公开(公告)号：EP2660675A4

公开(公告)日：2017-04-05

申请号：EP11852898

申请日：2011-12-15

Applicant: IBM

Inventor： HIDO SHOHEI ,  TAKAHASHI RIKIYA ,  SUZUKI SHOKO ,  YOSHIHAMA SACHIKO ,  HAMZAOUI KARIM

IPC: G05B23/02 ,  G21D3/00

CPC classification number: G05B9/02 ,  G05B23/0281 ,  G21D3/001 ,  Y02E30/40

Abstract: [Problem to be Solved] To efficiently determine the type, location, and cause of an abnormality in a complex system or support the determination. [Solution] A method applied to a computer that determines a situation of a system includes the steps of: receiving measurement data from each of a plurality of measurement targets in the system; computing a plurality of sets of anomaly values based on the measurement data and a predetermined computation algorithm according to a plurality of classifications corresponding to a plurality of properties of each measurement target; and determining the situation of the system based on the sets of anomaly values and a predetermined determination algorithm.

Abstract translation: [要解决的问题]有效地确定复杂系统中的异常的类型，位置和原因，或支持确定。 [解决方案]应用于确定系统的情况的计算机的方法包括以下步骤：从系统中的多个测量目标中的每一个接收测量数据; 基于所述测量数据和根据与每个测量目标的多个属性对应的多个分类的预定计算算法来计算多组异常值; 以及基于所述异常值集合和预定确定算法确定所述系统的状况。

公开(公告)号：JP2013205848A

公开(公告)日：2013-10-07

申请号：JP2012070503

申请日：2012-03-27

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Machines Corporation

Inventor： TAKAHASHI RIKIYA

IPC: G06F17/18

CPC classification number: G06F17/5009 ,  G06Q10/047

Abstract: PROBLEM TO BE SOLVED: To provide a technique for predicting a road link-categorized travel time with as high accuracy as possible even with respect to an edge in which the observed value of costs is insufficient.SOLUTION: Disclosed is a technology of, when the observed value of a certain probability variable is given to the edge of a part of a graph showing a road or the like, estimating the probability distribution of variables in an arbitrary edge in accordance with a probability density function capable of coping with arbitrary distribution. The probability density function is a formula of mixing a basic function estimated from edge-categorized data with an inter-edge similarity scalar and the significance scalar of each edge. That is, the probability density function is a form of interpolating the basic function with similar edges. According to one aspect, it is possible to quickly and stably optimize a parameter in the probability density function from finite data.

Abstract translation: 要解决的问题：为了提供一种用于以相对于观察到的成本值不足的边缘尽可能高的精度来预测道路链路分类行驶时间的技术。解决方案：公开了一种技术，当 将某一概率变量的观测值赋予示出道路等的图的一部分的边缘，根据能够应对任意分布的概率密度函数来估计任意边缘中的变量的概率分布。 概率密度函数是将从边缘分类数据估计的基本函数与边缘相似度标量和每个边缘的有效标量进行混合的公式。 也就是说，概率密度函数是内插具有相似边缘的基本函数的形式。 根据一个方面，可以从有限数据快速稳定地优化概率密度函数中的参数。

公开(公告)号：JP2008040522A

公开(公告)日：2008-02-21

申请号：JP2006209593

申请日：2006-08-01

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Maschines Corporation

Inventor： TAKAHASHI RIKIYA ,  OSOGAMI TAKAYUKI

IPC: G06Q40/00 ,  G06Q40/02 ,  G06Q40/04 ,  G06Q40/06

Abstract: PROBLEM TO BE SOLVED: To obtain probabilistic measures to minimize the risks of cumulative remuneration to be obtained from a plurality of objects undergoing status transition. SOLUTION: The transition probability of the status transition of an agent and a parameter showing the probability distribution of remuneration to be obtained in each status transition as the result of action taken by a user to a plurality of agents in each status are stored. Then, the input of measures to determine action probability for taking an action to each agent in each status is accepted. Then, a recurrence expression for searching the parameter showing the probability distribution of cumulative remuneration to be obtained in current and following terms from the plurality of agents based on the parameter showing the probability distribution of the cumulative remuneration to be obtained in the next and following terms is generated, and the parameter of the probability distribution is calculated by solving such an equation that the parameter of the probability distribution of the cumulative remuneration is converged into the same valve when the initial statuses are the same in the current and following terms and in the next and following terms in the recurrence expression. COPYRIGHT: (C)2008,JPO&INPIT

Abstract translation: 要解决的问题：采取概率措施，尽量减少从经历状态转换的多个对象获得的累积报酬风险。

解决方案：存储代理的状态转换的转移概率和表示在每个状态转换中获得的报酬的概率分布的参数作为用户对每个状态中的多个代理所采取的动作的结果被存储 。 然后，接受输入措施，以确定在每种状态下对每个代理人采取行动的行动概率。 然后，根据显示在下一个和后续条件中获得的累积报酬的概率分布的参数来搜索表示根据当前和后续条件从多个代理获得的累积报酬的概率分布的参数的递归表达式 并且通过求解这样的方程来计算概率分布的参数，即当初始状态在当前和后续条件下相同时，累积报酬的概率分布的参数被收敛在相同的阀中 递归表达式中的下一个和后续条款。 版权所有（C）2008，JPO＆INPIT

公开(公告)号：JP2012093976A

公开(公告)日：2012-05-17

申请号：JP2010241065

申请日：2010-10-27

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Maschines Corporation

Inventor： TAKAHASHI RIKIYA

IPC: G06F17/30

CPC classification number: G06F17/30705

Abstract: PROBLEM TO BE SOLVED: To provide a clustering technique which is efficiently in computational complexity, while guaranteeing global optimality of cluster.SOLUTION: A number of kernel elements are prepared each of which has a fixed center and a fixed bandwidth calculated on the basis of a distribution giving similarity between data of an input data group. A non-negative mixed weight is allocated to each kernel element. Next, a given kernel element and a kernel element having a fixed center and a fixed bandwidth closer thereto are selected and on the basis of a determination of monotonousness in a log-likelihood function of the mixed weight, trimming of an array element corresponding to one kernel element, trimming of an active array element corresponding to the other kernel element, or one-way optimization to one kernel element is executed. When processing to a pair of kernel elements is completed for overall array elements, at that time point, convergence of the mixed weight is determined and if converged, on the basis of the mixed weight, data of the input data group are clustered.

Abstract translation: 要解决的问题：提供一种高效计算复杂度的聚类技术，同时保证集群的全局最优性。 解决方案：准备好多个内核元素，每个内核元素具有基于输入数据组的数据之间的相似性的分布计算的固定中心和固定带宽。 非负混合权重被分配给每个内核元素。 接下来，选择具有固定中心和更接近固定带宽的给定内核元素和内核元素，并且基于对混合权重的对数似然函数中的单调性的确定，修整对应于一个的数组元素 内核元素，对应于另一个内核元素的活动数组元素的修剪或单向优化被执行到一个内核元素。 当对于整个阵列元素完成对一对内核元素的处理时，在该时间点，确定混合权重的收敛，并且如果收敛，则基于混合权重，输入数据组的数据被聚类。 版权所有（C）2012，JPO＆INPIT

公开(公告)号：JP2009103633A

公开(公告)日：2009-05-14

申请号：JP2007277207

申请日：2007-10-25

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Maschines Corporation

Inventor： KAJIMA HISATSUGU ,  SUZUKI SACHIKO ,  HITO SHOHEI ,  TSUBOI YUTA ,  TAKAHASHI TOSHIHIRO ,  IDE TAKESHI ,  TAKAHASHI RIKIYA ,  TAJIMA REI

IPC: G01S5/08

CPC classification number: G01S5/0252

Abstract: PROBLEM TO BE SOLVED: To estimate a location, with high accuracy, based on the strengths of the radio waves received from a plurality of access points.
SOLUTION: A label propagation method is used for estimating position. In particular, this high-accuracy position estimation method robust against the fluctuations of the radio wave intensity is attained, by using a q norm (0

Abstract translation: 要解决的问题：基于从多个接入点接收的无线电波的强度，高精度地估计位置。

解决方案：使用标签传播方法来估计位置。 特别地，通过使用q范数（0

公开(公告)号：JP2013143031A

公开(公告)日：2013-07-22

申请号：JP2012003233

申请日：2012-01-11

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Machines Corporation

Inventor： TAKAHASHI RIKIYA ,  MORIMURA TETSUO ,  OSOGAMI TAKAYUKI ,  KOJAKU SADAMORI

IPC: G06N5/04 ,  G06Q10/04 ,  G06Q50/04

CPC classification number: G06N5/02 ,  G06F17/18 ,  G06N99/005 ,  G06Q50/06

Abstract: PROBLEM TO BE SOLVED: To provide a method and system for accurately predicting data for explained variables.SOLUTION: This method includes: receiving explanatory variables as input data; searching training data for elements whose discrete variables match those of each element of sets contained in the input data; applying a function weighted by a scaling variable to each element in the input data and to each of one or more elements found for the element in the input data in order to calculate each function value; calculating the sum of the function values for every element in the input data; and applying every calculated sum to a prediction equation in order to calculate a prediction value of the explained variable.

Abstract translation: 要解决的问题：提供一种准确预测所说变量数据的方法和系统。解决方案：该方法包括：接收解释变量作为输入数据; 搜索其离散变量与输入数据中包含的集合的每个元素的元素匹配的元素的训练数据; 将由缩放变量加权的函数应用于输入数据中的每个元素以及为输入数据中的元素找到的一个或多个元素中的每个元素，以便计算每个函数值; 计算输入数据中每个元素的函数值之和; 并且将每个计算的和应用于预测方程，以便计算所述变量的预测值。

公开(公告)号：JP2008152321A

公开(公告)日：2008-07-03

申请号：JP2006336621

申请日：2006-12-14

Applicant: Internatl Business Mach Corp ,  インターナショナル・ビジネス・マシーンズ・コーポレーションInternational Business Maschines Corporation

Inventor： TAKAHASHI RIKIYA ,  OSOGAMI TAKAYUKI

IPC: G06Q10/00 ,  G06Q30/02 ,  G06Q50/00 ,  G06Q50/10 ,  G06Q90/00

CPC classification number: G06Q30/02 ,  G06Q30/0202

Abstract: PROBLEM TO BE SOLVED: To solve the problems of conventional techniques being unable to define customer states in consideration of marketing actions and to obtain, as parameters of customer state, information on what kinds of effects marketing actions produce in the short and long terms. SOLUTION: In order to obtain customer state transition probabilities and short-term rewards conditioned by actions, customer behaviors are modeled with a hidden Markov model (HMM) using composite states each composed of a pair of a customer sate and a marketing action. Parameters of the estimated hidden Markov model (the composite state transition probabilities and a reward distribution for each composite state) are further transformed into the customer state transition probabilities and the distribution of rewards for each customer state conditioned by marketing actions. In order to model purchase properties in more detail, an inter-purchase time is always included as an element in the customer state vector, thereby allowing the customer state to have information on the probability distribution of the inter-purchase time. COPYRIGHT: (C)2008,JPO&INPIT

Abstract translation: 要解决的问题：为了解决传统技术的问题，考虑到营销行为不能定义客户状态，并获得作为客户状态的参数的信息，关于营销行为在短期和长期产生什么样的效果的信息 条款。

解决方案：为了获得客户状态转移概率和以行动为条件的短期回报，客户行为使用隐藏的马尔科夫模型（HMM）建模，使用由客户同意和营销行为组成的复合状态 。 估计的隐马尔可夫模型的参数（复合状态转移概率和每个复合状态的回报分布）进一步转化为客户状态转移概率和由营销行为调节的每个客户状态的回报分配。 为了更详细地建模购买属性，总是将客户时间作为元素包含在客户状态向量中，从而允许客户状态具有关于购买间时间的概率分布的信息。 版权所有（C）2008，JPO＆INPIT