are modified by preceding musical scale

    Asuka Otsuka, Shinya Kuriki, Noboru Murata, Toshikazu Hasegawa

    NEUROSCIENCE RESEARCH   60 ( 1 ) 50 - 55  2008年01月  [査読有り]

     概要を見る

    Previous psychological studies have shown that musical chords primed by Western musical scale in a tonal and modal schema are perceived in a hierarchy of stability. We investigated such priming effects on auditory magnetic responses to tonic-major and submediant-minor chords preceded by major scales and tonic-minor and submediant-major chords preceded by minor scales. Musically trained subjects participated in the experiment. During MEG recordings, subjects judged perceptual stability of the chords. The tonic chords were judged to be stable, whereas the submediant chords were judged to be unstable. Dipole moments of N1m response originating in the auditory cortex were larger in the left hemisphere for the submediant chords than for the tonic chords preceded by the major but not minor scales. No difference in the N1m or P2m moment was found for the chords presented without preceding scales. These results suggest priming effects of the tonal schema, interacting with contextual modality, on neural activity of the auditory cortex as well as perceptual stability of the chords. It is inferred that modulation of the auditory cortical activity is associated with attention induced by tonal instability and modality shift, which characterize the submediant chords. (c) 2007 Elsevier Ireland Ltd and the Japan Neuroscience Society. All rights reserved.

    DOI

  • A Generalized Product Rule and Weak Independence based on Bregman Divergence

    Yu Fujimoto, Noboru Murata

    WMSCI 2008: 12TH WORLD MULTI-CONFERENCE ON SYSTEMICS, CYBERNETICS AND INFORMATICS, VOL V, PROCEEDINGS   5   248 - 253  2008年  [査読有り]

     概要を見る

    To describe the relation between some values, arithmetic operations like multiplication and division are important and conventional tools. These arithmetic operations for probabilities are characterized by the KL divergence, and hence, they can be generalized by using the Bregman divergence instead of the KL divergence. With this idea, independence of random variables is modified by generalized product rule, and a joint probability model is proposed based on this modified weak independence. Effectiveness of weak independent models is shown by numerical experiments on toy examples, and discussed from a geometrical viewpoint.

  • Robust loss functions for boosting

    Takafumi Kanamori, Takashi Takenouchi, Shinto Eguchi, Noboru Murata

    NEURAL COMPUTATION   19 ( 8 ) 2183 - 2244  2007年08月

     概要を見る

    Boosting is known as a gradient descent algorithm over loss functions. It is often pointed out that the typical boosting algorithm, Adaboost, is highly affected by outliers. In this letter, loss functions for robust boosting are studied. Based on the concept of robust statistics, we propose a transformation of loss functions that makes boosting algorithms robust against extreme outliers. Next, the truncation of loss functions is applied to contamination models that describe the occurrence of mislabels near decision boundaries. Numerical experiments illustrate that the proposed loss functions derived from the contamination models are useful for handling highly noisy data in comparison with other loss functions.

    DOI

  • Robust loss functions for boosting

    Takafumi Kanamori, Takashi Takenouchi, Shinto Eguchi, Noboru Murata

    NEURAL COMPUTATION   19 ( 8 ) 2183 - 2244  2007年08月  [査読有り]

     概要を見る

    Boosting is known as a gradient descent algorithm over loss functions. It is often pointed out that the typical boosting algorithm, Adaboost, is highly affected by outliers. In this letter, loss functions for robust boosting are studied. Based on the concept of robust statistics, we propose a transformation of loss functions that makes boosting algorithms robust against extreme outliers. Next, the truncation of loss functions is applied to contamination models that describe the occurrence of mislabels near decision boundaries. Numerical experiments illustrate that the proposed loss functions derived from the contamination models are useful for handling highly noisy data in comparison with other loss functions.

    DOI

  • A modified EM algorithm for mixture models based on Bregman divergence

    Yu Fujimoto, Noboru Murata

    ANNALS OF THE INSTITUTE OF STATISTICAL MATHEMATICS   59 ( 1 ) 3 - 25  2007年03月  [査読有り]

     概要を見る

    The EM algorithm is a sophisticated method for estimating statistical models with hidden variables based on the Kullback-Leibler divergence. A natural extension of the Kullback-Leibler divergence is given by a class of Bregman divergences, which in general enjoy robustness to contamination data in statistical inference. In this paper, a modification of the EM algorithm based on the Bregman divergence is proposed for estimating finite mixture models. The proposed algorithm is geometrically interpreted as a sequence of projections induced from the Bregman divergence. Since a rigorous algorithm includes a nonlinear optimization procedure, two simplification methods for reducing computational difficulty are also discussed from a geometrical viewpoint. Numerical experiments on a toy problem are carried out to confirm appropriateness of the simplifications.

    DOI

  • Tutorial series on brain-inspired computing Part 6: Geometrical structure of boosting algorithm

    Takafumi Kanamori, Takashi Takenouchi, Noboru Murata

    NEW GENERATION COMPUTING   25 ( 1 ) 117 - 141  2007年  [査読有り]

     概要を見る

    In this article, several boosting methods are discussed, which are notable implementations of the ensemble learning. Starting from the firstly introduced "boosting by filter" which is an embodiment of the proverb "Two heads are better than one", more advanced versions of boosting methods "AdaBoost" and "U-Boost" are introduced. A geometrical structure and some statistical properties such as consistency and robustness of boosting algorithms are discussed, and then simulation studies are presented for confirming discussed behaviors of algorithms.

    DOI

  • Robust estimation for mixture of probability tables based on β-likelihood

    Fujimoto Y, Murata N

    Proceedings of the Sixth SIAM International Conference on Data Mining   2006   519 - 523  2006年  [査読有り]

  • Geometrical properties of Nu support vector machines with different norms

    K Ikeda, N Murata

    NEURAL COMPUTATION   17 ( 11 ) 2508 - 2529  2005年11月  [査読有り]

     概要を見る

    By employing the L-1 or L-infinity norms in maximizing margins, support vector machines (SVMs) result in a linear programming problem that requires a lower computational load compared to SVMs with the L-2 norm. However, how the change of norm affects the generalization ability of SVMs has not been clarified so far except for numerical experiments. In this letter, the geometrical meaning of SVMs with the L-p norm is investigated, and the SVM solutions are shown to have rather little dependency on p.

    DOI

  • A Gaussian process robust regression

    N Murata, Y Kuroda

    PROGRESS OF THEORETICAL PHYSICS SUPPLEMENT   157 ( 157 ) 280 - 283  2005年  [査読有り]

     概要を見る

    A modified Gaussian process regression is proposed aiming at making regressors robust against outliers. The proposed method is based on U-loss, which is introduced as a natural extension of Kullback-Leibler divergence. The robustness is examined based on the influence function, and numerical experiments are conducted for contaminated data sets and it is shown that the practical performance agrees with the theoretical analysis.

  • Effects of norms on learning properties of support vector machines

    K Ikeda, N Murata

    2005 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH, AND SIGNAL PROCESSING, VOLS 1-5   V   241 - 244  2005年  [査読有り]

     概要を見る

    Support Vector Machines (SVMs) are known to have a high generalization ability, yet a heavy computational load since margin maximization results in a quadratic programming problem. It is known that this maximization task results in a pth-order programming problem if we employ the LP norm instead of the L-2 norm. In this paper, we theoretically show the effects of p on the learning properties of SVMs by clarifying its geometrical meaning.

    DOI

  • Information geometry of U-Boost and Bregman divergence

    N Murata, T Takenouchi, T Kanamori, S Eguchi

    NEURAL COMPUTATION   16 ( 7 ) 1437 - 1481  2004年07月  [査読有り]

     概要を見る

    We aim at an extension of AdaBoost to U-Boost, in the paradigm to build a stronger classification machine from a set of weak learning machines. A geometric understanding of the Bregman divergence defined by a generic convex function U leads to the U-Boost method in the framework of information geometry extended to the space of the finite measures over a label set. We propose two versions of U-Boost learning algorithms by taking account of whether the domain is restricted to the space of probability functions. In the sequential step, we observe that the two adjacent and the initial classifiers are associated with a right triangle in the scale via the Bregman divergence, called the Pythagorean relation. This leads to a mild convergence property of the U-Boost algorithm as seen in the expectation-maximization algorithm. Statistical discussions for consistency and robustness elucidate the properties of the U-Boost methods based on a stochastic assumption for training data.

    DOI

  • 独立性の検定を用いた,独立成分のグルーピング手法の提案

    伊東 大祐, 向井 卓也, 村田 昇

    日本生体磁気学会論文誌   16 ( 2 ) 23 - 31  2004年06月

  • Improving generalization performance of natural gradient learning using optimized regularization by NIC

    H Park, N Murata, S Amari

    NEURAL COMPUTATION   16 ( 2 ) 355 - 382  2004年02月  [査読有り]

     概要を見る

    Natural gradient learning is known to be efficient in escaping plateau, which is a main cause of the slow learning speed of neural networks. The adaptive natural gradient learning method for practical implementation also has been developed, and its advantage in real-world problems has been confirmed. In this letter, we deal with the generalization performances of the natural gradient method. Since natural gradient learning makes parameters fit to training data quickly, the overfitting phenomenon may easily occur, which results in poor generalization performance. To solve the problem, we introduce the regularization term in natural gradient learning and propose an efficient optimizing method for the scale of regularization by using a generalized Akaike information criterion (network information criterion). We discuss the properties of the optimized regularization strength by NIC through theoretical analysis as well as computer simulations. We confirm the computational efficiency and generalization performance of the proposed method in real-world applications through computational experiments on benchmark problems.

    DOI

  • The most robust loss function for boosting

    T Kanamori, T Takenouchi, S Eguchi, N Murata

    NEURAL INFORMATION PROCESSING   3316   496 - 501  2004年  [査読有り]

     概要を見る

    Boosting algorithm is understood as the gradient descent algorithm of a loss function. It is often pointed out that the typical boosting algorithm, Adaboost, is seriously affected by the outliers. In this paper, loss functions for robust boosting are studied. Based on a concept of the robust statistics, we propose a positive-part-truncation of the loss function which makes the boosting algorithm robust against extreme outliers. Numerical experiments show that the proposed boosting algorithm is useful for highly noisy data in comparison with other competitors.

  • Nonlinear PCA/ICA for the structure from motion problem

    J Fujiki, S Akaho, N Murata

    INDEPENDENT COMPONENT ANALYSIS AND BLIND SIGNAL SEPARATION   3195   750 - 757  2004年  [査読有り]

     概要を見る

    Recovering both camera motion and object shape from multiple images, called structure from motion problem, is an important and essential problem in computer vision. Generally, the result of the structure from motion problem has an ambiguity represented by a three-dimensional rotation matrix. We present two kinds of specific criteria such as independence of parameters to fix the ambiguity by choosing an appropriate rotation matrix in the sense of computer vision. Once some criterion is defined, the fixing of the ambiguity is reduced to a nonlinear extension of the PCA/ICA. We examine the efficiency through synthetic experiments.

  • Learning properties of support vector machines with p-norm

    K Ikeda, N Murata

    2004 47TH MIDWEST SYMPOSIUM ON CIRCUITS AND SYSTEMS, VOL III, CONFERENCE PROCEEDINGS   3   69 - 72  2004年  [査読有り]

     概要を見る

    Support Vector Machines (SVMs) are a new classification technique which has a high generalization ability, yet a heavy computational load since margin maximization results in a quadratic programming problem. It is known that this maximization task results in a pth-order programming problem if we employ the p-norm instead of the Euclidean norm, that is. When p = 1, for example, it is a linear programming problem with a much lower computational load. In this article, we theoretically show that p has very little affect on the generalization performance of SVMs in practice by considering its geometrical meaning.

  • An approach of moment-based algorithm for noisy ICA models

    D Ito, N Murata

    INDEPENDENT COMPONENT ANALYSIS AND BLIND SIGNAL SEPARATION   3195   343 - 349  2004年  [査読有り]

     概要を見る

    Factor analysis is well known technique to uncorrelate observed signals with Gaussina noises before ICA (Independent Component Analysis) algorithms are applied. However, factor analysis is not applicable when the number of source signals are more than that of Ledermann's bound, and when the observations are contaminated by non-Gaussian noises. In this paper, an approach is proposed based on higher-order moments of signals and noises in order to overcome those constraints.

  • A robust approach to independent component analysis of signals with high-level noise measurements

    J Cao, N Murata, S Amari, A Cichocki, T Takeda

    IEEE TRANSACTIONS ON NEURAL NETWORKS   14 ( 3 ) 631 - 645  2003年05月  [査読有り]

     概要を見る

    In this paper, we propose a robust approach for independent component analysis (ICA) of signals that observations are contaminated with high-level additive noise and/or outliers. The source signals may contain mixtures of both sub-Gaussian and super-Gaussian components, and the number of sources is unknown. Our robust approach includes two procedures. In the first procedure, a robust prewhitening technique is used to reduce the power of additive noise, the dimensionality and the correlation among sources. A cross-validation technique is introduced to estimate the number of sources in this first procedure. In the second procedure, a nonlinear function is derived using the parameterized t-distribution density model. This nonlinear function is robust against the undue influence of outliers fundamentally. Moreover, the stability of the proposed algorithm and the robust property of misestimating the parameters (kurtosis) have been studied. By combining the t-distribution model with a family of light-tailed distributions (sub-Gaussian) model, we can separate the mixture of sub-Gaussian and super-Gaussian source components. Through the analysis of artificially synthesized data and real-world magnetoencephalographic (MEG) data, we illustrate the efficacy of this robust approach.

    DOI

  • Independent component analysis for unaveraged single-trial MEG data decomposition and single-dipole source localization

    JT Cao, N Murata, S Amari, A Cichocki, T Takeda

    NEUROCOMPUTING   49 ( “1-4” ) 255 - 277  2002年12月  [査読有り]

     概要を見る

    This paper presents a novel method for decomposing and localizing unaveraged single-trial magnetoencephalographic data based on the independent component analysis (ICA) approach associated with pre- and post-processing techniques. In the pre-processing stage, recorded single-trial raw data are first decomposed into uncorrelated signals with the reduction of high-power additive noise. In the stage of source separation, the decorrelated source signals are further decomposed into independent source components. In the post-processing stage, we perform a source localization procedure to seek a single-dipole map of decomposed individual source components, e.g., evoked responses. The first results of applying the proposed robust ICA approach to single-trial data with phantom and auditory evoked field tasks indicate the following. (1) A source signal is successfully extracted from unaveraged single-trial phantom data. The accuracy of dipole estimation for the decomposed source is even better than that of taking the average of total trials. (2) Not only the behavior and location of individual neuronal sources can be obtained but also the activity strength (amplitude) of evoked responses corresponding to a stimulation trial can be obtained and visualized. Moreover, the dynamics of individual neuronal sources, such as the trial-by-trial variations of the amplitude and location, can be observed. (C) 2002 Elsevier Science B.V. All rights reserved.

    DOI

  • On-line learning in changing environments with applications in supervised and unsupervised learning

    N Murata, M Kawanabe, A Ziehe, KR Muller, S Amari

    NEURAL NETWORKS   15 ( 4-6 ) 743 - 760  2002年06月  [査読有り]

     概要を見る

    An adaptive on-line algorithm extending the learning of learning idea is proposed and theoretically motivated. Relying only on gradient flow information it can be applied to learning continuous functions or distributions, even when no explicit loss function is given and the Hessian is not available. The framework is applied for unsupervised and supervised learning. Its efficiency is demonstrated for drifting and switching non-stationary blind separation tasks of acoustic signals. Furthermore applications to classification (US postal service data set) and time-series prediction in changing environments are presented. (C) 2002 Elsevier Science Ltd. All rights reserved.

    DOI

  • Support vector machines with different norms: motivation, formulations and results

    JP Pedroso, N Murata

    PATTERN RECOGNITION LETTERS   22 ( 12 ) 1263 - 1272  2001年10月  [査読有り]

     概要を見る

    We introduce two formulations for training support vector machines, based on considering the L-1 and L-infinity norms instead of the currently used L-2 norm, and maximising the margin between the separating hyperplane and each data sets using L-1 and L-infinity distances. We exploit the geometrical properties of these different norms. and propose what kind of results should be expected for them. Formulations in mathematical programming for linear problems corresponding to L-1 and L-infinity norms are also provided, for both the separable and non-separable cases. We report results obtained for some standard benchmark problems. which confirmed that the performance of all the formulations is similar. As expected, the CPU time required for machines solvable with linear programming is much shorter. (C) 2001 Elsevier Science B.V. All rights reserved.

    DOI

  • Multiplicative nonholonomic/Newton-like algorithm

    T Akuzawa, N Murata

    CHAOS SOLITONS & FRACTALS   12 ( 4 ) 785 - 793  2001年03月  [査読有り]

     概要を見る

    We construct new algorithms, which use the fourth order cumulant of stochastic variables for the cost function. The multiplicative updating rule here constructed is natural from the homogeneous nature of the Lie group and has numerous merits for the rigorous treatment of the dynamics. As one consequence, the second order convergence is shown. For the cost function, functions invariant under the componentwise scaling are chosen. By identifying points which can be transformed to each other by the scaling, we assume that the dynamics is in a coset space. In our method, a point can move toward any direction in this coset. Thus, no prewhitening is required. (C) 2001 Elsevier Science Ltd. All rights reserved.

    DOI

  • An approach to blind source separation based on temporal structure of speech signals

    Noboru Murata, Shiro Ikeda, Andreas Ziehe

    Neurocomputing   41 ( 1-4 ) 1 - 24  2001年  [査読有り]

     概要を見る

    In this paper, we introduce a new technique for blind source separation of speech signals. We focus on the temporal structure of the signals. The idea is to apply the decorrelation method proposed by Molgedey and Schuster in the time-frequency domain. Since we are applying separation algorithm on each frequency separately, we have to solve the amplitude and permutation ambiguity properly to reconstruct the separated signals. For solving the amplitude ambiguity, we use the matrix inversion and for the permutation ambiguity, we introduce a method based on the temporal structure of speech signals. We show some results of experiments with both artificially controlled data and speech data recorded in the real environment. © 2001 Elsevier Science B.V. All rights reserved.

    DOI

  • Sequential extraction of minor components

    TP Chen, SI Amari, N Murata

    NEURAL PROCESSING LETTERS   13 ( 3 ) 195 - 201  2001年  [査読有り]

     概要を見る

    Principal component analysis (PCA) and Minor component analysis (MCA) are similar but have different dynamical performances. Unexpectedly, a sequential extraction algorithm for MCA proposed by Luo and Unbehauen [11] does not work for MCA, while it works for PCA. We propose a different sequential-addition algorithm which works for MCA. We also show a conversion mechanism by which any PCA algorithms are converted to dynamically equivalent MCA algorithms and vice versa.

    DOI

  • Single-trial magnetoencephalographic data decomposition and localization based on independent component analysis approach

    JT Cao, N Murata, S Amari, A Cichocki, T Takeda, H Endo, N Harada

    IEICE TRANSACTIONS ON FUNDAMENTALS OF ELECTRONICS COMMUNICATIONS AND COMPUTER SCIENCES   E83A ( 9 ) 1757 - 1766  2000年09月

     概要を見る

    Magnetoencephalography (MEG) is a powerful and non-invasive technique for measuring human brain activity with a high temporal resolution. The motivation for studying MEG data analysis is to extract the essential features from measured data and represent them corresponding to the human brain functions. In this paper a novel MEG data analysis method based on independent component analysis (ICA) approach with pre-processing: and post-processing multistage procedures is proposed. Moreover, several kinds of ICA algorithms are investigated for analyzing MEG; tingle-trial data which is recorded in the experiment of phantom. The analyzed results are presented to illustrate the effectiveness and high performance both in source decomposition by IC:A approaches and source localization by equivalent current dipoles fitting method.

  • Single-trial magnetoencephalographic data decomposition and localization based on independent component analysis approach

    JT Cao, N Murata, S Amari, A Cichocki, T Takeda, H Endo, N Harada

    IEICE TRANSACTIONS ON FUNDAMENTALS OF ELECTRONICS COMMUNICATIONS AND COMPUTER SCIENCES   E83A ( 9 ) 1757 - 1766  2000年09月  [査読有り]

     概要を見る

    Magnetoencephalography (MEG) is a powerful and non-invasive technique for measuring human brain activity with a high temporal resolution. The motivation for studying MEG data analysis is to extract the essential features from measured data and represent them corresponding to the human brain functions. In this paper a novel MEG data analysis method based on independent component analysis (ICA) approach with pre-processing: and post-processing multistage procedures is proposed. Moreover, several kinds of ICA algorithms are investigated for analyzing MEG; tingle-trial data which is recorded in the experiment of phantom. The analyzed results are presented to illustrate the effectiveness and high performance both in source decomposition by IC:A approaches and source localization by equivalent current dipoles fitting method.

  • Independent component analysis algorithm for online blind source separation and blind equalization systems

    Jianting Cao, Noboru Murata, Andrew Cichocki

    Journal of Signal Processing   4 ( 2 ) 131 - 140  2000年03月

  • Optimisation on support vector machines

    JP Pedroso, N Murata

    IJCNN 2000: PROCEEDINGS OF THE IEEE-INNS-ENNS INTERNATIONAL JOINT CONFERENCE ON NEURAL NETWORKS, VOL VI     399 - 404  2000年  [査読有り]

     概要を見る

    In this paper we deal with the optimisation problem involved in determining the maximal margin separation hyperplane in support vector machines. We consider three different formulations, based on L-2 norm distance (the standard case), L-1 norm, and L-infinity norm. We consider separation in the original space of the data (i.e., there are no kernel transformations).
    For any of these cases, we focus on the following problem: having the optimal solution for a given training data set, one is given a new training example. The purpose is to use the information about the solution of the problem without the additional example in order to speed up the new optimisation problem. We also consider the case of reoptimisation after removing an example from the data set.
    We report results obtained for some standard benchmark problems.

  • Population decoding based on an unfaithful model

    S Wu, H Nakahara, N Murata, S Amari

    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 12   12   192 - 198  2000年  [査読有り]

     概要を見る

    We study a population decoding paradigm in which the maximum likelihood inference is based on an unfaithful decoding model (UMLI), This is usually the case for neural population decoding because the encoding process of the brain is not exactly known, or because a simplified decoding model is preferred for saving computational cost. We consider an unfaithful decoding model which neglects the pair-wise correlation between neuronal activities, and prove that UMLI is asymptotically efficient when the neuronal correlation is uniform or of limited-range. The performance of UMLI is compared with that of the maximum likelihood inference based on a faithful model and that of the center of mass decoding method. It turns out that UMLI has advantages of decreasing the computational complexity remarkablely and maintaining a high-level decoding accuracy at the same time. The effect of correlation on the decoding accuracy is also discussed.

  • A stable and robust ICA algorithm based on t-distribution and generalized Gaussian distribution models

    Jianting Cao, Noboru Murata

    Neural Networks for Signal Processing   IX   283 - 292  1999年08月

    DOI

  • Statistical analysis of learning dynamics

    N Murata, S Amari

    SIGNAL PROCESSING   74 ( 1 ) 3 - 28  1999年04月  [査読有り]

     概要を見る

    Learning is a flexible and effective means of extracting the stochastic structure of the environment. It provides an effective method for blind separation and deconvolution in signal processing. Two different types of learning are used, namely batch learning and on-line learning. The batch learning procedure uses all the training examples repeatedly so that its performance is compared to the statistical estimation procedure. On-line learning is more dynamical, updating the current estimate by observing a new datum one by one. On-line learning is slow in general but works well in the changing environment. The present paper gives a unified framework of statistical analysis for batch and on-line learning. The topics include the asymptotic learning curve, generalization error and training error, over-fitting and over-training, efficiency of learning, and an adaptive method of determining learning rate. (C) 1999 Elsevier Science B.V. All rights reserved.

    DOI

  • Stable and robust ICA algorithm based on t-distribution and generalized Gaussian distribution models

    Cao Jianting, Murata Noboru

    Neural Networks for Signal Processing - Proceedings of the IEEE Workshop     283 - 292  1999年  [査読有り]

  • Statistical study on on-line learning

    Noboru Murata

    In David Saad editor, On-line Learning in Neural Networks, Cambridge University Press    1998年12月

  • Statistical inference: learning in artificial neural networks

    HH Yang, N Murata, S Amari

    TRENDS IN COGNITIVE SCIENCES   2 ( 1 ) 4 - 10  1998年01月  [査読有り]

     概要を見る

    Artificial neural networks (ANNs) are widely used to model low-level neural activities and high-level cognitive functions. In this article, we review the application of statistical inference for learning in ANNs. Statistical inference provides an objective way to derive learning algorithms both for training and for evaluation of the performance of trained ANNs. Solutions to the over-fitting problem by model- selection methods, based on either conventional statistical approaches or on a Bayesian approach, are discussed. The use of supervised and unsupervised learning algorithms for ANNs are reviewed. Training a multilayer ANN by supervised learning is equivalent to nonlinear regression. The ensemble methods, bagging and arching, described here, can be applied to combine ANNs to form a new predictor with improved performance. Unsupervised learning algorithms that are derived either by the Hebbian law for bottom-up self-organization, or by global objective functions for top-down self-organization are also discussed.

    DOI

  • Asymptotic statistical theory of overtraining and cross-validation

    S Amari, N Murata, KR Muller, M Finke, HH Yang

    IEEE TRANSACTIONS ON NEURAL NETWORKS   8 ( 5 ) 985 - 996  1997年09月  [査読有り]

     概要を見る

    A statistical theory for overtraining is proposed. The analysis treats general realizable stochastic neural networks, trained with Kullback-Leibler divergence in the asymptotic case of a large number of training examples. It is shown that the asymptotic gain in the generalization error Is small if we perform early stopping, even if we have access to the optimal stopping time. Considering cross-validation stopping we answer the question: In what ratio the examples should be divided into training and cross-validation sets in order to obtain the optimum performance. Although cross-validated early stopping is useless in the asymptotic region, it surely decreases the generalization error in the nonasymptotic region. Our large scale simulations done on a CM5 are in nice agreement with our analytical findings.

    DOI

  • Adaptive on-line learning in changing environments

    N Murata, KR Muller, A Ziehe, SI Amari

    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 9   9   599 - 605  1997年  [査読有り]

     概要を見る

    An adaptive on-line algorithm extending the learning of learning idea is proposed and theoretically motivated. Relying only on gradient flow information it can be applied, to learning continuous functions or distributions, even when no explicit loss function is given and the Hessian is not available. Its efficiency is demonstrated for a non-stationary blind separation task of acoustic signals.

  • Statistical analysis of regularization constant - From Bayes, MDL and NIC points of view

    S Amari, N Murata

    BIOLOGICAL AND ARTIFICIAL COMPUTATION: FROM NEUROSCIENCE TO TECHNOLOGY   1240   284 - 293  1997年  [査読有り]

     概要を見る

    In order to avoid overfitting in neural learning, a regularization term is added to the loss function to be minimized. It is naturally derived from the Bayesian standpoint. The present paper studies how to determine the regularization constant from the points of view of the empirical Bayes approach, the maximum description length (MDL) approach, and the network information criterion (NIC) approach. The asymptotic statistical analysis is given to elucidate their differences. These approaches are tightly connected with the method of model selection. The superiority of the NIC is shown from this analysis.

  • Approximation bounds of three-layered neural networks -- A theorem on an integral transform with ridge functions

    Noboru Murata

    Electronics and Communications in Japan   79 ( 3 ) 23 - 33  1996年08月  [査読有り]

    DOI

  • A numerical study on learning curves in stochastic multilayer feedforward networks

    KR Muller, M Finke, N Murata, K Schulten, S Amari

    NEURAL COMPUTATION   8 ( 5 ) 1085 - 1106  1996年07月  [査読有り]

     概要を見る

    The universal asymptotic scaling laws proposed by Amari et al. are studied in large scale simulations using a CM5. Small stochastic multilayer feedforward networks trained with backpropagation are investigated. In the range of a large number of training patterns t, the asymptotic generalization error scales as 1/t as predicted. For a medium range t a faster 1/t(2) scaling is observed. This effect is explained by using higher order corrections of the likelihood expansion. It is shown for small t that the scaling law changes drastically, when the network undergoes a transition from strong overfitting to effective learning.

    DOI

  • Approximation bounds of three-layered neural networks - A theorem on an integral transform with ridge functions

    N Murata

    ELECTRONICS AND COMMUNICATIONS IN JAPAN PART III-FUNDAMENTAL ELECTRONIC SCIENCE   79 ( 3 ) 23 - 33  1996年03月  [査読有り]

     概要を見る

    Neural networks have attracted attention due to their capability to perform nonlinear function approximation. In this paper, in order to better understand this capability, a new theorem on an integral transform was derived by applying ridge functions to neural networks. From the theorem, it is possible to obtain approximation bounds which clarify the quantitative relationship between the function approximation accuracy and the number of nodes in the hidden layer. The theorem indicates that the approximation accuracy depends on the smoothness of the target function. Furthermore, the theorem also shows that this type of approximation method differs from usual methods and is able to escape the so-called ''curse of dimensionality,'' in which the approximation accuracy depends strongly of the input dimension of the function and deteriorates exponentially.

  • An integral representation of functions using three-layered networks and their approximation bounds

    Noboru Murata

    Neural Networks   9 ( 6 ) 947 - 956  1996年  [査読有り]

     概要を見る

    Neural networks are widely known to provide a method of approximating nonlinear functions. In order to clarify its approximation ability, a new theorem on an integral transform of ridge functions is presented. By using this theorem, an approximation bound, which evaluates the quantitative relationship between the approximation accuracy and the number of elements in the hidden layer, can be obtained. This result shows that the approximation accuracy depends on the smoothness of target functions. It also shows that the approximation methods which use ridge functions are free from the 'curse of dimensionality'.

    DOI

  • Statistical theory of overtraining - Is cross-validation asymptotically effective?

    S Amari, N Murata, KR Muller, M Finke, H Yang

    ADVANCES IN NEURAL INFORMATION PROCESSING SYSTEMS 8   8   176 - 182  1996年

  • Ridge 関数による積分表現と3層ネットワークの近似誤差

    村田 昇

    電子情報通信学会論文誌   J78-A ( 9 ) 1204 - 1212  1995年09月

  • NETWORK INFORMATION CRITERION - DETERMINING THE NUMBER OF HIDDEN UNITS FOR AN ARTIFICIAL NEURAL-NETWORK MODEL

    N MURATA, S YOSHIZAWA, S AMARI

    IEEE TRANSACTIONS ON NEURAL NETWORKS   5 ( 6 ) 865 - 872  1994年11月  [査読有り]

     概要を見る

    The problem of model selection, or determination of the number of hidden units, can be approached statistically, by generalizing Akaike's information Criterion (AIC) to be applicable to unfaithful (i.e., unrealizable) models with general loss criteria including regularization terms. The relation between the training error and the generalization error is studied in terms of the number of the training examples and the complexity of It network which reduces to the number of parameters in the ordinary statistical theory of the AIC. This relation leads to a new Network Information Criterion (NIC) which is useful for selecting the optimal network model based on a given training set.

    DOI

  • UNIVERSAL PROPERTIES OF LEARNING-CURVES

    S AMARI, N MURATA, K IKEDA

    COGNITIVE PROCESSING FOR VISION & VOICE     77 - 87  1994年  [査読有り]

  • PREDICTION ERROR OF STOCHASTIC LEARNING MACHINE

    K IKEDA, N MURATA, S AMARI

    1994 IEEE INTERNATIONAL CONFERENCE ON NEURAL NETWORKS, VOL 1-7   2   1159 - 1162  1994年  [査読有り]

  • Learning curves, model selection and complexity of neural networks

    Noboru Murata, Shuji Yoshizawa, Shun-ichi Amari

    In Stephen Jose Hanson, Jack D. Cowan, and C. Lee Giles, editors, Advances in Neural Information Processing Systems, Morgan Kaufmann Publishers, San Mateo, CA   5   607 - 614  1993年

  • STATISTICAL-THEORY OF LEARNING-CURVES UNDER ENTROPIC LOSS CRITERION

    S AMARI, N MURATA

    NEURAL COMPUTATION   5 ( 1 ) 140 - 153  1993年01月

     概要を見る

    The present paper elucidates a universal property of learning curves, which shows how the generalization error, training error, and the complexity of the underlying stochastic machine are related and how the behavior of a stochastic machine is improved as the number of training examples increases. The error is measured by the entropic loss. It is proved that the generalization error converges to H0, the entropy of the conditional distribution of the true machine, as H0 + m*/(2t), while the training error converges as H0 - m*/(2t), where t is the number of examples and m* shows the complexity of the network. When the model is faithful, implying that the true machine is in the model, m* is reduced to m, the number of modifiable parameters. This is a universal law because it holds for any regular machine irrespective of its structure under the maximum likelihood estimator. Similar relations are obtained for the Bayes and Gibbs learning algorithms. These learning curves show the relation among the accuracy of learning, the complexity of a model, and the number of training examples.

    DOI

  • A CRITERION FOR DETERMINING THE NUMBER OF PARAMETERS IN AN ARTIFICIAL NEURAL NETWORK MODEL

    N MURATA, S YOSHIZAWA, S AMARI

    ARTIFICIAL NEURAL NETWORKS, VOLS 1 AND 2     9 - 14  1991年  [査読有り]

▼全件表示

書籍等出版物

  • パターン認識と機械学習 : ベイズ理論による統計的予測 上

    Bishop Christopher M, 元田 浩, 栗田 多喜夫, 樋口 知之, 松本 裕治, 村田 昇

    丸善出版  2012年 ISBN: 9784621061220

  • 電気・電子・情報のための基礎数学

    村田 純一, 村田 昇

    オーム社  2011年 ISBN: 9784274210877

  • パターン認識

    金森 敬文, 竹之内 高志, 村田 昇

    共立出版  2009年 ISBN: 9784320019256

  • パターン認識と機械学習 : ベイズ理論による統計的予測 上

    Bishop Christopher M, 元田 浩, 栗田 多喜夫, 樋口 知之, 松本 裕治, 村田 昇

    シュプリンガー・ジャパン  2007年 ISBN: 9784431100133

  • 確率と統計 : 情報学への架橋

    渡辺 澄夫, 村田 昇

    コロナ社  2005年 ISBN: 9784339060775

  • 情報理論の基礎 - 情報と学習の直観的理解のために.

    村田昇

    サイエンス社  2005年 ISBN: 9784781912127

  • 入門 独立成分分析.

    村田昇

    東京電機大学出版局  2004年 ISBN: 4501537507

  • パターン認識と学習の統計学 - 新しい概念と手法.

    麻生英樹, 津田宏治, 村田昇

    岩波書店  2003年 ISBN: 4000068466

  • 独立成分分析 - 多変量データ解析の新しい方法.

    甘利俊一, 村田昇 共編著

    サイエンス社  2002年

  • パターン認識と機械学習 : ベイズ理論による統計的予測 下

    Bishop Christopher M, 元田 浩, 栗田 多喜夫, 樋口 知之, 松本 裕治, 村田 昇

    シュプリンガー・ジャパン  ISBN: 9784431100317

  • パターン認識と機械学習 : ベイズ理論による統計的予測 下

    Bishop Christopher M, 元田 浩, 栗田 多喜夫, 樋口 知之, 松本 裕治, 村田 昇

    丸善出版  ISBN: 9784621061244

▼全件表示

Misc

  • The global optimum of shallow neural network is attained by ridgelet transform

    Sho Sonoda, Isao Ishikawa, Masahiro Ikeda, Kei Hagihara, Yoshihiro Sawano, Takuo Matsubara, Noboru Murata

       2018年05月

    機関テクニカルレポート,技術報告書,プレプリント等  

     概要を見る

    We prove that the global minimum of the backpropagation (BP) training problem<br />
    of neural networks with an arbitrary nonlinear activation is given by the<br />
    ridgelet transform. A series of computational experiments show that there<br />
    exists an interesting similarity between the scatter plot of hidden parameters<br />
    in a shallow neural network after the BP training and the spectrum of the<br />
    ridgelet transform. By introducing a continuous model of neural networks, we<br />
    reduce the training problem to a convex optimization in an infinite dimensional<br />
    Hilbert space, and obtain the explicit expression of the global optimizer via<br />
    the ridgelet transform.

  • 積分表現とKernel HerdingによるNeural Networkの学習 (情報論的学習理論と機械学習)

    松原 拓央, 園田 翔, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   116 ( 500 ) 25 - 31  2017年03月

    CiNii

  • 部分観測されたスパイクからの神経細胞間の結合推定 (ニューロコンピューティング)

    岩崎 泰士, 日野 英逸, 龍野 正実, 赤穂 昭太郎, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   116 ( 313 ) 21 - 26  2016年11月

    CiNii

  • 無限層デノイジング・オートエンコーダーの輸送理論解釈 (情報論的学習理論と機械学習) -- (情報論的学習理論ワークショップ(IBIS2016))

    園田 翔, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   116 ( 300 ) 297 - 304  2016年11月

    CiNii

  • 重回帰を用いた高次局所潜在的次元推定 (情報論的学習理論と機械学習)

    日野 英逸, 藤木 淳, 赤穂 昭太郎, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   116 ( 209 ) 233 - 240  2016年09月

    CiNii

  • 深層学習のリッジレット解析にむけた取組み (ウェーブレット解析と信号処理)

    園田 翔, 村田 昇

    数理解析研究所講究録   2001   64 - 73  2016年07月

    CiNii

  • 確率質量関数の二次展開とポアソン誤差構造に基づくエントロピー推定 (情報論的学習理論と機械学習)

    日野 英逸, 赤穂 昭太郎, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   115 ( 511 ) 47 - 53  2016年03月

    CiNii

  • グレーボックスモデルとブートストラップフィルタによる不確定な製鋼プロセスの溶鋼温度予測

    AHMAD Iftikhar, 加納学, 長谷部伸治, 北田宏, 村田昇

    化学工学会秋季大会研究発表講演要旨集(CD-ROM)   47th   ROMBUNNO.N214  2015年09月

    J-GLOBAL

  • 二重スパース性に基づくマルチフレーム超解像 (パターン認識・メディア理解)

    加藤 利幸, 日野 英逸, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   115 ( 98 ) 119 - 124  2015年06月

    CiNii

  • 二重スパース性に基づくマルチフレーム超解像

    加藤 利幸, 日野 英逸, 村田 昇

    聴覚研究会資料 = Proceedings of the auditory research meeting   45 ( 4 ) 323 - 328  2015年06月

    CiNii

  • 時間推移する定常分布の潜在構造モデル化

    千葉 智暁, 日野 英逸, 赤穂 昭太郎, 村田 昇

    研究報告数理モデル化と問題解決(MPS)   2014 ( 6 ) 1 - 6  2014年12月

     概要を見る

    購買者が限定されているマーケットでの製品の売上や株式市場などをモデル化するには,有限な資源を複数の関係者が奪い合うような条件を想定することが必要となる.本稿では,この条件下で存在しうる,資源の移動経路の動的な潜在構造と,関係者が持つ資源の量の変動に着目することで,潜在構造と時系列の関係性を非斉時マルコフ連鎖の遷移確率行列とこれによる分布の遷移でモデル化し,この表現に基づき時間変化する潜在構造を多変量時系列から推定する手法を提案する.

    CiNii

  • 確率質量関数の二次展開と単回帰に基づくエントロピー推定 (情報論的学習理論と機械学習 情報論的学習理論ワークショップ)

    日野 英逸, 越島 健介, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   114 ( 306 ) 33 - 40  2014年11月

     概要を見る

    エントロピーは分布の高次統計量の情報を反映するため,データの特性の把握に有用である.本稿では,観測データから微分エントロピーを高精度に推定する手法を3通り提案する.提案する推定量は,複数の半径における超球内の確率質量関数を二次展開することで得られる確率密度関数値の,線形回帰による推定に基づく.人工データを用いた実験により,既存のエントロピー推定手法に対する提案手法の優位性を示す.

    CiNii

  • 予測・運用・制御の一貫したGEMSの電圧制御とHEMSの電熱運用手法との協調EMS手法の評価 (電力技術 電力系統技術合同研究会・(1)電力技術・電力系統技術一般,(2)分散電源)

    芳澤 信哉, 河野 俊介, 吉田 彬, 藤本 悠, 村田 昇, 若尾 真治, 田辺 新一, 天野 嘉春, 林 泰弘

    電気学会研究会資料. PE   2014 ( 66 ) 63 - 67  2014年09月

    CiNii

  • S0840101 快適性を考慮した家庭用エネルギーシステムの最適運用方策の検討([S084]分散型エネルギーシステム,動力エネルギーシステム部門)

    吉田 彬, 藤本 悠, 村田 昇, 若尾 真治, 田辺 新一, 天野 嘉春

    年次大会 : Mechanical Engineering Congress, Japan   2014   "S0840101 - 1"-"S0840101-5"  2014年09月

     概要を見る

    The main objectives of this study are to consider both thermal comfort and energy consumption on operational planning problem of residential energy system, and to handle uncertainty of energy demand and PV output for the future as a scenario-based stochastic programming problem. The energy system consists of photovoltaic power generator, electrical-driven room air-conditioner and PEM type fuel-cell cogeneration system. We introduce mathematical optimization theory for operational planning problem, because the targeted system has many alternative operations. As a result, the optimal operational planning problem, which is extended by considering PMV and future scenarios as constraints, concludes that the residential energy system achieves 33% of energy saving ratio in the case of 1.0 of PMV.

    CiNii

  • オラクル分布を用いたサンプリング学習アルゴリズム (パターン認識・メディア理解)

    園田 翔, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   114 ( 197 ) 137 - 142  2014年09月

     概要を見る

    ニューラルネットの新規学習法を提案する.Murataによるニューラルネットの積分表現に基づき,中間層素子パラメータの確率分布(オラクル分布)を具体的に計算するアルゴリズムを構築する.オラクル分布に従ってサンプルを生成することで,バックプロパゲーションの有力な初期値を与えられる.一般に,オラクル分布からのサンプリングは数値的に不安定だが,近似的に線形時間でサンプリングする方法を開発した.人工データおよび実データに対するベンチマークでは,正規分布を用いて初期化する方法と比較して,高速にバックプロパゲーションが収束することを示した.

    CiNii

  • オラクル分布を用いたサンプリング学習アルゴリズム

    園田 翔, 村田 昇

    研究報告コンピュータビジョンとイメージメディア(CVIM)   2014 ( 24 ) 1 - 6  2014年08月

     概要を見る

    二ューラルネットの新規学習法を提案する.Murata によるニューラルネットの積分表現に基づき,中間層素子パラメータの確率分布 (オラクル分布) を具体的に計算するアルゴリズムを構築する.オラクル分布に従ってサンプルを生成することで,バックプロパゲーションの有力な初期値を与えられる.一般に,オラクル分布からのサンプリングは数値的に不安定だが,近似的に線形時間でサンプリングする方法を開発した.人工データおよび実データに対するベンチマークでは,正規分布を用いて初期化する方法と比較して,高速にバックプロパゲーションが収束することを示した.A new sampling learning algorithm for neural networks is proposed. Based on the integral representation of neural networks, a practical algorithm for calculating an oracle probability distribution of hidden parameters is developed. The samples drawn from the oracle distribution would be good initial parameters for backpropagation. In general rigorous sampling from the oracle distribution holds numerical difficulty, a linear-time sampling algorithm is also developed. Numerical experiments showed that when hidden parameters were initialized by the oracle distribution, following backpropagation converged faster to better parameters than when parameters were initialized by a normal distribution.

    CiNii

  • 重み付き有向グラフモデリングによるスパイクデータ解析 (情報論的学習理論と機械学習)

    樋口 翔, 日野 英逸, 龍野 正実, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   114 ( 105 ) 193 - 200  2014年06月

     概要を見る

    近年の技術の発展により,脳からニューロンの発火活動のデータを得られるようになった.このデータを解析することで,脳の情報処理の仕組みを理解することが期待されている.脳は外部からの情報を処理する際,ニューロンの単独の活動で情報を処理しているのではなく,多数のニューロンが相互に影響を及ぼし合い,協調的な発火活動をすることで情報を処理していると考えられている.つまり,脳から得られるデータを解析するには,協調の様子を捉えることができるものが望ましい.また,ニューロンのシナプス結合には向きと強さ,及び興奮性・抑制性の違いが存在するため,これらを表現できることも求められる.本研究では脳内の多数のニューロンが相互に影響を与え合う様子を,重み付き有向グラフによりモデル化した上で,そのグラフ構造推定手法を提案する.ニューロンモデルから作成した擬似スパイクデータに提案手法を適用し,重み付き有向グラフを推定した結果を示す.

    CiNii

  • 複数粒子フィルタとモデル選択を用いたEEGデータの電流ダイポール推定 (情報論的学習理論と機械学習)

    金田 有紀, 園田 翔, 日野 英逸, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   114 ( 105 ) 91 - 96  2014年06月

     概要を見る

    本研究ではEEGデータから脳内の電流源の位置・モーメントおよび個数を近似ダイポールとして推定する.電流ダイポール推定問題をEEGの生成モデルの逆問題として定式化して,粒子フィルタを用いてダイポールの位置・モーメントを推定する.更に,ダイポール数を変えて複数の粒子フィルタを用意し,モデル選択規準によって適切なダイポール数を選択する.人工EEGデータ実験により,提案手法を用いて正しいダイポール数とそれらの位置・モーメントが推定できることを確認した.また,パターンリバーサルVEPに提案手法を適用して,生理学的知見から結果を考察した.その結果,実EEGデータに対しても提案手法によってダイポールの数・位置・モーメントを推定可能であることを示した.

    CiNii

  • マーク付き点過程間の距離計算手法と判別への応用 (情報論的学習理論と機械学習)

    高野 健, 小林 芽依, 日野 英逸, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   114 ( 105 ) 83 - 89  2014年06月

     概要を見る

    時系列データの表現のひとつとしてイベントの生起時刻とそのイベントに何らかの値が付随している「マーク付き点過程」がある.本研究ではイベントの生起時刻のみで表現される「マークなし点過程」間の既存の距離計算手法を,マーク付き点過程間の距離計算手法へと拡張する.また,その応用として地震データの解析を行う.ここでは与えられたデータを窓幅で区切り,その窓のあと12時間以内に地震が発生するか否かをクラスラベルとする二値のクラス判別問題を解く.特徴量には拡張したマーク付き点過程間の距離を用い,ランダムフォレストを用いて判別問題を解くことで拡張した距離計算手法の有効性を確認した.

    CiNii

  • C131 確率計画法を用いたエネルギー需要シナリオに対する家庭用PEFCシステムの最適運用方策の検討(OS4 省エネルギー・コジェネ・ヒートポンプ(3))

    吉田 彬, 小方 亮平, 村田 昇, 天野 嘉春

    動力・エネルギー技術の最前線講演論文集 : シンポジウム   2014 ( 19 ) 93 - 96  2014年06月

     概要を見る

    The subject of this study is to plan the operation of residential energy system with uncertain parameters based on ex-ante decision before uncertain parameters are realized. This paper applies a scenario-based stochastic programming framework to the operational planning problem having uncertain energy demand as parameters. Based on predicted energy demand scenarios, the operational strategies, which indicates the prime mover's start-stop status and the level of hot water tank, are decided. The decided operational strategies and realized values of energy demand are input to optimal control problem, which is formulated by dynamic programming. Finally we consider the operational performance of dynamic programming using the decided operational strategies based on predicted demand scenarios. As a result, the proposed stochastic programming framework decided the optimal strategies, and showed that effectiveness.

    CiNii

  • C132 家庭用PEFCシステムにおける給湯需要予測誤差が省エネルギー性に及ぼす影響の評価(OS4 省エネルギー・コジェネ・ヒートポンプ(3))

    小方 亮平, 吉田 彬, 村田 昇, 天野 嘉春

    動力・エネルギー技術の最前線講演論文集 : シンポジウム   2014 ( 19 ) 97 - 100  2014年06月

     概要を見る

    Nowadays, PEFC-CGS is getting attention as a distributed energy system. It has high efficiency, and a lot of studies of PEFC-CGS were reported. But almost of them didn't consider energy prediction error. In this study, in order to evaluate energy-saving performance of PEFC-CGS for residential use with energy prediction error, I made PEFC-CGS control low input optimal operational plan based on energy prediction value. Consequently, operational plan to start PEFC-CGS in the morning or avoiding unneeded power runup and low load operation of PEFC-CGS can make control low of PEFC-CGS improved.

    CiNii

  • 重み付き有向グラフモデリングによるスパイクデータ解析

    樋口 翔, 日野 英逸, 龍野 正実, 村田 昇

    研究報告数理モデル化と問題解決(MPS)   2014 ( 35 ) 1 - 8  2014年06月

     概要を見る

    近年の技術の発展により,脳からニューロンの発火活動のデータを得られるようになった.このデータを解析することで,脳の情報処理の仕組みを理解することが期待されている.脳は外部からの情報を処理する際,ニューロンの単独の活動で情報を処理しているのではなく,多数のニューロンが相互に影響を及ぼし合い,協調的な発火活動をすることで情報を処理していると考えられている.つまり,脳から得られるデータを解析するには,協調の様子を捉えることができるものが望ましい.また,ニューロンのシナプス結合には向きと強さ,及び興奮性・抑制性の違いが存在するため,これらを表現できることも求められる.本研究では脳内の多数のニューロンが相互に影響を与え合う様子を,重み付き有向グラフによりモデル化した上で,そのグラフ構造推定手法を提案する.ニューロンモデルから作成した擬似スパイクデータに提案手法を適用し,重み付き有向グラフを推定した結果を示す.With recent developments in multielectrode recording technology, neural spike data can be obtained from a brain. It is expected to understand the mechanism of information processing in a brain by analyzing these neural data. It is consider that a brain processes information by neural cooperative activity. In order to analyze neural data, the method that can take into account of neural cooperative activity is desirable. Since neural connections are asymmetric and have different connection strengths, it is required that the method is able to represent these features. In this paper, neural connections are represented by using the weighted directed graph, and a state that a lot of neurons affect each other is modeled. Then a method for estimating graph structures is proposed. Experimental results using artificial neural spike data show that the proposed method is able to estimate weighted directed graph structure.

    CiNii

  • 重み付き有向グラフモデリングによるスパイクデータ解析

    樋口 翔, 日野 英逸, 龍野 正実, 村田 昇

    研究報告バイオ情報学(BIO)   2014 ( 35 ) 1 - 8  2014年06月

     概要を見る

    近年の技術の発展により,脳からニューロンの発火活動のデータを得られるようになった.このデータを解析することで,脳の情報処理の仕組みを理解することが期待されている.脳は外部からの情報を処理する際,ニューロンの単独の活動で情報を処理しているのではなく,多数のニューロンが相互に影響を及ぼし合い,協調的な発火活動をすることで情報を処理していると考えられている.つまり,脳から得られるデータを解析するには,協調の様子を捉えることができるものが望ましい.また,ニューロンのシナプス結合には向きと強さ,及び興奮性・抑制性の違いが存在するため,これらを表現できることも求められる.本研究では脳内の多数のニューロンが相互に影響を与え合う様子を,重み付き有向グラフによりモデル化した上で,そのグラフ構造推定手法を提案する.ニューロンモデルから作成した擬似スパイクデータに提案手法を適用し,重み付き有向グラフを推定した結果を示す.With recent developments in multielectrode recording technology, neural spike data can be obtained from a brain. It is expected to understand the mechanism of information processing in a brain by analyzing these neural data. It is consider that a brain processes information by neural cooperative activity. In order to analyze neural data, the method that can take into account of neural cooperative activity is desirable. Since neural connections are asymmetric and have different connection strengths, it is required that the method is able to represent these features. In this paper, neural connections are represented by using the weighted directed graph, and a state that a lot of neurons affect each other is modeled. Then a method for estimating graph structures is proposed. Experimental results using artificial neural spike data show that the proposed method is able to estimate weighted directed graph structure.

    CiNii

  • 複数粒子フィルタとモデル選択を用いたEEGデータの電流ダイポール推定

    金田 有紀, 園田 翔, 日野 英逸, 村田 昇

    研究報告バイオ情報学(BIO)   2014 ( 15 ) 1 - 6  2014年06月

     概要を見る

    本研究では EEG データから脳内の電流源の位置・モーメントおよび個数を近似ダイポールとして推定する.電流ダイポール推定問題を EEG の生成モデルの逆問題として定式化して,粒子フィルタを用いてダイポールの位置・モーメントを推定する.更に,ダイポール数を変えて複数の粒子フィルタを用意し,モデル選択規準によって適切なダイボール数を選択する.人工 EEG データ実験により,提案手法を用いて正しいダイポール数とそれらの位置.モーメントが推定できることを確認した.また,パターンリバーサル VEP に提案手法を適用して,生理学的知見から結果を考察したその結果,実 EEG データに対しても提案手法によってダイポールの数・位置・モーメントを推定可能であることを示した.In this study, the location, the moment and the number of ionic current modeled as dipoles are estimated from EEG data. The source localization problem is formulated as an inverse problem of the EEG generative model and the location and moment of dipoles are estimated by multiple particle filters. The appropriate number of dipoles is selected by a model selection criterion. The proposed method is shown to estimate accurately the locations and moments of dipoles of appropriate numbers. The proposed method is also applied to pattern reversal VEP data and shown to estimate the locations and the moments and the number of the dipoles.

    CiNii

  • マーク付き点過程間の距離計算手法と判別への応用

    高野 健, 小林 芽依, 日野 英逸, 村田 昇

    研究報告バイオ情報学(BIO)   2014 ( 14 ) 1 - 7  2014年06月

     概要を見る

    時系列データの表現のひとつとしてイベントの生起時刻とそのイベントに何らかの値が付随している 「マーク付き点過程」 がある.本研究ではイベントの生起時刻のみで表現される 「マークなし点過程」 間の既存の距離計算手法を,マーク付き点過程間の距離計算手法へと拡張する.また,その応用として地震データの解析を行う.ここでは与えられたデータを窓幅で区切り,その窓のあと 12 時間以内に地震が発生するか否かをクラスラベルとする二値のクラス判別問題を解く.特徴量には拡張したマーク付き点過程間の距離を用い,ランダムフォレストを用いて判別問題を解くことで拡張した距離計算手法の有効性を確認した.A simple point process is a type of a random process expressed only by times at which events happen. A marked point process is an extension of a simple point process, which is expressed by times and values of events. Distance measures between simple and marked point processes are of importance for various data analyses. In this work, a method for calculating distance between time-windowed marked point processes is proposed. The proposed method is based on the extensions of conventional measure for simple point processes to marked point processes. The method combines various distance matrices of marked point processes by using the random forest algorithm taking into account the locality expressed by each distance. The proposed method is applied to seismic data analysis and shown to improve conventional methods.

    CiNii

  • ユーザ評価データを用いたアイテム選好度の推定 (情報論的学習理論と機械学習)

    望月 駿一, 藤本 悠, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   113 ( 476 ) 89 - 94  2014年03月

     概要を見る

    ユーザが店や商品などのアイテムに点数をつけたデータに基づき,他のユーザに対して好ましいアイテムを提示するサービスが存在する.このようにユーザが付けた点数から,アイテム間の確率的表現を求めるものとしてBradley-Terry (BT)モデルの応用が検討されている.アイテム間の優劣の確率的表現を与え,そのアイテムの真の良さ(選好度)を定量化する.本研究では,分割統合型の最適化を行うことで数万件のアイテムを扱う場合でも現実的な時間で選好度の推定を行う方法を提案し,人工データと実データを用いた実験によって推定可能性を示した.

    CiNii

  • スパイクデータ解析のためのグラフ構造モデリング (ニューロコンピューティング)

    樋口 翔, 野田 淳史, 日野 英逸, 村田 昇

    電子情報通信学会技術研究報告 = IEICE technical report : 信学技報   113 ( 111 ) 145 - 150  2013年06月

     概要を見る

    動物は脳内のニューロンの協調的な活動により情報を処理している.この協調の様子をグラフとして表現し,その構造を推定することは,脳内における情報処理の仕組みの理解につながる.ニューロンのシナプス結合には向きが存在するため,ニューロンの結合モデルとしては有向グラフが適切であり,また,グラフ構造推定手法としても有向グラフに適用可能な手法が要求される.本研究では,digraph Laplacianによって有向グラフを簡潔に表現し,グラフ上で情報が遷移する様子をモデル化する.さらに,モデルのパラメータ推定手法を提案する.ニューロンモデルから作成したデータを用いて,提案手法によって有向グラフ構造の推定が可能であることを実験的に示す.

    CiNii

  • スパイクデータ解析のためのグラフ構造モデリング

    樋口 翔, 野田 淳史, 曰野 英逸, 村田 昇

    研究報告バイオ情報学(BIO)   2013 ( 26 ) 1 - 6  2013年06月

     概要を見る

    動物は脳内のニューロンの協調的な活動により情報を処理している.この協調の様子をグラフとして表現し,その構造を推定することは,脳内における情報処理の仕組みの理解につながる.ニューロンのシナプス結合には向きが存在するため,ニューロンの結合モデルとしては有向グラフが適切であり,また,グラフ構造推定手法としても有向グラフに適用可能な手法が要求される.本研究では,digraph Laplacian によって有向グラフを簡潔に表現し,グラフ上で情報が遷移する様子をモデル化する.さらに,モデルのパラメータ推定手法を提案する.ニューロンモデルから作成したデータを用いて,提案手法によって有向グラフ構造の推定が可能であることを実験的に示す.Information in the brain is processed by neural cooperative activity. Estimation of neural graph structures is important to understand the mechanism of information processing in the brain. Since neural connections are asymmetric, directed graphs are appropriate to represent neural graph structures. In this paper, directed graphsare represented using the digraph Laplacian, and information transition on graphs is modeled by an exponential map of the digraph Laplacian. Moreover, a parameter estimation method is proposed. At last, the proposed method is experimentally shown to be able to estimate directed graph structures using artificial neural spike data.

    CiNii

  • スパースコーディングを用いたマルチフレーム超解像

    加藤利幸, 日野英逸, 村田昇

    研究報告コンピュータビジョンとイメージメディア(CVIM)   2013 ( 3 ) 1 - 9  2013年05月

     概要を見る

    複数の低解像度の観測画像から高解像度画像の推定を行う,マルチフレーム超解像の一手法を提案する.提案手法は高解像度画像を推定するために,信号のスパース表現を用いる.スパース表現を用いる超解像手法の従来手法の多くはシングルフレーム超解像の手法である.また,スパース表現を用いるマルチフレーム超解像の従来手法は,画像間のサブピクセル精度の位置関係を考慮しないものであったため,複数枚の画像を効率的に利用できなかった.提案手法は低解像度画像に対応する辞書を,画像の劣化過程に従って作成し,その際にサブピクセル精度の位置ずれを取り入れることでこの問題を解決した.観測された低解像度画像の位置関係は,サブピクセル精度のブロックマッチングを用いて求める.超解像に用いる低解像度パッチのサブセットはマッチングの結果として得られる類似度に応じて選択する.提案手法の優れている点は,入力画像の中の一部の画像しか超解像に利用できない場合は,その条件に応じて出来る限り良い推定画像を作成することができる点である.特に一つのみの低解像度パッチが選択される場合,提案手法はシングルフレームの超解像手法としても実行される.実画像を使用した実験により,提案手法は従来のシングルフレームやマルチフレームの超解像手法と同等もしくはより良い結果を示すことを確認した.

    CiNii

  • 溶鋼温度推定を目的としたグレイボックスモデルの比較

    AHMAD Iftikhar, 加納学, 長谷部伸治, 北田宏, 村田昇

    材料とプロセス(CD-ROM)   26 ( 1 ) ROMBUNNO.TO17  2013年03月

    J-GLOBAL

  • カーネル層別逆回帰のためのモデル選択手法

    日野英逸, 越島健介, 村田昇

    研究報告数理モデル化と問題解決(MPS)   2012 ( 6 ) 1 - 6  2012年11月

     概要を見る

    データの次元を適切に削減することは,計算コスト及び記憶領域の削減,データの本質的構造の把握に繋がる.特に回帰問題においては,データが拘束されている部分空間を推定する sufficient dimension reduction の問題が盛んに研究されている.その代表例である層別逆回帰は,説明変数が楕円分布に従う場合は,応答変数の値によって層別した説明変数の中心ベクトルを用いて線型次元削減行列を推定することの正当性が理論的に保証されている.層別逆回帰の拡張として,カーネルトリックを利用することで非線型次元削減部分空間を推定する手法も提案されているが,利用するカーネル関数を適切に選択する必要がある.本研究では,カーネル層別逆回帰に利用するカーネル関数の選択手法を提案する.カーネル関数を定めることで,付随する特徴空間における説明変数の分布が決定されることに着目し,説明変数が特徴空間で正規分布に従うようにカーネルを選択する.正規分布は楕円分布の一種であり,これにより層別逆回帰における仮定が満たされる.特徴空間における分布の正規性を特性関数によって評価し,カーネル関数の凸結合の結合係数を最適化するアルゴリズムを導出する.幾つかの実データを用いた実験から,提案手法の有用性を示す.

    CiNii

  • カーネル層別逆回帰のためのモデル選択手法

    日野英逸, 越島健介, 村田昇

    研究報告バイオ情報学(BIO)   2012 ( 6 ) 1 - 6  2012年11月

     概要を見る

    データの次元を適切に削減することは,計算コスト及び記憶領域の削減,データの本質的構造の把握に繋がる.特に回帰問題においては,データが拘束されている部分空間を推定する sufficient dimension reduction の問題が盛んに研究されている.その代表例である層別逆回帰は,説明変数が楕円分布に従う場合は,応答変数の値によって層別した説明変数の中心ベクトルを用いて線型次元削減行列を推定することの正当性が理論的に保証されている.層別逆回帰の拡張として,カーネルトリックを利用することで非線型次元削減部分空間を推定する手法も提案されているが,利用するカーネル関数を適切に選択する必要がある.本研究では,カーネル層別逆回帰に利用するカーネル関数の選択手法を提案する.カーネル関数を定めることで,付随する特徴空間における説明変数の分布が決定されることに着目し,説明変数が特徴空間で正規分布に従うようにカーネルを選択する.正規分布は楕円分布の一種であり,これにより層別逆回帰における仮定が満たされる.特徴空間における分布の正規性を特性関数によって評価し,カーネル関数の凸結合の結合係数を最適化するアルゴリズムを導出する.幾つかの実データを用いた実験から,提案手法の有用性を示す.

    CiNii

  • ランダムウォークに基づいたグラフ構造モデリング

    野田淳史, 石田諒, 日野英逸, 龍野正実, 赤穂昭太郎, 村田昇

    研究報告バイオ情報学(BIO)   2012 ( 21 ) 1 - 6  2012年11月

     概要を見る

    グラフ構造とは,データの変数間の関係を表す構造であり,鉄道路線,インターネット,神経回路網などが代表例として挙げられる.これらのグラフの変数間の依存関係が得られた際に,変数間の繋がりの強度を推定する問題は実用上重要であるが,ある変数間の依存関係には複数の変数から受けた影響が含まれているため,直接的な繋がりは分かりにくい.そこで本稿では.繋がりの有無のみを表現した行列でグラフを簡潔に表現し,そのグラフ上で情報が遷移する様子をモデル化することで,変数間の依存関係を近似する.モデルには,遷移の回数を表すパラメータとグラフ構造を表すパラメータが含まれており,それらを推定するアルゴリズムを提案する.最後に,提案手法の優位性を実験的に示す.

    CiNii

  • スパースコーディングにおける基底生成のための単一母基底の学習

    有竹 俊光, 日野 英逸, 村田 昇

    電子情報通信学会技術研究報告. IBISML, 情報論的学習理論と機械学習 = IEICE technical report. IBISML, Information-based induction sciences and machine learning   112 ( 279 ) 343 - 350  2012年10月

     概要を見る

    スパースコーディングは少数の基底の線形和で信号を表現する方法論である.信号の表現に利用する基底集合の選択は信号の近似性能を左右する重要な問題であり,観測した信号から基底を学習することで柔軟かつ適応的な信号の近似が可能となる.本研究では,ウェーブレットのように単一の母基底からシフトとスケーリングによって構造化した基底を生成する方法と,その母基底を学習する手法を提案する.また,提案手法を人工データ及び実データに対して適用し,提案手法の利点と特徴について述べる.

    CiNii

  • 漸近展開による近似精度の予測可能性

    野村 亮介, 日野 英逸, 村田 昇, 吉田 朋広

    電子情報通信学会技術研究報告. IBISML, 情報論的学習理論と機械学習 = IEICE technical report. IBISML, Information-based induction sciences and machine learning   112 ( 279 ) 319 - 325  2012年10月

     概要を見る

    デリバティブの価格付けはファイナンス分野で重要な問題である.高精度であるが計算コストの高いモンテカルロ法に対して,決定論的な近似に基づく漸近展開法が有望な手法である.本研究は種々の確率微分方程式モデルに対し,漸近展開法の有効性を判断することを目的とする.漸近展開法とモンテカルロ法の誤差率の大小を予測し,漸近展開法を適用する基準を作成することが可能であることを実験的に示す.

    CiNii

  • スパース表現の数理とその応用 (パターン認識・メディア理解)

    日野 英逸, 村田 昇

    電子情報通信学会技術研究報告 : 信学技報   112 ( 197 ) 133 - 142  2012年09月

     概要を見る

    スパースコーディングは生物の一次視覚野の情報処理を数学的にモデル化したものであり,与えられた画像を少数の基底の線型結合で表現する手法である.観測信号のスパース表現は,工学的にも効率的な情報の保持・伝達,あるいはノイズに対して頑健な情報表現を実現する手法として注目を集めている.本稿では,スパースコーディングを始めとする種々の行列分解手法の数理的側面を,その確率モデルを介して統一的に論じる.また,スパースコーディングの代表的なアルゴリズムと幾つかの応用を紹介する.

    CiNii

  • スパース表現の数理とその応用 (情報論的学習理論と機械学習)

    日野 英逸, 村田 昇

    電子情報通信学会技術研究報告 : 信学技報   112 ( 198 ) 133 - 142  2012年09月

     概要を見る

    スパースコーディングは生物の一次視覚野の情報処理を数学的にモデル化したものであり,与えられた画像を少数の基底の線型結合で表現する手法である.観測信号のスパース表現は,工学的にも効率的な情報の保持・伝達,あるいはノイズに対して頑健な情報表現を実現する手法として注目を集めている.本稿では,スパースコーディングを始めとする種々の行列分解手法の数理的側面を,その確率モデルを介して統一的に論じる.また,スパースコーディングの代表的なアルゴリズムと幾つかの応用を紹介する.

    CiNii

  • スパース表現の数理とその応用

    日野英逸, 村田昇

    研究報告コンピュータビジョンとイメージメディア(CVIM)   2012 ( 20 ) 1 - 10  2012年08月

     概要を見る

    スパースコーディングは生物の一次視覚野の情報処理を数学的にモデル化したものであり,与えられた画像を少数の基底の線型結合で表現する手法である.観測信号のスパース表現は,工学的にも効率的な情報の保持・伝達,あるいはノイズに対して頑健な情報表現を実現する手法として注目を集めている.本稿では,スパースコーディングを始めとする種々の行列分解手法の数理的側面を,その確率モデルを介して統一的に論じる.また,スパースコーディングの代表的なアルゴリズムと幾つかの応用を紹介する.

    CiNii

  • タンディッシュ内溶鋼温度制御用グレイボックスモデルの開発

    阪下翔太, 加納学, 北田宏, 村田昇

    計測自動制御学会制御部門大会(CD-ROM)   12th   ROMBUNNO.3GATSU14NICHI,DAI2SHITSU,16:00-17:40,SAKASHITA  2012年03月

    J-GLOBAL

  • Dirichlet過程と変分Bayes法による交叉率推定 (情報論的学習理論と機械学習)

    蓬田 裕菜, 村田 昇, 井上 真郷

    電子情報通信学会技術研究報告 : 信学技報   111 ( 275 ) 181 - 185  2011年11月

     概要を見る

    ヒトゲノムが解明され,ゲノム配列と形質にはどのような関連があるか,様々な研究がなされている.その中でも薬剤感受性や疾患との関連解析は,オーダーメイド医療の実現に向けた大きな課題である.現在では,これらの関連解析は,haplotype blockと比較する方が効率的であると知られてきている.したがってhaplotype blockを推定するために,組換hot spotを正しく検出することは大変重要である.本報告では組換hot spotを検出する為に,各SNP間における交叉率の推定を行う.従来の推定手法として,EMアルゴリズムが挙げられる.しかし,EMアルゴリズムでは局所解への収束が疑われることや,最適な混合数の選択が難しいという課題があった.そこで本報告では,Dirichlet過程(DP)と変分Bayes法(VB)を用いた推定手法を提案する.これらを用いることで,混合数を自動推定した.

    CiNii

  • ガウス性に基づく多重カーネル学習 (情報論的学習理論と機械学習)

    日野 英逸, 村田 昇

    電子情報通信学会技術研究報告 : 信学技報   111 ( 275 ) 99 - 104  2011年11月

     概要を見る

    カーネル法におけるカーネル関数の最適化手法として,多重力-ネル学習(MKL)が近年盛んに研究されている.カーネル法は様々な判別器の非線形化に適用されているが,その一つにFisherの判別分析(FDA)がある.2クラスのデータが同一の共分散行列を有する正規分布に従っている時,FDAによってBayes誤りを達成する最適な判別局面が得られることが知られている.本研究ではこの事実を利用し,ガウス性に基づくMKLの枠組みを提案する.カーネル関数に付随する特徴空間における正規性の尺度としては,経験特性関数を利用する.また,一般には無限次元空間となる特徴空間を扱うために,経験カーネル写像を利用する.ベンチマークデータを用いた判別実験の結果,提案する枠組みに基づくMKLアルゴリズムが良好な判別性能を有することが確認された.

    CiNii

  • Just-In-Time Modeling を用いた日射量予測における信頼度推定

    寺園 隆宏, 若尾 真治, 沈 浩洋, 日野 英逸, 村田 昇

    電気学会研究会資料. MES, メタボリズム社会・環境システム研究会 = The papers of Technical Meeting on Metabolism Society and Environmental Systems, IEE Japan   2011 ( 15 ) 37 - 42  2011年11月

    CiNii

  • ランダムサンプリングに基づく超曲面あてはめ

    藤木 淳, 赤穂 昭太郎, 日野 英逸, 村田 昇

    電子情報通信学会技術研究報告. PRMU, パターン認識・メディア理解   111 ( 48 ) 69 - 74  2011年05月

     概要を見る

    本稿では,N次元空間のデータに対してN-1次元超曲面をあてはめる問題について考察する.一般にこのあてはめ問題は特徴写像を介してM次元特徴空間における1次元低い超平面あてはめ問題に帰着される.このとき,もとの空間におけるユークリッド距離は特徴空間において重み附きユークリッド距離で近似できるので,もとの空間におけるユークリッド距離を反映した特徴空間におけるあてはめ問題は特徴空間において重み附きユークリッド距離に基づくあてはめ問題に帰着できる.一方,1次元低い超平面あてはめ問題について,重み附きユークリッド距離のk乗和を最小とする超平面あてはめは0&lt;k≦1のときに最適抽出可能である,すなわちM次元特徴空間において,M-1次元アフィン超平面をあてはめる場合はM個のデータ点を通る大域的最適解が存在する.本稿では,このことを利用したランダムサンプリングによりあてはめ問題を近似する手法である最小k乗偏差推定,及び最小二乗中央値推定を拡張した最小二乗α百分位点推定を提案する.これら提案手法と,同様にランダムサンプリングに基づく手法であるRANSACとの関係について議論する.

    CiNii

  • 取鍋およびタンディッシュ内溶鋼温度推定モデルの構築

    大倉才昇, 加納学, 北田宏, 村田昇

    材料とプロセス(CD-ROM)   24 ( 1 ) ROMBUNNO.TO8  2011年03月

    J-GLOBAL

  • ブートストラップフィルタによる溶鋼温度分布の予測と制御

    園田翔, 村田昇, 日野英逸, 進藤史裕, 北田宏, 加納学

    材料とプロセス(CD-ROM)   24 ( 1 ) ROMBUNNO.TO9  2011年03月

    J-GLOBAL

  • High performance prediction of molten steel temperature through gray-box model

    Toshinori Okura, Manabu Kano, Shinji Hasebe, Hiroshi Kitada, Noboru Murata

    Computing and Systems Technology Division - Core Programming Topic at the 2011 AIChE Annual Meeting   2   1153 - 1154  2011年  [査読有り]

  • マルチカーネル学習を用いた話者認識における最適化の検討

    小川 哲司, 日野 英逸, レイハニニマ, 村田 昇, 小林 哲則

    研究報告音声言語情報処理(SLP)   2010 ( 27 ) 1 - 6  2010年12月

     概要を見る

    本稿では,マルチカーネル学習を話者認識システムに適用した場合における,最適化アルゴリズムと認識性能の関係について調査を行った.話者認識システムにカーネル法を適用する場合,与えられたデータに対して適切なカーネル関数やパラメータを決定する必要がある.マルチカーネル学習は,複数のカーネル関数を凸結合することで,カーネル関数やパラメータを厳密に決定する必要性を減じることができる.本稿では,このマルチカーネル学習で用いる最適化基準および最適化アルゴリズムに焦点を当て,条件付きエントロピー最小化に基づくアルゴリズムと従来多く用いられているマージン最大化に基づくアルゴリズムを話者認識において比較した.その結果,条件付きエントロピー最小化に基づくシステムは,マージン最大化に基づくシステムの誤りを削減することがわかった.We investigated the relation between the optimization algorithm for multiple kernel learning (MKL) and the speaker recognition performance. Most of the kernel methods applied to speaker recognition systems require a suitable kernel function and its parameters to be determined for a given data set. In contrast, MKL eliminates the need for strict determination of the kernel function and parameters by using a convex combination of element kernels. In the present paper, we focused on the optimization criterion and algorithm applied to MKL. We compared an MKL algorithm based on conditional entoropy minimization (MCEM) with a conventional maximum-margin-based MKL algorithm in terms of speaker recognition accuracy; the MCEM-based system reduced the speaker error rate as compared to the maximum-margin-based system.

    CiNii

  • クロスエントロピー最適化を用いた株価予測値の安定化手法

    三浦 和起, 日野 英逸, 村田 昇

    研究報告バイオ情報学(BIO)   2010 ( 9 ) 1 - 6  2010年12月

     概要を見る

    時系列の予測は古くからある重要な問題であり,特に株価の予測は経済動向の予測や資産運用の指針として需要が高い.コンピュータ性能の発達と共に,学習理論を用いた経済時系列データに関する研究が活発に行われているが,株価のメカニズムを捉えることは依然として困難な問題である.本稿では,単一の予測モデルにより株価を一点で予測するのではなく,複数の予測モデルの学習を行い,各モデルに適切な重みを付けることで予測値の分散を低減する手法を提案する.基礎となる予測モデルは遺伝的プログラミングを用いて構成する.各予測モデルの重みは,学習用データと予測モデルの出力値とのクロスエントロピーが最小となるように定める.提案した予測手法の有用性を,人工データ及び日経平均株価の 1 分足の予測によって検証する.Prediction of time series data is a long standing important problem. Especially, prediction of stock price is much in demand for forecasting the economic trend and guideline for asset maintenance. Although there are growing number of studies on learning theory based time series prediction, the prediction of stock prices is still being very difficult task. In this study, the stock prices is predicted not only using one predictor, but using a set of predictors generated by the method of Genetic Programming (GP). Each element predictor is given non-negative weight, and the weight is optimized to minimize the cross entropy between the true learning stock prices and the weighted sum of predicted values. The proposed stock price prediction method is evaluated using both an artificial data and real-world stock price data.

    CiNii

  • 極射影平面を利用した放射対称歪曲の較正

    藤木 淳, 日野 英逸, 宇佐見 由美, 赤穂 昭太郎, 村田 昇

    電子情報通信学会技術研究報告. PRMU, パターン認識・メディア理解   109 ( 470 ) 377 - 382  2010年03月

     概要を見る

    魚眼レンズ等の全方位カメラにおける放射対称歪曲の較正パラメータは,一般的に測鉛線原理,つまり実空間における直線の像である歪曲直線が直線へと写像されるように推定される.これは歪曲画像から理想的なピンホールカメラで撮影された画像を推定することに相当する.筆者らはこの測鉛線原理に基づく線型演算で構成される較正手法を提案したが,その手法は視野角が1800以上のカメラには適用できないという欠点がある.そこで本稿では測鉛線原理を拡張した極射影測鉛線原理に基づく,視野角が1800以上のカメラにも適用できる線型演算で構成される較正手法を提案する.

    CiNii

  • 頑健なヤコビ核主成分分析に向けて

    藤木 涼, 赤穂 昭太郎, 曰野 英逸, 村田 昇

    研究報告コンピュータビジョンとイメージメディア(CVIM)   2009 ( 29 ) 191 - 196  2009年03月

     概要を見る

    元来の核主成分分析では特徴空間におけるユークリッド計量に基づいた最小二乗推定によって主たる部分空間を求めるが,ヤコビ核主成分分析では人力空間におけるユークリット計量にもとづいて計算される重みを用いた重み付き最小二乗推定によって主たる部分空間を求める.一般に (重み付き) 最小二乗推定量は外乱値に弱いため,本稿では RANSAC と x2 検定を利用して外乱値を検出することによりヤコビ核主成分分析の頑健化を行なう.Conventional kernel principal component analysis finds the principal subspace of the data by means of least squares estimation with respect to the metric defined in the feature space, on the other hand, Jacobian kernel principle component analysis (JKPCA) finds the principal subspace based on the metric in the input space by weighted least squares associated with the Jacobian of the feature map. For some applications such as image processing, it is more natural to utilize the metric in the input space, however, the weighted least squares estimation is sometimes too sensitive to outliers. To overcome this drawback of JKPCA and make it robust to outliers, an outlier detection method based on random sampling consensus (RANSAC) and X2 test is discussed, and its validity is confirmed by numerical experiments.

    CiNii

  • 頑健なヤコビ核主成分分析に向けて

    藤木 淳, 赤穂 昭太郎, 日野 英逸, 村田 昇

    電子情報通信学会技術研究報告. PRMU, パターン認識・メディア理解   108 ( 484 ) 191 - 196  2009年03月

     概要を見る

    元来の核主成分分析では特徴空間におけるユークリッド計量に基づいた最小二乗推定によって主たる部分空間を求めるが,ヤコビ核主成分分析では人力空間におけるユークリッド計量に基づいて計算される重みを用いた重み付き最小二乗推定によって主たる部分空間を求める.一般に(重み付き)最小二乗推定量は外乱値に弱いため,本稿ではRANSACとχ^2検定を利用して外乱値を検出することによりヤコビ核主成分分析の頑健化を行なう.

    CiNii

  • 主成分曲線のあてはめによる放射対称歪曲の較正

    藤木 淳, 赤穂 昭太郎, 日野 英逸, 村田 昇

    電子情報通信学会技術研究報告. PRMU, パターン認識・メディア理解   108 ( 363 ) 13 - 18  2008年12月

     概要を見る

    魚眼レンズなどがもつ放射方向の歪みを較正する際,直線の像が直線となるように歪みを表現する助変数を決定することが一般的に行なわれている.本稿では,放射対称歪曲の較正問題を直線の像である曲線に対して主成分曲線をあてはめる問題として捉え,核主成分分析を利用して定式化する.

    CiNii

  • F-050 GESによる補正を行った情報量に基づくパラメタ推定法の評価(F分野:人工知能・ゲーム)

    藤本 悠, 村田 昇

    情報科学技術フォーラム一般講演論文集   6 ( 2 ) 461 - 464  2007年08月

    CiNii

  • LF_001 e-混合モデルの推定(F分野:人工知能・ゲーム)

    藤本 悠, 村田 昇

    情報科学技術レターズ   5   93 - 96  2006年08月

    CiNii

  • LH-004 少数サンプルを基にした独立な確率表の混合表現(H分野:生体情報科学)

    藤本 悠, 村田 昇

    情報科学技術レターズ   4   129 - 130  2005年08月

    CiNii

  • G-022 ベイジアンネットによる即興音楽生成システム : 少ない曲列からのモデル推定(G.音声・音楽)

    藤本 悠, 村田 昇

    情報科学技術フォーラム一般講演論文集   3 ( 2 ) 395 - 396  2004年08月

    CiNii

  • D-2-1 L_pノルムを用いたサポート・ベクトル・マシンの学習特性(D-2. ニューロコンピューティング)

    池田 和司, 村田 昇, 大西 隆治, 青石 勉

    電子情報通信学会総合大会講演論文集   2004 ( 1 ) 11 - 11  2004年03月

    CiNii

  • 推定量を組み合わせる-バギングとブースティング, パターン認識と学習の統計学-新しい概念と手法

    村田昇

    統計科学のフロンティア    2003年

    CiNii

▼全件表示

共同研究・競争的資金等の研究課題

  • 統計的因果推論と非正規性

    科学研究費助成事業(大阪大学)  科学研究費助成事業(基盤研究(C))

  • 逐次周辺尤度オンライン変化検出:粒子フィルタ的接近

    科学研究費助成事業(早稲田大学)  科学研究費助成事業(基盤研究(C))

  • 情報量に基づく重み付きデータ縮約

    科学研究費助成事業(早稲田大学)  科学研究費助成事業(挑戦的萌芽研究)

  • 統計的正則化理論と神経生理学:脳科学との接点

    科学研究費助成事業(大阪大学)  科学研究費助成事業(挑戦的萌芽研究)

  • 神経回路モデルに基づく学習情報処理機構の研究

    科学研究費助成事業(東京大学)  科学研究費助成事業(一般研究(C))

  • 「検証的」独立成分分析と独立因子分析の研究

    科学研究費助成事業(大阪大学)  科学研究費助成事業(基盤研究(C))

  • 統計学,ニューラルネット,機械学習の新しい融合

    科学研究費助成事業(統計数理研究所)  科学研究費助成事業(基盤研究(B))

  • 「署名・顔・声」マルチモーダルバイオメトリック個人認証の基礎研究

    科学研究費助成事業(早稲田大学)  科学研究費助成事業(基盤研究(C))

  • ベイズ統計と集団学習の統計的解析

    科学研究費助成事業(早稲田大学)  科学研究費助成事業(特定領域研究)

  • Between ICA and SEM

    科学研究費助成事業(大阪大学)  科学研究費助成事業(基盤研究(C))

  • ゲノム多様性解析のための新しい統計的方法

    科学研究費助成事業(統計数理研究所)  科学研究費助成事業(基盤研究(B))

▼全件表示

 

現在担当している科目

▼全件表示