2022/10/01 更新

写真a

イノウエ マサト
井上 真郷
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

  • 附属機関・学校   グローバルエデュケーションセンター

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

    理工学術院総合研究所   兼任研究員

学歴

  •  
    -
    2003年

    京都大学   医学研究科   外科系専攻  

  •  
    -
    2003年

    京都大学  

学位

  • 京都大学   博士(医学)

 

研究分野

  • 情報学基礎論

  • ソフトコンピューティング

  • 統計科学

  • 感性情報学

  • 生命、健康、医療情報学

研究キーワード

  • バイオインフォマティクス

  • 情報統計力学

  • 確率的情報処理

  • Bioinformatics

  • Probabilistic Information Processing

論文

  • An Innovative Voice Analyzer “ VA ” Smart Phone Program for Quantitative Analysis of Voice Quality

    Tsuyoshi Kojima, Shintaro Fujimura, Ryusuke Hori, Yusuke Okanoue, Kazuhiko Shoji, Masato Inoue

    Journal of Voice   33 ( 5 ) 642 - 648  2018年  [査読有り]

    DOI PubMed

  • Traffic Velocity Estimation From Vehicle Count Sequences

    Takayuki Katsuki, Tetsuro Morimura, Masato Inoue

    IEEE TRANSACTIONS ON INTELLIGENT TRANSPORTATION SYSTEMS   18 ( 7 ) 1700 - 1712  2017年07月  [査読有り]

     概要を見る

    Traffic velocity is a fundamental metric for inferring traffic conditions. This paper proposes a new velocity estimation approach from temporal sequences of vehicle count that does not require tracking any vehicles or using any labeled data. It is useful for measuring traffic velocities with low quality and inexpensive sensors such as web cameras in general use. We formalize the task as a density estimation problem by introducing a new model for temporal sequences of vehicle counts wherein the correlation between the sequences is directly related to the traffic velocity. We also derive a sampling-based algorithm for the density estimation. We show the effectiveness of our method on artificial and real-world data sets.

    DOI

  • A latent class model for competing risks

    M. Rowley, H. Garmo, M. Van Hemelrijck, W. Wulaningsih, B. Grundmark, B. Zethelius, N. Hammar, G. Walldius, M. Inoue, L. Holmberg, A. C. C. Coolen

    STATISTICS IN MEDICINE   36 ( 13 ) 2100 - 2119  2017年06月  [査読有り]

     概要を見る

    Survival data analysis becomes complex when the proportional hazards assumption is violated at population level or when crude hazard rates are no longer estimators of marginal ones. We develop a Bayesian survival analysis method to deal with these situations, on the basis of assuming that the complexities are induced by latent cohort or disease heterogeneity that is not captured by covariates and that proportional hazards hold at the level of individuals. This leads to a description from which risk-specific marginal hazard rates and survival functions are fully accessible, 'decontaminated' of the effects of informative censoring, and which includes Cox, random effects and latent classmodels as special cases. Simulated data confirm that our approach can map a cohort's substructure and remove heterogeneity-induced informative censoring effects. Application to data from the Uppsala Longitudinal Study of Adult Men cohort leads to plausible alternative explanations for previous counter-intuitive inferences on prostate cancer. The importance of managing cardiovascular disease as a comorbidity in women diagnosed with breast cancer is suggested on application to data from the Swedish Apolipoprotein Mortality Risk Study. Copyright (C) 2017 John Wiley & Sons, Ltd.

    DOI

  • Bayesian Regression Selecting Valuable Subset from Mixed Bag Training Data

    Takayuki Katsuki, Masato Inoue

    2016 23RD INTERNATIONAL CONFERENCE ON PATTERN RECOGNITION (ICPR)     2580 - 2585  2016年  [査読有り]

     概要を見る

    This paper addresses a problem in which we learn a regression model from sets of training data. Each of the sets has an only single label, and only one of the training data in the set reflects the label. This is particularly the case when the label is attached to a group of data, such as time-series data. The label is not attached to the point of the sequence but rather attached to particular time window of the sequence. As such, a small part of the time window likely reflects the label, whereas the other larger part of the time window likely does not reflect it. We design an algorithm for estimating which of the training data in each of the sets corresponds to the label, as well as for training the regression model on the basis of Bayesian modeling and posterior inference with variational Bayes. Our experimental results show that our approach perform better than baseline methods on an artificial dataset and on a real-world dataset.

    DOI

  • A novel high-speed digital imaging system for assessing vocal fold vibration

    Kaneko K, Watanabe T, Inoue M, Takahashi H

    Acta Medica Nagasakiensia   59 ( 1 ) 37 - 40  2014年  [査読有り]

     概要を見る

    Observing vocal fold vibration is essential when investigating the causes of hoarseness of voice and determining the appropriate treatment. However, in Japan, it is difficult at present to obtain commercial high-speed digital imaging (HSDI) systems for this purpose because of a lack of marketing routes. Accordingly, we devised an HSDI system using a high-speed camera (monochrome HiSpec1). We chose this camera as it has been an established instrument in industrial and laboratory use; it also provided excellent light sensitivity, and used a 400-watt xenon light source for the acquisition of bright and clear images. The HSDI system enabled video capture of up to 4.4 s with our usual setting of 422 × 384 pixels at 3000 frames per second (fps). It can also capture videos in various settings such as 396 × 256 pixels at 4000 fps. In addition, we equipped the system with devices for concurrent audio waveform acquisition. Our novel laryngeal HSDI system could be a useful clinical tool for vocal fold examination in patients with voice problems when access to commercially available devices is limited.

    DOI CiNii

  • Low-Cost High-Speed Imaging System for Observing Vocal Fold Vibration in Voice Disorders

    Kenichi Kaneko, Koichi Sakaguchi, Masato Inoue, Haruo Takahashi

    ORL-JOURNAL FOR OTO-RHINO-LARYNGOLOGY HEAD AND NECK SURGERY   74 ( 4 ) 208 - 210  2012年09月  [査読有り]

     概要を見る

    Purpose:The aim of this study was to establish a method to observe vocal fold vibration using a low-cost high-speed laryngeal imaging system. Procedures: We assembled a highspeed imaging system with a consumer digital camera and a rigid laryngeal endoscope. The camera can shoot digital images at a rate of 1,200 frames per second and be purchased for about USD 1,000 in Japan. Results: We examined the normal and pathological vocal folds of 215 subjects with our new system and analyzed the vocal fold vibration in these subjects by playback of a video and kymograph images. Conclusions: Our high-speed laryngeal imaging system is highly cost-effective and can be a useful tool for examining the vocal folds of patients with voice disorders. Copyright (C) 2012 S. Karger AG, Basel

    DOI

  • Posterior-Mean Super-Resolution With a Causal Gaussian Markov Random Field Prior

    Takayuki Katsuki, Akira Torii, Masato Inoue

    IEEE TRANSACTIONS ON IMAGE PROCESSING   21 ( 7 ) 3182 - 3193  2012年07月  [査読有り]

     概要を見る

    We propose a Bayesian image super-resolution (SR) method with a causal Gaussian Markov random field (MRF) prior. SR is a technique to estimate a spatially high-resolution image from given multiple low-resolution images. An MRF model with the line process supplies a preferable prior for natural images with edges. We improve the existing image transformation model, the compound MRF model, and its hyperparameter prior model. We also derive the optimal estimator-not the joint maximum a posteriori (MAP) or the marginalized maximum likelihood (ML) but the posterior mean (PM)-from the objective function of the L2-norm-based (mean square error) peak signal-to-noise ratio. Point estimates such as MAP and ML are generally not stable in ill-posed high-dimensional problems because of overfitting, whereas PM is a stable estimator because all the parameters in the model are evaluated as distributions. The estimator is numerically determined by using the variational Bayesian method. The variational Bayesian method is a widely used method that approximately determines a complicated posterior distribution, but it is generally hard to use because it needs the conjugate prior. We solve this problem with simple Taylor approximations. Experimental results have shown that the proposed method is more accurate or comparable to existing methods.

    DOI

  • POSTERIOR MEAN SUPER-RESOLUTION WITH A COMPOUND GAUSSIAN MARKOV RANDOM FIELD PRIOR

    Takayuki Katsuki, Masato Inoue

    2012 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP)     841 - 844  2012年  [査読有り]

     概要を見る

    This manuscript proposes a posterior mean (PM) super-resolution (SR) method with a compound Gaussian Markov random field (MRF) prior. SR is a technique to estimate a spatially high-resolution image from observed multiple low-resolution images. A compound Gaussian MRF model provides a preferable prior for natural images that preserves edges. PM is the optimal estimator for the objective function of peak signal-to-noise ratio (PSNR). This estimator is numerically determined by using variational Bayes (VB). We then solve the conjugate prior problem on VB and the exponential-order calculation cost problem of a compound Gaussian MRF prior with simple Taylor approximations. In experiments, the proposed method roughly overcomes existing methods.

    DOI

  • New VA software program quantitatively analyzes voice quality

    Masanobu Mizuta, Kazuhiko Shoji, Ryusuke Hori, Mami Morita, Yusuke Okanoue, Shintaro Fujimura, Tsuyoshi Kojima, Masato Inoue

    Practica Oto-Rhino-Laryngologica   104 ( 4 ) 297 - 302  2011年04月  [査読有り]

     概要を見る

    The voice analyzer (VA) software program we developed for quantitatively analyzing voice quality uses zero-cross-picking to find individual basic pitch periods. Two such periods extracted from a voice waveform were analyzed to calculate the acoustic energy of harmonics separately from that of noise components. The program determines the ratio of harmonics to noise (Ra), FO, jitter, shimmer, APQ, and PPQ. Sustained vowel/a/from 28 subjects without evident vocal problems such as hoarseness and 49 subjects with hoarseness were examined. Average female and male Ra among the nonhoarse were 31. 9 dB and 34. 4 dB, jitter 0. 254% and 0. 246%, and shimmer 0.168% and 0.116%. Among those with hoarseness, all parameter scored worse than in the nonhoarse in proportion to auditory impression.

    DOI

  • A haplotype inference method based on sparsely connected multi-body Ising model

    Masashi Kato, Qian Ji Gao, Hiroshi Chigira, Hiroyuki Shindo, Masato Inoue

    INTERNATIONAL WORKSHOP ON STATISTICAL-MECHANICAL INFORMATICS 2010 (IW-SMI 2010)   233  2010年

     概要を見る

    Statistical haplotype inference is an indispensable technique in the field of medical science. The method usually has two steps: inference of haplotype frequencies and inference of diplotype for each subject. The first step can be done by using the expectation-maximization (EM) algorithm, but it incurs an unreasonably large calculation cost when the number of single-nucleotide polymorphism (SNP) loci of concern is large. In this article, we describe an approximate probabilistic model of haplotype frequencies. The model is constructed by using several distributions of nearby local SNPs. This approximation seems good because SNPs are generally more strongly correlated when they are close to one another on a chromosome. To implement this approach, we use a log linear model, the Walsh-Hadamard transform, and a combinatorial optimization method. Artificial data suggested that the overall haplotype inference of our method is good if there are nine or more local consecutive SNPs. Some minor problems should be dealt with before this method can be applied to real data.

    DOI J-GLOBAL

  • Identifying hidden confounders in gene networks by Bayesian networks

    Tomoya Higashigaki, Kaname Kojima, Rui Yamaguchi, Masato Inoue, Seiya Imoto, Satoru Miyano

    10th IEEE International Conference on Bioinformatics and Bioengineering 2010, BIBE 2010     168 - 173  2010年  [査読有り]

     概要を見る

    In the estimation of gene networks from microarray gene expression data, we propose a statistical method for quantification of the hidden confounders in gene networks, which were possibly removed from the set of genes on the gene networks or are novel biological elements that are not measured by microarrays. Due to high computational cost of the structural learning of Bayesian networks and the limited source of the microarray data, it is usual to perform gene selection prior to the estimation of gene networks. Therefore, there exist missing genes that decrease accuracy and interpretability of the estimated gene networks. The proposed method can identify hidden confounders based on the conflicts of the estimated local Bayesian network structures and estimate their ideal profiles based on the proposed Bayesian networks with hidden variables with an EM algorithm. From the estimated ideal profiles, we can identify genes which are missing in the network or suggest the existence of the novel biological elements if the ideal profiles are not significantly correlated with any expression profiles of genes. To the best of our knowledge, this research is the first study to theoretically characterize missing genes in gene networks and practically utilize this information to refine network estimation. © 2010 IEEE.

    DOI

  • Decoding Algorithm of Low-density Parity-check Codes based on Bowman-Levin Approximation

    Ken-ichi Tamura, Miho Komiya, Masato Inoue, Yoshiyuki Kabashima

    NEW GENERATION COMPUTING   27 ( 4 ) 347 - 363  2009年08月

     概要を見る

    Belief propagation (BP) and the concave-convex procedure (CCCP) are algorithms that use the Bethe free energy as a cost function and are used to solve information processing tasks. We have developed a new algorithm that also uses the Bethe free energy but changes the roles of the master and slave variables. This is called the Bowman-Levin (BL) approximation in the domain of statistical physics. When we applied the BL approximation to decode the regular low-density parity-check (LDPC) codes over an additive white Gaussian noise (AWGN) channel, its average performance was roughly similar to that of either BP or CCCP, but slightly outperforms them if the vast calculation cost is not prohibitive. This implies that our algorithm based on the BL approximation can be successfully applied to other problems to which BP or CCCP have already been applied. We also found that the decoding dynamics of the BL algorithm particularly depend on the number of inner loops. These differences from BP may be important in understanding the complicated landscape of the Bethe free energy.

    DOI CiNii J-GLOBAL

  • Grouping preprocess for haplotype inference from SNP and CNV data

    Hiroyuki Shindo, Hiroshi Chigira, Tomoyo Nagaoka, Naoyuki Kamatani, Masato Inoue

    INTERNATIONAL WORKSHOP ON STATISTICAL-MECHANICAL INFORMATICS 2009 (IW-SMI 2009)   197  2009年

     概要を見る

    The method of statistical haplotype inference is an indispensable technique in the field of medical science The authors previously reported Hardy-Weinberg equilibrium-based haplotype inference that could manage single nucleotide polymorphism (SNP) data We recently extended the method to cover copy number variation (CNV) data Haplotype inference from mixed data is important because SNPs and CNVs are occasionally in linkage disequilibrium. The idea underlying the proposed method is simple, but the algorithm for it needs to be quite elaborate to reduce the calculation cost Consequently, we have focused on the details on the algorithm in this study. Although the main advantage of the method is accuracy, in that it does not use any approximation, its main disadvantage is still the calculation cost, which is sometimes intractable for large data sets with missing values.

    DOI J-GLOBAL

  • Vibration of the vocal folds observed using a high-speed movie system with consumer digital video camera

    Ken-Ichi Kaneko, Koichi Sakaguchi, Fujinobu Tanaka, Atsushi Takáno, Shinya Jinnouchi, Hideaki Nishi, Haruo Takahashi, Masato Inoue

    Practica Oto-Rhino-Laryngologica   102 ( 5 ) 379 - 384  2009年  [査読有り]

     概要を見る

    Observing vibration of the vocal folds is essential to find causes of hoarseness. We assembled the high-speed movie system from a consumer digital video camera (EXILIM PRO EX-Fl, Casio Computer Co., LTD., Tokyo) and a conventional rigid laryngeal endoscope. This system enables us to observe each vocal fold vibration at a frame rate of 1, 200 images per second. We examined the vocal folds of 10 healthy subjects and two patients with hemilateral vocal fold paralysis with our system. The high-speed movies were saved in the camera as MOV format files. On slow playback of the movies and kymographic images constructed from the movie files, the vocal fold vibrations of healthy subjects showed regularity, symme-try, complete glottal closure, normal amplitude and normal mucosal waves. However, those of the sub-jects with vocal fold paralysis showed irregularity, asymmetry, incomplete glottal closure, abnormal am-plitude and abnormal mucosal waves. Our system will be useful for a routine clinical examination of the vocal folds.

    DOI

  • Grouping preprocess to accurately extend application of EM algorithm to haplotype inference

    Hiroyuki Shindo, Hiroshi Chigira, Junji Tanaka, Naoyuki Kamatani, Masato Inoue

    JOURNAL OF HUMAN GENETICS   53 ( 8 ) 747 - 756  2008年08月

     概要を見る

    Haplotype inference is an indispensable technique in medical science, especially in genome-wide association studies. Although the conventional method of inference using the expectation-maximization (EM) algorithm by Excoffier and Slatkin is one standard approach, as its calculation cost is an exponential function of the maximum number of heterozygous loci, it has not been widely applied. We propose a method of haplotype inference that can empirically accommodate up to several tens of single nucleotide polymorphism loci in a single haplotype block while maintaining criteria that are exactly equivalent to those of the EM algorithm. The idea is to reduce the cost of calculating the EM algorithm by using a haplotype-grouping preprocess exploiting the symmetrical and inclusive relationships of haplotypes based on the Hardy-Weinberg equilibrium. Testing of the proposed method using real data sets revealed that it has a wider range of applications than the EM algorithm.

    DOI PubMed CiNii J-GLOBAL

  • Bayesian-optimal image reconstruction for translational-symmetric filters

    Satohiro Tajima, Masato Inoue, Masato Okada

    JOURNAL OF THE PHYSICAL SOCIETY OF JAPAN   77 ( 5 ) 1 - 54803  2008年05月

     概要を見る

    Translational-symmetric filters provide a foundation for various kinds of image processing. When a filtered image containing noise is observed, the original one can be reconstructed by Bayesian inference. Furthermore, hyperparameters such as the smoothness of the image and the noise level in the communication channel through which the image observed can be estimated from the observed image by setting a criterion of maximizing marginalized likelihood. In this article we apply a diagonalization technique with the Fourier transform to this image reconstruction problem. This diagonalization not only reduces computational costs but also facilitates theoretical analyses of the estimation and reconstruction performances. We take as an example the Mexican-hat shaped neural cell receptive field seen in the early visual systems of animals, and we compare the reconstruction performances obtained under various hyperparameter and filter parameter conditions with each other and with the corresponding performances obtained under no-filter conditions. The results show that the using a Mexican-hat filter can reduce reconstruction error.

    DOI J-GLOBAL

  • LDPC decoding dynamics from a PCA viewpoint

    Shinpei Hara, Yuta Akira, Eisuke Ishii, Masato Inoue, Masato Okada

    Interdisciplinary Information Sciences   13 ( 1 ) 43 - 48  2007年

     概要を見る

    Low-density parity-check (LDPC) code has recently become of great interest. The statistical mechanics approach has been used to reveal some characteristics of LDPC in the thermodynamic limit. In this paper, we analyze this system for finite size rather than within the thermodynamic limit through a principal component analysis (PCA) approach. Specifically, both the decoding dynamics of belief propagation (BP) and the phases of the system are visualized and discussed. The result implies that the decoding dynamics roughly corresponds to the system temperature we introduced, and this system has several phases such as ferromagnetic, paramagnetic, and 1RSB spin-glass phases.

    DOI CiNii J-GLOBAL

  • Analysis method combining Monte Carlo simulation and principal component analysis ―application to Sourlas code―

    Masato INOUE, Koji HUKUSHIMA, Masato OKADA

    Journal of the Physical Society of Japan   75 ( 8 ) 1 - 84003  2006年08月

    DOI J-GLOBAL

  • Isolation of gene sets affected specifically by polyglutamine expression: implication of the TOR signaling pathway in neurodegeneration

    B Nelson, S Nishimura, H Kanuka, E Kuranaga, M Inoue, G Hori, H Nakahara, M Miura

    CELL DEATH AND DIFFERENTIATION   12 ( 8 ) 1115 - 1123  2005年08月

     概要を見る

    Transcriptional dysregulation as a result of sequestration of essential transcription factors into protein aggregates formed by polyglutamine (polyQ) expansions can lead to late-onset progressive neurodegeneration. DNA microarray analysis of Drosophila expressing polyQ in the compound eye over time revealed large numbers of transcriptional changes at the earliest stages of the disease including repression of the transient receptor potential calcium channels in a polyQ-induced cell death specific manner. While significant differences in expression profiles were found between the Drosophila compound eye and polyQ-sensitive neural cells, a number of possible key overlapping regulators were extracted. Among these, PDK1 was shown to act as a mediator for polyQ-toxicity, suggesting the involvement of the TOR pathway in polyQ-induced neurodegeneration.

    DOI J-GLOBAL

  • Slow dynamics due to singularities of hierarchical learning machines

    H Park, M Inoue, M Okada

    PROGRESS OF THEORETICAL PHYSICS SUPPLEMENT   157 ( 157 ) 275 - 279  2005年  [査読有り]

     概要を見る

    Recently, slow dynamics in learning of neural networks has been known to be closely related to singularities, which exist in parameter spaces of hierarchical learning models. To show the influence of singular structure on learning dynamics, we take statistical mechanical approaches and investigate online-learning dynamics under various learning scenario with different relationship between optimum and singularities. From the investigation, we found a quasi-plateau phenomenon which differs from the well known plateau. The quasi-plateau and plateau become extremely serious when an optimal point is in a neighborhood of a singularity. The quasi-plateau and plateau disappear in the natural gradient learning, which takes singular structures into account and uses Riemannian measure for the parameter space.

  • A PCA approach to Sourlas code analysis

    M Inoue, K Hukushima, M Okada

    PROGRESS OF THEORETICAL PHYSICS SUPPLEMENT   157 ( 157 ) 246 - 249  2005年  [査読有り]

     概要を見る

    The statistical mechanical approach is a powerful method for understanding large degrees of freedom problems, such as those related to the spin-glass model, but its application is usually limited to the class of mean-field models. We try a new general and computational approach - instead of an exact calculation of the Boltzmann distribution, we use an empirical spin state distribution obtained through simulation and extract potentially useful axes by principal component analysis (PCA). We adopted a three-body Sourlas code to evaluate this PCA approach compared with existing replica theory. The empirical spin distribution projected to these PCA axes showed distinctive patterns corresponding to the phase given by the replica method. Moreover, the first principal component roughly coincided with one of the order parameters (averaged spin) under a certain condition. These results suggest that this PCA approach could be effective even in more complicated systems that we cannot investigate analytically.

  • Improved parameter estimation for variance-stabilizing transformation of gene-expression microarray data

    Masato Inoue, Shin-Ichi Nishimura, Gen Hori, Hiroyuki Nakahara, Michiko Saito, Yoshihiro Yoshihara, Shun-Ichi Amari

    Journal of Bioinformatics and Computational Biology   2 ( 4 ) 669 - 679  2004年12月

     概要を見る

    A gene-expression microarray datum is modeled as an exponential expression signal (log-normal distribution) and additive noise. Variance-stabilizing transformation based on this model is useful for improving the uniformity of variance, which is often assumed for conventional statistical analysis methods. However, the existing method of estimating transformation parameters may not be perfect because of poor management of outliers. By employing an information normalization technique, we have developed an improved parameter estimation method, which enables statistically more straightforward outlier exclusion and works well even in the case of small sample size. Validation of this method with experimental data has suggested that it is superior to the conventional method. © Imperial College Press.

    DOI PubMed J-GLOBAL

  • Dynamics of the adaptive natural gradient descent method for soft committee machines

    Masato Inoue, Hyeyoung Park, Masato Okada

    Physical Review E   69 ( 5,Pt.2 ) 1 - 56120  2004年

    DOI J-GLOBAL

  • A case of giant calcifying epithelioma (pilomatrixoma) in the subauricular region

    Kazuhisa Yuki, Ryo Asato, Masato Inoue, Ichiro Tateya, Kennichi Kaneko, Kazuhiko Shoji, Hisayoshi Kojima

    Practica Oto-Rhino-Laryngologica   97 ( 8 ) 731 - 734  2004年  [査読有り]

     概要を見る

    Calcifying epithelioma (pilomatrixoma) is a relatively uncommon benign tumor. It usually presents as a slow-growing subcutaneous or intradermal nodule measuring less than 3 cm. In this report, we present a 20-year-old male with giant pilomatrixoma in the subauricular region. We performed en bloc resection of the tumor, which measured 12 x 9 cm. Histopathologically, this tumor did not show any malignant findings. There has not been any evidence of recurrence since the treatment.

    DOI

  • Learning dynamics of neural networks with singularity - Standard gradient vs. natural gradient

    H Park, M Inoue, M Okada

    PRICAI 2004: TRENDS IN ARTIFICIAL INTELLIGENCE, PROCEEDINGS   3157   282 - 291  2004年  [査読有り]

     概要を見る

    In hierarchical models, such as neural networks, there exist complex singular structures. The singularity is known to affect estimation performances and learning dynamics of the models. Recently, there have been a number of studies on properties of obtained estimators for the models, but there are few studies on the dynamical properties of learning used for obtaining the estimators. Using two-layer neural networks, we investigate influences of singularities on dynamics of standard gradient learning and natural gradient learning under various learning conditions. In the standard gradient learning, we found a quasi-plateau phenomenon, which is severer than the well known plateau in some cases. The slow convergence due to the quasi-plateau and plateau becomes extremely serious when an optimal point is in a neighborhood of a singularity. In the natural gradient learning, however, the quasi-plateau and plateau are not observed and convergence speed is hardly affected by singularity.

  • Stimulus-induced behavior in F1 hybrids of seizure-sensitive and seizure-resistant gerbils

    A Seto-Ohshima, S Kitajima, M Ito, M Inoue, YL Murashima, K Yamakawa, S Itohara

    ZOOLOGICAL SCIENCE   20 ( 12 ) 1439 - 1445  2003年12月

     概要を見る

    We previously established two strains of Mongolian gerbil: a seizure-sensitive strain, established by selective inbreeding for motor seizures elicited by a stimulus called the S method and a seizure-resistant strain that does not exhibit inducible seizures. The behavior of the seizure-sensitive strain is characterized by a progressive increase in responsiveness to weekly application of the S method, from repetitive backward ear movements appearing after postnatal day 40, to a full-blown seizure, while the seizure-resistant strain is apparently unaffected by the stimulation. The difference between these two strains is presumably genetic. To determine the genetic factors underlying this difference, we first examined developmental changes in the stimulus-induced behavior of the F1 hybrids. When the S method was applied, most F1 hybrids had repetitive movements of the ears (and head) similar to the seizure-sensitive gerbils, but generalized seizures emerged considerably later than in seizure-sensitive gerbils. These results suggest that a half dose of the gene products involved renders most gerbils susceptible to the stimulus but is insufficient for the rapid accumulation of an as yet undefined change needed to spread the abnormal electrophysiologic activity to elicit generalized seizures.

    DOI PubMed CiNii J-GLOBAL

  • Cortical correlates of vestibulo-ocular reflex modulation: a PET study

    Y Naito, Tateya, I, S Hirano, M Inoue, K Funabiki, H Toyoda, M Ueno, K Ishizu, Y Nagahama, H Fukuyama, J Ito

    BRAIN   126   1562 - 1578  2003年07月

     概要を見る

    To elucidate cortical correlates of vestibulo-ocular reflex (VOR) modulation, we observed cortical activation during fixation suppression and habituation of caloric vestibular nystagmus in 12 normal subjects, using PET. Significant positive correlation between regional cerebral blood flow (rCBF) and slow phase eye velocity of caloric nystagmus was observed in the middle and posterior insula, inferior parietal lobule, temporal pole, right fusiform gyrus, lingual gyrus, and cerebellar vermis and hemisphere. The rCBF increase in the insular region and the inferior parietal lobule was lateralized depending on the direction of the nystagmus. Caloric nystagmus was suppressed as a result of visual fixation, during which time the area around the right frontal eye field, temporal pole, inferior temporal gyrus, a broad area in the visual cortex, including fusiform and lingual gyrus, cerebellar uvula/nodulus and flocculus, exhibited positive correlation with fixation suppression of caloric nystagmus, while vestibular cortices exhibited negative correlation. The caloric nystagmus habituated with repetition of stimulation. With habituation, we observed activation in the right anterior cingulate gyrus, left superior parietal lobule and right cuneus, and deactivation in the anterior insula, cingulate gyrus, inferior parietal lobule and occipito-temporal visual cortex. The region that showed significant co-activation with fixation suppression and habituation of caloric nystagmus was the right cuneus, and significant co-deactivation was observed in the anterior insula, cingulate gyrus, inferior parietal lobule and middle temporal visual cortex. The present results support previous observations that the parieto-insular cortex and inferior parietal lobule are involved in processing of vestibular information, and, in addition, suggest that activation may depend on the direction of nystagmus. Deactivation of vestibular cortices during visual fixation supports the concept of inhibitory visual-vestibular interaction in the cortex. Significant activation of the cingulate, superior parietal and visual cortices, and cerebellar vermis accompanying reduction of caloric response with repeated stimuli suggests possible involvement of these regions in vestibular habituation. Common activation of the cuneus in visual cortex and deactivation of vestibular and visuo-spatial association cortices by both visual suppression and habituation of VOR suggests that these two mechanisms are not completely independent but may share some cortical and subcortical regions.

    DOI PubMed

  • Gene interaction in DNA microarray data is decomposed by information geometric measure

    H Nakahara, S Nishimura, M Inoue, G Hori, S Amari

    BIOINFORMATICS   19 ( 9 ) 1124 - 1131  2003年06月

     概要を見る

    Motivation: Given the vast amount of gene expression data, it is essential to develop a simple and reliable method of investigating the fine structure of gene interaction. We show how an information geometric measure achieves this.
    Results: We introduce an information geometric measure of binary random vectors and show how this measure reveals the fine structure of gene interaction. In particular, we propose an iterative procedure by using this measure (called IPIG). The procedure finds higher-order dependencies which may underlie the interaction between two genes of interest. To demonstrate the method, we investigate the interaction between the two genes of interest in the data from human acute lymphoblastic leukemia cells. The method successfully discovered biologically known findings and also selected other genes as hidden causes that constitute the interaction.

    DOI J-GLOBAL

  • On-line learning theory of soft committee machines with correlated hidden units - Steepest gradient descent and natural gradient descent

    M Inoue, H Park, M Okada

    JOURNAL OF THE PHYSICAL SOCIETY OF JAPAN   72 ( 4 ) 805 - 810  2003年04月

     概要を見る

    The permutation symmetry of the hidden units in multilayer perceptrons causes the saddle structure and plateaus of the learning dynamics in gradient learning methods. The correlation of the weight vectors of hidden units in a teacher network is thought to affect this saddle structure, resulting in a prolonged learning time, but this mechanism is still unclear. In this paper, we discuss it with regard to soft committee machines and on-line learning using statistical mechanics. Conventional gradient descent needs more time to break the symmetry as the correlation of the teacher weight vectors rises. On the other hand, no plateaus occur with natural gradient descent regardless of the correlation for the limit of a low learning rate. Analytical results support these dynamics around the saddle point.

    DOI CiNii J-GLOBAL

  • Inner ear hearing loss modulates ipsilateral temporal lobe activation by monaural speech stimuli

    Tateya, I, Y Naito, S Hirano, H Kojima, M Inoue, K Kaneko, H Toyoda, M Ueno, K Ishizu, J Ito

    NEUROREPORT   14 ( 5 ) 763 - 767  2003年04月

     概要を見る

    We examined cortical activation by speech in patients with moderate inner ear hearing loss using PET to investigate the response of the language network to insufficient speech input. We made two word lists, well-perceived words and poorly-perceived words, and measured rCBF during monaural presentation of these words. Well-perceived words activated bilateral temporal lobes, bilateral inferior frontal gyri (IFG) and left angular gyrus (AG) regardless of the ear stimulated, Poorly-perceived words activated contralateral temporal lobe and bilateral IFG, while little or no activation was observed in the ipsilateral temporal lobe and left AG. Insufficient activation of the temporal lobe ipsilateral to the ear stimulated might correlated with less accurate word comprehension in patients with inner ear hearing loss.

    DOI J-GLOBAL

  • On-line learning dynamics of multilayer perceptrons with unidentifiable parameters

    Hyeyoung Park, Masato Inoue, Masato Okada

    Journal of Physics A   36 ( 47 ) 11753 - 11764  2003年

    DOI J-GLOBAL

  • Nonlinear digital hearing aid with near instantaneous amplitude compression

    K Kaneko, K Shoji, H Kojima, M Inoue, R Asato, S Hirano, Tateya, I

    EUROPEAN ARCHIVES OF OTO-RHINO-LARYNGOLOGY   258 ( 10 ) 523 - 528  2001年12月

     概要を見る

    We designed a new type of nonlinear digital hearing aid (TD-1) with near-instantaneous amplitude compression for sensorineural hearing-impaired people. The amplification method utilized by the TD-1 normalizes the signal waveforms within a time window bounded by two adjacent positive going zero-crossing points. The signals within this window are amplified so that the maximum amplitude reaches a preset value, thus allowing a consistent peak amplitude to be maintained. To test this, we examined the efficacy of the TD-1 device in 31 patients with sensorineural hearing loss. When tested for recognition of Japanese sentences in a quiet environment, 7 of 25 (28.0%) subjects wearing the TD-1 showed a 20% or greater increase in the maximum intelligibility score than that obtained when wearing their own linear hearing aids. In addition, 7 of 31 (22.6%) subjects showed the same increase with the TD-1 relative to that obtained when they wore HA70 or HA73 devices (Rion Co., Ltd., Tokyo, Japan), which are conventional analogue hearing aids. These results indicate that TD-1 can be effective for patients with sensorineural hearing loss.

    DOI J-GLOBAL

  • Cortical dysfunction of the supplementary motor area in a spasmodic dysphonia patient

    S Hirano, H Kojima, Y Naito, Tateya, I, K Shoji, K Kaneko, M Inoue, S Nishizawa, J Konishi

    AMERICAN JOURNAL OF OTOLARYNGOLOGY   22 ( 3 ) 219 - 222  2001年05月

     概要を見る

    The etiology of spasmodic dysphonia (SD) is still unknown. In the present study, cortical function of a 59-year-old male patient with adductor type SD was examined during phonation with positron emission tomography (PET). Magnetic resonance imaging showed no organic abnormality in the brain. However, PET showed remarkable activities during phonation in the left motor cortex, Broca's area, the cerebellum, and the auditory cortices, whereas the supplementary motor area (SMA) was not activated. The SMA is known to function for motor planning and programming and is usually activated in normal phonation. Several previous reports have shown that the damage of the SMA caused a severe disturbance of voluntary vocalization, In the present case, it was suggested that the functional deficit of the SMA might be related to SD. (Am J Otolaryngol 2001;22:219-222. Copyright (C) 2001 by W.B. Saunders Company).

    DOI J-GLOBAL

  • Functional differentiation of the auditory association area in prelingually deaf subjects

    Shigeru Hirano, Yasushi Naito, Hisayoshi Kojima, Iwao Honjo, Masato Inoue, Kazuhiko Shoji, Ichiro Tateya, Nobuya Fujiki, Sadahiko Nishizawa, Junji Konishi

    Auris Nasus Larynx   27 ( 4 ) 303 - 310  2000年10月

     概要を見る

    Background: It is believed that the number of neurons of the human cortex increases rapidly in the first postnatal year, and then decreases gradually towards adult level as their functions are revised up to 11 years of age ('synaptic revision'). It is also confirmed that regional cerebral blood flow (rCBF) at rest represents the density of the neurons and decreases in accordance with the synaptic revision in process. If synaptic revision does not occur, rCBF remains at high level. Thus, we can evaluate whether functional differentiation has occurred in the human cortex by measuring rCBF at resting state. Objective: To examine functional differentiation of the auditory association area (A2) in prelingually deaf subjects. Methods: Six postlingually and six prelingually deaf subjects who had undergone cochlear implant (CI) were involved in the current study. All prelingually deaf subjects underwent CI over 8 years old. The rCBF in A2 was examined during resting and listening to speech sounds using positron emission tomography (PET) and H2/15O intravenous injection. Twelve normal subjects' rCBFs were also measured as control. Furthermore, three prelingually deaf subjects underwent follow up PET studies in which cortical activities in A2 for listening and lipreading were examined. Results: In the examination of rCBF at rest, rCBFs of prelingually deaf subjects in A2 showed significantly higher than those of either the postlingually deaf subjects or normal subjects. During listening, rCBFs in A2 increased in postlingually deaf subjects and normal subjects, while there was no significant rCBF increase in the prelingually deaf. High rCBF level in A2 at rest observed in prelingually deaf subjects implied a lack of synaptic revision, and it was suggested that the functional differentiation for auditory processing was little in the A2 of prelingually deaf subjects. In the follow up study for three prelingually deaf subjects, activation of A2 was observed during lipreading but not during listening in two cases, who had developed the skill of lipreading while speech recognition was not improved by CI. In contrast, the other case had not used any visual clues in daily communication prior to CI, and the hearing acuity was well improved by CI. This case demonstrated an activity in A2 during listening, while lipreading induced no activation. Conclusion: It is suggested that functional differentiation of A2 should differ according to which of visual and auditory clue is chiefly used during critical periods for speech acquisition. The findings are thought to be important for us to schedule the education and treatment for prelingually deaf children. (C) 2000 Elsevier Science Ireland Ltd.

    DOI PubMed J-GLOBAL

  • Three-dimensional reconstruction of functional brain images

    Inoue M, Shoji K, Kojima H, Hirano S, Naito Y, Honjo I

    Nippon Jibiinkoka Gakkai kaiho   102 ( 8 ) 971 - 975  1999年  [査読有り]

     概要を見る

    様々な言語活動を行っている時の脳の活動部位を特定する有力な方法として, 脳血流をPET (positron emission tomography) で計測しSPM (Statistical Parametric Mapping) で解析する手法がある. SPMは脳全体にわたって有意な活動部位を検出する優れた解析方法である. しかし, 従来これらの表示方法は平面スライス, 3方向投影図, 脳表面着色図がほとんどで, これらでは立体的位置関係が分かりにくく読影が困難であった. そこで今回我々は脳機能画像をより分かり易く読影できるよう, これらを3次元再構築した画像作成を試みた.<BR>対象は正常ボランティア12名で, 日常会話文聴取時のPET画像をSPMで解析した後, 1) SPM付属の従来の表示, 2) 3次元静止画像, 3) 3次元動画の作成を試みた. 2), 3) の3次元表示にはVTK (The Visualization Toolkit) のvolume rendering機能を使用したのが特徴である. また, 脳機能画像には元の脳画像が含まれないため, MRI脳画像との合成などを自作C++プログラムによって行った. 3) の動画は2) の静止画で一コマーコマを作成し, 市販ソフトウエアを用いて動画にした. 2), 3) の画像作成はパソコンを用いて行った.<BR>この結果, 従来の表示方法に比べ, 3次元再構築を行うことで脳活動部位の同定がより容易となった.<BR>脳機能画像は従来神経内科, 脳外科などの領分であったが, 耳鼻咽喉科でも言語の聴取, 表出を扱っている点から今後ますます重要な領域になると考えられる. 今回のように, 従来分かりにくかった脳機能画像を正確な3次元表示画像にすることが, 脳機能画像を正確で直感的に読影するのに重要であり, ひいてはこの分野での発展につながると考えられた. また, 3次元表示で一般的に知られている手法はsurface modelであるが, 脳のようなものの表示には今回用いたvolume renderingの方が優れていると考えられた.

    DOI CiNii

  • Effects of a nonlinear digital hearing aid amplifying in the time domain

    Ken-Ichi Kaneko, Kazuhiko Shoji, Hisayoshi Kojima, Masato Inoue, Ryo Asato, Iwao Honjo

    Practica Otologica   91 ( 2 ) 139 - 144  1998年  [査読有り]

     概要を見る

    The effects of a nonlinear, digital hearing aid on the perception of sentences in a quiet environment were examined for 12 patients with profound sensorineural hearing impairments. This hearing aid was equipped with a compression circuit, which divides the waveform in the time domain by zero-cross points, and then amplifies each waveform to an fixed amplitude. To reduce the susceptibility to background noise, input signals less than a certain level were amplified linearly, and this level was easily adjustable. Three out of 12 (25.0%) patients understood more than 50% of the phrases in typical Japanese sentences at all of the pre-set sound level (50, 60, 70 and 80 dBSPL). Furthermore, this device offered better perception than linear amplification (Rion HA70 or 73) in 8 out of 12 (66.7%) subjects, and was superior to the subjects' own linear hearing aids in 5 out of 10 (50.0%) cases. These results indicate that this hearing aid is beneficial for some subjects.

    DOI

  • AN ANDROGEN RECEPTOR MUTATION CAUSING ANDROGEN RESISTANCE IN UNDERVIRILIZED MALE SYNDROME

    T TSUKADA, M INOUE, S TACHIBANA, Y NAKAI, H TAKEBE

    JOURNAL OF CLINICAL ENDOCRINOLOGY & METABOLISM   79 ( 4 ) 1202 - 1207  1994年10月

     概要を見る

    The molecular basis of androgen resistance was investigated in a patient with undervirilized male syndrome. Binding studies of the androgen receptors in the patient's genital skin fibroblasts revealed a normal binding capacity of 5 alpha-dihydrotestosterone, although the affinity to androgen was slightly lower than the normal control value. The androgen binding of the patient's receptor showed a moderate thermal instability when the assay temperature was raised from 30 to 41 C.
    Nucleotide sequencing analysis of the androgen receptor gene revealed a single nucleotide substitution in exon F, resulting in an amino acid alteration from leucine (CTC) to phenylalanine (TTC) at position 789 within the steroid-binding domain of androgen receptor. When expressed in COS-7 cells, the mutant androgen receptor harboring phenylalanine at position 789 showed thermolabile androgen-binding properties similar to those observed in the patient's genital skin fibroblasts. Cotransfection experiments with an androgen-inducible reporter gene demonstrated a decreased transactivational capability of the mutant receptor. These results indicate that this point mutation modified the receptor function and caused androgen resistance in this patient. This mutation caused the mildest form of all androgen insensitivity syndromes ever examined for mutations in the androgen receptor gene.

    DOI

▼全件表示

書籍等出版物

  • International Workshop on Statistical-mechanical Informatics 2009 (IW-SMI 2009)

    Masato Inoue, Shin Ishii, Yoshiyuki Kabashima, Masato Okada( 担当: 共編者(共編著者))

    Institute of Physics Publishing  2009年

Misc

  • International Workshop on Statistical-mechanical Informatics 2009 (IW-SMI 2009) PREFACE

    Masato Inoue, Shin Ishii, Yoshiyuki Kabashima, Masato Okada

    Journal of Physics: Conference Series   197  2009年

    DOI J-GLOBAL

共同研究・競争的資金等の研究課題

  • -

  • -

特定課題研究

  • 大規模遺伝型・形質型データの解析における低頻度型検出問題の研究

    2013年  

     概要を見る

     本研究では,集団中低頻度である遺伝型・形質型において,特定の疾患や治療との因果関係を確認するランダム化比較試験をどのように構成すれば治験人数等のコストの最小化を図れるかを研究した.結果として,1)既存のblocking等の均等割付法が正しくランダム比較試験を構成していないこと(因果関係を過剰に検出することがある),2)ランダム化比較試験の割付乱数系列を工夫することで,比較試験の最後で行われる統計的検定(カイ二乗検定,Fisherの正確確率検定,G検定等の独立性の検定)の検出力を改善することができることが分かった. ランダム化比較試験は,二つの事項の因果関係を統計的に証明する手法であり,通常,原因と思われる事項をランダムかつ強制的に割り付け,結果と思われる事項との間に統計的に優位な相関関係が見られる場合に,因果関係ありと判断するものである.ランダム化比較試験の利点は,因果関係によらず相関関係を生み出す如何なる事項(例えば,共通の原因など)が存在しようとも,その影響を確率的に排除できる点にあり,科学的な因果関係の証明に広く用いられている.しかしながら,サンプル数が少数の場合,割付が過度に偏る可能性があり,このような場合,検出力の低下が起こり,真実として因果関係が存在するにもかかわらずそれを検出できないことがあり,問題となっている. 既存手法では,blocking等,強制的にほぼ同数を割り付ける簡易な手法が存在するが,このような均等割付法を採用した場合,因果関係によらず相関関係を生み出す事項が存在すると,この影響を排除しきれず,結果のp値が誤って有意に出てしまうことがある.更に,このような過誤を修正する統計的手法は一般に知られておらず,誤ったp値がランダム化比較試験の結果として論文等で報告されているケースが少なからずあるのではないかと思われる. 本研究では,G検定においてこの過誤を過小評価せずに扱い,更に,検出力を最大にするような割付乱数の確率分布を求める手法を考案した.この手法は具体的には,G検定の統計量は二つの事項の相互情報量で与えられ,相互情報量の上限は結果が未知の場合は割付乱数のエントロピーで与えられるため,これを最大化するような最適な割付分布を求めるものである.また,ランダム化比較試験は,サンプルを性別等で分類した上で,各分類項目において検定を行うこともあるため,このような場合にも対応できるよう重みづけを行うオプションも考案した.本手法は,現在行われているランダム化比較試験において用いられており,約一年後に結果を纏める予定である.  これ以外に,関連する統計的推論手法についても研究を進めた.

  • タンパク質間相互作用モチーフの情報論的推定手法の開発

    2012年  

     概要を見る

     タンパク質の多くは体内で特定の化学反応を促進する酵素として働く.また,酵素としての機能を発揮するには,同種または異種のタンパク質と立体的に結合することが必要なものも多い.タンパク質同士が結合する際は,一般に結合モチーフと呼ばれる数~数十アミノ酸配列が重要であることが分かっている.この配列は,それぞれのタンパク質の組合せに固有であるが,同じタンパク質の同じ部位に結合するタンパク質同士には,似たアミノ酸配列から成る結合モチーフがあると思われる.特定のタンパク質と他の多数のタンパク質が結合する/しないというデータから,この結合モチーフを推定することが本研究の最終的な目的である. これに対し,いくつかのアプローチを試みた.まず,近年はshort linear motifと呼ばれる,10アミノ酸程度の短い配列についての研究が進んでおり,例えば真核細胞のlinear motifデータベースには2000ほどのモチーフが登録されている.これらの中から,特定のタンパク質に結合するタンパク質が共通に持つ結合モチーフを推定することを考えた.結合モチーフは一種類とは限らないため,また,各結合モチーフは正確に同じアミノ酸配列でなくとも,少々の変異を許すため,20種類のアミノ酸の離散分布をアミノ酸数だけ繋げ,更にこれをいくつか混合させた確率モデルを構築した.これにより,これまで結合する/しないの二値分類を行っていたアプローチよりも,より正確に結合モチーフを推定できることが期待できる.また,この問題を変分Bayes法用いて近似解を得る際に,1次のTaylor近似を用いたため,解の収束性が悪く,交互に極端な値を行ったり来たりする現象が見られたため,これを防ぐために更新割合を限定するような修正を加えた. また,やや異なる問題設定であるが,多数のタンパク質について,互いに結合する/しないの二値データから成る行列が与えられた際に,これを互いに結合する者同士でクラスタリングする問題についても研究を行った.互いに結合するタンパク質同士は,同じもしくは似たような機能を担っている可能性が高いため,機能未知のタンパク質の機能推定に有用である.従来法として,Markov chain Monte Carlo (MCMC)法を用いて解く方法があったが,これを平均場近似を用いて高速に解くアルゴリズムを構築した.この手法は,初期値のみランダムで設定する必要があるが,収束解の初期値依存性が強く,解を求める試行を多数行わなければならないという欠点が分かった. 上記研究は,英国King’s College LondonのACC Coolen教授との共同研究で進めた.

  • 大規模遺伝型・形質型データの解析における低頻度型検出問題の研究

    2012年  

     概要を見る

     Fisherの正確確率検定は二つの事象間の非独立性を調べる検定法で,一定の仮定のもと,正確なp値を算出できることが利点である.しかし,サンプル数が多い場合や,各事象の起こる場合の数が多い場合は,p値の算出にかかる計算量が指数的に増加するという問題があった.また,genome-wide association study (GWAS)の枠組みでは,ある疾患の原因遺伝子を2万遺伝子,もしくは300万single nucleotide polymorphism (SNP)の中から探し出すという問題を解く必要があるが,これには一疾患対多遺伝子という多重検定を考慮しなければならず,更に難しい問題となっている. 先行研究では,この多重検定補正を行って正確なp値の算出を行うアルゴリズムを構築したが,依然計算量の問題は残っていた.そこで本研究では,実際のp値の使われ方としては,ある閾値以下であるかどうかが重要となる点に着目し,p値が閾値以下であるかどうかのみを正確に判定するアルゴリズムの構築を試みた.集合の包除原理を用いると,一次(単体)の項はp値の上界を,二次の項は下界を,三次の項は再び上界を与えることから,一次の近似で棄却できる場合や,二次の近似で棄却できない場合,三次の近似で棄却できる場合などは,その時点で判定できることが分かった.更に,二次,三次等の高次相関の組合せ数が膨大となることから,より低次の相関情報から,計算する高次相関項を絞り込むことを考えたが,現実的なアルゴリズムの構築にまでは至らなかった. 次に,各事象が三値以上の値を取り得る場合は,順序関係がある場合とない場合があることに着目し,順序関係がある場合にこれを利用して効率の良いアルゴリズムが構築できないか試みた.これはより基本的な問題として,順序回帰と呼ばれる問題があるため,これを研究した.こちらは,従来ある複数の閾値を組み合わせたモデルを排し,ロジスティック回帰モデルを複数組み合わせた確率モデルを構築することができた.また,全サンプルデータを解に反映させるのではなく,一部の重要なデータだけを反映させることで,より汎用的な解を得ることを目的とする疎なBayes学習アルゴリズムを構築できた.また,この研究を手伝った学生(長島 主尚)を2013年1月8日~3月27日の期間,英国King’s College LondonのACC Coolen教授の元へ派遣し,研究を進めた. 次に,SNPデータはhaplotype推定されていた方がより検定の検出力が向上するため,haplotype推定問題にも取り組んだ.こちらも,先の高次相関問題と共通の側面を持つ,モーメント母関数を用いたアルゴリズムの構築を試みた.結果,高次相関まで全て計算できるような規模の小さい問題については,現実的なアルゴリズムの構築ができた.また,推定精度は従来のEMアルゴリズムによる方法に劣るものの,計算量をサンプル数に対して比例でしか増加しない高速アルゴリズムを構築することができた.また,この研究の学会発表を行った学生(小野 司寿加)が IEEE Computational Intelligence Society Japan Chapter より Young Researcher Award を受賞することができた.

  • 多体相互作用モデルのモデル選択に関する情報論的研究

    2006年  

     概要を見る

     多体相互作用モデル,特に二体結合モデルで,その結合が疎な場合の研究を進めた.その結果,以下の三つの系で新しい知見得られ,また手法が開発でき,それぞれ成果発表を行った.また,モデル選択の領域にまでは十分踏み込むことができなかった. 低密度パリティ検査符号の復号問題を解析した.対象にした系は厳密には六体であるが,低次偶数体という点では共通点が多い.このような系では,条件によってスピングラス相と呼ばれる特徴的なマクロ状態をとるが,その状態を主成分解析したところ,結合が密の系と同様の三角形構造が生ずることが分かった.また,磁場下では,この三角形構造の大きさが磁場に応じて小さくなることが分かった.このことについて成果発表を行った. haplotype推定問題での推定手法の開発を行った.この系ではかなり異なったタイプの二体結合モデルを構築し,最良解を近似的に求める一手法を開発した.その後,この二体結合モデルは系のサイズが比較的小さいことが多いため,厳密に解くことが不可能ではない場合があることが分かり,厳密解を求めるアルゴリズムを構築し,成果発表を行った. 脳の一次視覚野からの視覚画像推定を行う推定手法を開発した.この系では事前分布に隣り合うスピン同士のみに二体結合があるというモデルが構築されており,以前から良く研究されている.この系に対し,一次視覚野で行っていると言われるGaborフィルタをかけ,フィルタ出力からの原画像推定を行う具体的な推定手法を開発した.結果,質的には期待通りの推定結果が得られ,成果発表を行った.

  • 遺伝子 haplotype 推定アルゴリズムの開発

    2005年  

     概要を見る

     複数の遺伝母集団が想定される場合に,SNPデータからhaplotype推定を行う手法をEMアルゴリズムに基づいて構成した. 近年,黒人のみに処方可能な薬剤といった人種依存薬が開発されている.一般に投薬対象を限定すれば効果や副作用がより明確に分かるため,投薬対象限定により,従来捨てられてきた薬剤候補が復活する可能性や,今後の薬剤開発において開発が成功する確率が高くなると思われる.しかし,人種は社会的に分かり易い区分であるが,混血の場合など難しい場合もある.薬剤の効能・副作用という観点からは,薬剤のターゲット・代謝等に関わる諸遺伝子のhaplotypeによって,適用を区分する方が理にかなっていると思われる.従来,一塩基多型(SNP)データからターゲット遺伝子のhaplotypeを推定する際に,人種の違いのような複数の遺伝母集団を仮定して推定する手法がなかったため,これを行う一手法をEMアルゴリズムに基づいて構成した.また,計算課程で近似を用いることで必要メモリ量を少なくした.複数遺伝子座での人工的SNPデータに対して本手法を適用したところ,正しくdiplotype及び起源遺伝母集団を推定した.本手法は遺伝子座数が20~30を超えると計算量爆発により計算しきれなくなるため,今後この点を改善する手法が求められる.また,交差の可能性を考慮していないため,この点についても同様である.

海外研究活動

  • ランダム化比較試験における割付乱数系列および独立性検定の研究

    2014年04月
    -
    2014年09月

    英国   King's College London

 

現在担当している科目

▼全件表示