Search

При обработке больших объемов информации часто возникает проблема уменьшения размерности данных с максимальным сохранением их информативности. В работе приведены эффективные алгоритмы для сокращения размерности данных, описаны их принципы работы: метод главных компонент, ядерный метод главных компонент, метод многомерного шкалирования, метод локально-линейного встраивания, метод изометрического отображения, метод упругих карт. Разработано программное обеспечение, реализующее рассмотренные алгоритмы. Разработанное программное обеспечение было применено для анализа медицинских данных., At processing large amounts of information, there is often a problem of reducing the dimensionality of data and keep its informative value. The paper presents the effective algorithms for reducing the dimensionality of data, describes their operating principles: the principal component method, the kernel method of principal components, the multidimensional scaling method, the local-linear embedding method, the isometric mapping method, and the flexible maps method. The software implementing the above mentioned algorithms is developed, this application was used to analyze medical data.

歌声特有の音響特徴量と歌声知覚の関係を検討するために,歌声らしさの知覚モデルを提案する。このモデルは,「歌声らしさという聴覚印象が複数の基本的な心理的特徴の知覚に起因する」という仮説のもと,歌声らしさと音響特徴量の対応関係の間に基本的な心理的特徴を介した3層で構成される階層構造モデルである。第1層(歌声らしさ)と第2層(基本的な心理的特徴)の関係については,多次元尺度構成法と重回帰分析によって調査した。第2層と第3層(音響特徴量)の関係については, STRAIGHTを用いた音響分析・合成と心理物理実験によって調査した。その結果, "揺れ," "響き"といった基本的な心理的特徴が歌声らしさの聴覚印象に大きく寄与しており,両者の聴覚印象には基本周波数の準周期的な振動成分であるヴィブラートとそれに同期したホルマントの振幅変調成分,及び3kHz付近の顕著なスペクトルピーク成分と同帯域の強い高調波成分がそれぞれ寄与していることが明らかとなった。更に,これらの音響特徴量を話声に付与することで歌声らしさの聴覚印象が向上する結果を得た。以上から,歌声らしさの知覚モデルを構築することで,歌声知覚における歌声特有の音響特徴量の役割について詳細に検討することが可能であることを示した。

The main objects of this paper are to introduce how to organize the dance therapy session intended for inpatients of a collaborative psychiatric hospital in Ibaragi prefecture; and to develop an outcome measurement of the session on their interactive physical communication through expressive bady movements. In order to develop the quantitative model assessing how patients' interactive physical communication is promoted through the sessions, we collect 20 principal indicators based on our previous studies on "interactive physical synchronization (kyoshin)" on expressive body movements. We apply a hierarchical cluster method (called multidimensional scaling method) evaluating the degree of interactive physical communication for 7 male and 17 female patients through the dance thrapy performed four-times from January through March in 2004. Cosequently, we found that our therapy program would motivate patients to achieve their creative and unique expressive body movements and it would lead them to higher degree of interactive physical communication. Also, we foud that the quantitative model is applicable to locate relative relation between "activity" and "passivity" among patients.

The main object of this paper is to investigate how to establish the education program for preschool teachers, in order for developing hearty and sympathetic body awareness when interacting with children. Due to the lack of decent discussion on how to measure the quality of care provided by preschool teachers that is associated with their sensitivity and creativity, to our best knowledge, there has been little empirical research on how efficient a specific education program is. Numbers of studies have proved that proceeding "interactive psycho-physical synchronization (kyoshin)" plays a significant role for preschool teachers to develop their "adaptable and compassionate body attentiveness." In order to develop the empirical model assessing how the "interactive physical synchronization" on expressive body movement will be expanding, we collect 15 principal indicators measuring the degree of "interactive psycho-physical synchronization" for approximately 250 males and females at age of 20 and older years old and we apply hierarchical cluster method for our data. Finally, we examine the effectiveness of the established model for 12 students who are taking our course at Toyo Eiwa University. Consequently, we found that our training program would motivate students to achieve higher degree of interactive physical synchronization compared to the status before the class, with respect to 4 out of 15 indicators representing intersubjectivity" and "physical sensitivity," in average. Further, using multidimensional scaling method, we observed that 8 out of 12 students have attained higher degree of interactive psycho-physical synchronization. Based on our empirical results in this study, therefore, we conclude that a specific professional training program might be very effective for preschool teachers to develop hearty and sympathetic body awareness when interacting with children.

Magistro baigiamajame darbe Monte Karlo metodu buvo siekiama atlikti įvairių daugiamačių neparametrinių tankių įvertinių tikslumo lyginamąją analizę naudojantis histogramos, menkai nagrinėtu gradientiniu ir k artimiausio kaimyno metodais, tačiau pagrindinis dėmesys buvo skiriamas nežinomo skirstinio tankio formos nustatymui., This Master‘s work examines the use of flexible methods to approximate an unknown density function, and techniques appropriate for visualization of densities in up to d - dimensions. In the following sections, these algorithms and theory of non parametric density estimation will be described: histogram method, k - nearest neighbour and poorly studied kernel gradient method. For the visualization of multivariate data were implement multidimensional scaling method.

Rad predstavlja rezultate istraživanja klasifi kacije poslovnih problema trgovačkih društava dužnika banke. Istraživanje je provedeno na polaznom uzorku veličine 168 jedinica – dužnika banke, a korišteno je 15 omjera do-bijenih iz osnovnih fi nancijskih izvješća. U radu su primijenjeni višestruka diskriminacijska analiza, logit model i metoda višedimenzionalnih skala za razvrstavanje trgovačkih društava prema urednosti izmirenja obveza prema banci. Budući da diskriminacijskom analizom i logit modelom nije bilo moguće uspješno razlikovati dobra i srednja društva, polazni je uzorak sužen na 141 jedinicu, pa su ga činila trgovačka društva grupirana na dobra i loša. Svi kreirani modeli pokazali su visok stupanj pouzdanosti predviđanja poslovnih problema društava u poslovnim odnosima s bankom, što je potvrđeno i na kontrolnom uzorku. Istraživanje je također pokazalo da se i metodom višedimenzionalnih skala može uspješno koristiti pri grupiranju društava prema razini poslovnih problema uz korištenje odgovarajućih financijskih omjera., The paper presents research of models for classification of bank debtor distress based on ratios calculated from official financial statements. Basic sample consists of 168 company’s debtors of middle size bank, and 15 financial ratios were used. In research multiple discriminate analyses, logit model and multidimensional scaling method was used for classifying companies based on orderliness in payment of their liabilities. Since discriminate analysis and logit model could not successfully discriminate “good” and “medium” companies, basic sample was appropriately limited to 141 companies classified as “good” or “dubious”. All estimated models have high rate of reliability in prediction of orderliness in payment of liabilities that can express the degree of company’s efficiency or distress. Reliability of all created models was confirmed on control sample as well. This research also demonstrated that multidimensional scaling methods can be successfully used for classifying of companies based on magnitude of their business problems using appropriate financial ratios.