ERLERNEN EINES ERKENNUNGSMODELLS UNTER VERWENDEN EINER VERLUSTFUNKTION

    公开(公告)号:DE112020000279T5

    公开(公告)日:2021-10-14

    申请号:DE112020000279

    申请日:2020-03-13

    Applicant: IBM

    Abstract: Es besteht ein Wunsch nach einem genauen Erlernen eines Erkennungsmodells (130). Es wird ein durch einen Computer implementiertes Verfahren bereitgestellt, das umfasst: Erfassen eines Eingangsbildes; Erfassen eines kommentierten Bildes, das einen Bereich von Interesse im Eingangsbild kennzeichnet; Eingeben des Eingangsbildes in ein Erkennungsmodell (130), das ein Ausgangsbild erzeugt, das einen Zielbereich aus dem Eingangsbild zeigt; Berechnen eines Fehlers zwischen dem Ausgangsbild und dem kommentierten Bild unter Verwenden einer Verlustfunktion, die einen Fehler innerhalb des Bereichs von Interesse stärker gewichtet als einen Fehler außerhalb des Bereichs von Interesse; und Aktualisieren des Erkennungsmodells derart, dass der Fehler verringert wird.

    Verfahren und Programmprodukt zum Entwerfen einer Quelle und einer Maske für die Lithographie

    公开(公告)号:DE112014000486B4

    公开(公告)日:2021-08-19

    申请号:DE112014000486

    申请日:2014-05-09

    Applicant: IBM

    Abstract: Computerimplementiertes Verfahren für das Entwerfen einer Quelle und einer Maske für die Lithographie, das aufweist:einen Schritt (S1) des Auswählens eines Satzes von Strukturen (10);einen Schritt (S2) des Durchführens einer Quellen-Masken-Optimierung (SMO), die den Satz von Strukturen (10) verwendet, unter einer Beschränkungsregel der optischen Lagekorrektur (OPC), die für ein selektives Beschränken eines Verschiebens (40) einer Kantenposition eines Polygons (20) verwendet wird, wenn die OPC auf den Satz von Strukturen (10) angewendet wird; undeinen Schritt (S4) des Festlegens eines Layouts der Maske für die Lithographie mittels Anwendens (S3) der OPC auf sämtliche Strukturen (10), welche die Maske für die Lithographie bilden, wobei die durch die SMO optimierte Quelle verwendet wird.

    KLASSIFIZIEREN NEURONALER NETZE
    4.
    发明专利

    公开(公告)号:DE112018002822T5

    公开(公告)日:2020-02-13

    申请号:DE112018002822

    申请日:2018-05-31

    Applicant: IBM

    Abstract: Ein Klassifizieren neuronaler Netze kann ausgeführt werden durch Eingeben eines Trainingsdatensatzes in jedes aus einer Mehrzahl von ersten neuronalen Netzen, wobei der Trainingsdatensatz eine Mehrzahl von Proben enthält, Erhalten einer Mehrzahl von Sätzen von Ausgabewerten aus der Mehrzahl von ersten neuronalen Netzen, wobei jeder Satz von Ausgabewerten eine Mehrzahl von Ausgabewerten enthält, die einer aus der Mehrzahl von Proben entsprechen, wobei jeder Ausgabewert in Reaktion auf das Eingeben von einer der Proben des Trainingsdatensatzes von einem entsprechenden ersten neuronalen Netz ausgegeben wird, Eingeben der Mehrzahl von Sätzen von Ausgabewerten in ein zweites neuronales Netz und Trainieren des zweiten neuronalen Netzes, so dass es in Reaktion auf das Eingeben eines entsprechenden Satzes von Ausgabewerten ein erwartetes Ergebnis ausgibt, das jeder Probe entspricht.

    System and method for managing community provided in information processing system
    5.
    发明专利
    System and method for managing community provided in information processing system 有权
    用于管理信息处理系统中提供的社区的系统和方法

    公开(公告)号:JP2007140578A

    公开(公告)日:2007-06-07

    申请号:JP2005329070

    申请日:2005-11-14

    CPC classification number: G06Q10/107

    Abstract: PROBLEM TO BE SOLVED: To properly manage the content of information registered in a user community. SOLUTION: To provide a system for managing a user community provided on an information processing system for allowing the other users refer to information provided by a user. This system is provided with a storage part for storing an audit policy for specifying the content of information for permitting registration in the user community in association with each of a plurality of user communities; a detection part for detecting the user community of a provision destination according to the provision of information by the user; and a registration control part for preventing the information from being registered in the user community under such conditions that the information provided by the user is against the audit policy corresponding to the detected user community. COPYRIGHT: (C)2007,JPO&INPIT

    Abstract translation: 要解决的问题:正确管理在用户社区中注册的信息的内容。 解决方案:提供一种用于管理在信息处理系统上提供的用户社区的系统,以允许其他用户参考由用户提供的信息。 该系统设置有存储部件,用于存储用于指定与多个用户社区中的每一个相关联地在用户社区中注册的信息的内容的审计策略; 检测部,其根据用户的信息提供来检测提供目的地的用户群; 以及注册控制部分,用于在用户所提供的信息与所检测的用户社区相对应的审计策略的条件下,防止在用户社区中注册信息。 版权所有(C)2007,JPO&INPIT

    SPEECH SYNTHESIZING METHOD AND SYSTEM THEREFOR

    公开(公告)号:JPH0895589A

    公开(公告)日:1996-04-12

    申请号:JP22666794

    申请日:1994-09-21

    Applicant: IBM JAPAN

    Abstract: PURPOSE: To provide a stable speech synthesis processing with the reduced tremble of pitch in a speech synthesizer system utilizing a pitch synchronized waveform superposition method. CONSTITUTION: A glottis closing point is made to be the reference point of superposition (pitch mark). Since the glottis closing point is stably and accurately extracted by using a dynamic wavelet transformation, a speech with few trembling and few gurgling is synthesized by its stability. By setting the reference point of superposition on another position from the reference point of waveform cutout, the softer cutout of a wave form is enabled. Extraction of the glottis closing point is performed by searching the local peak of dynamic wavelet transformation, however, preferably the threshold for searching the local peak of dynamic wavelet transformation is adaptively controlled every time when dynamic wavelet transformation is obtained.

    Mask design method, program, and mask design system
    7.
    发明专利
    Mask design method, program, and mask design system 有权
    MASK设计方法,程序和MASK设计系统

    公开(公告)号:JP2013213973A

    公开(公告)日:2013-10-17

    申请号:JP2012084657

    申请日:2012-04-03

    CPC classification number: G03F1/70

    Abstract: PROBLEM TO BE SOLVED: To provide a mask design method, a program thereof, and a mask design system.SOLUTION: The mask design system according to the present invention includes: an optimization unit 120 having a global mask optimization section 122 and a global light source optimization section 123; and an optical domain simultaneous optimization unit (FDJO) 124. The optimization unit 120 performs a non-linear optimization for an optical domain representation of a mask pattern under a constraint condition that a value of a negative deviation of the object domain representation at a prescribed evaluation point of a restored object domain representation is smaller than a value of a predetermined negative threshold of the evaluation point.

    Abstract translation: 要解决的问题:提供掩模设计方法,程序和掩模设计系统。根据本发明的掩模设计系统包括:具有全局掩模优化部分122和全局光的优化单元120 源优化部分123; 和光域同步优化单元(FDJO)124.优化单元120在限制条件下对掩模图案的光域表示执行非线性优化,所述约束条件是目标域表示在规定的负值偏差的值 恢复的对象域表示的评估点小于评估点的预定负阈值的值。

    Managing technology of encryption key in encryption communication
    8.
    发明专利
    Managing technology of encryption key in encryption communication 有权
    加密技术在加密通信中的关键

    公开(公告)号:JP2008078952A

    公开(公告)日:2008-04-03

    申请号:JP2006255183

    申请日:2006-09-21

    CPC classification number: H04L9/0836 H04L2209/601

    Abstract: PROBLEM TO BE SOLVED: To transfer authority of encryption communication without increasing leakage risk of an encryption key. SOLUTION: A system with a first communication apparatus and a management apparatus which makes encryption communications perform between the first communication apparatus and an external second communication apparatus, is offered. The management apparatus is provided with: a key memory unit which memorizes a first encryption key which is shared with the second communication apparatus; a key creation unit which performs operation which gives the first encryption key which is read out from the key memory unit to one direction function with a key, and creates the operation result as the second encryption key; and a key output unit which outputs the second encryption key and the key of the one direction function to the first communication apparatus. The first communication apparatus is provided with a key creation control unit which transmits the key of the one direction function received from the management apparatus to the second communication apparatus, makes the second communication apparatus perform operation to give the memorized first encryption key to the one direction function with the key, and creates the second encryption key; and an encryption communication unit which performs encryption communications with the second communication apparatus using the second encryption key received from the management apparatus. COPYRIGHT: (C)2008,JPO&INPIT

    Abstract translation: 要解决的问题:传递加密通信的权限,而不增加加密密钥的泄漏风险。 提供一种具有在第一通信装置和外部第二通信装置之间进行加密通信的第一通信装置和管理装置的系统。 管理装置设置有:密钥存储单元,其存储与第二通信装置共享的第一加密密钥; 密钥创建单元,执行将从密钥存储器单元读出的第一加密密钥与密钥一个方向函数进行的操作,并创建作为第二加密密钥的操作结果; 以及键输出单元,其向第一通信设备输出第二加密密钥和单向功能的密钥。 第一通信装置设置有密钥创建控制单元,其将从管理装置接收的一个方向功能的密钥发送到第二通信装置,使第二通信装置执行操作以将存储的第一加密密钥提供给一个方向 使用键功能,并创建第二个加密密钥; 以及加密通信单元,其使用从所述管理装置接收的所述第二加密密钥来执行与所述第二通信装置的加密通信。 版权所有(C)2008,JPO&INPIT

    Method for computing manufacturability of lithographic mask using continuous differentiability of manufacturability defined on continuous scale
    9.
    发明专利
    Method for computing manufacturability of lithographic mask using continuous differentiability of manufacturability defined on continuous scale 有权
    使用连续定标定义的可制造性的连续差异化计算拼接掩模的制造方法

    公开(公告)号:JP2010140021A

    公开(公告)日:2010-06-24

    申请号:JP2009256225

    申请日:2009-11-09

    CPC classification number: G03F1/76 G03F1/36

    Abstract: PROBLEM TO BE SOLVED: To provide a method for computing manufacturability of a lithographic mask to be used for manufacturing a semiconductor device. SOLUTION: Target edge pairs are selected from the mask layout data of a lithographic mask for computinging a manufacturing penalty as an index indicating the difficulty of manufacturing in making a lithographic mask. The mask layout data includes polygons, and each polygon has a large number of edges. Each target edge pair is defined by two edges of edges of one or more of the polygons. The manufacturability of the lithographic mask, including the manufacturing penalty in making the lithographic mask, is determined. Determining the manufacturing penalty is based on the target edge pairs selected. Determining the manufacturability of the lithographic mask uses continuous differentiability of the maufacturability defined on a continuous scale. The manufacturability of the lithographic mask is output. The manufacturability of the lithographic mask is dependent on the manufacturing penalty in making the lithographic mask. COPYRIGHT: (C)2010,JPO&INPIT

    Abstract translation: 要解决的问题:提供一种用于计算用于制造半导体器件的光刻掩模的可制造性的方法。

    解决方案:从用于计算制造罚款的光刻掩模的掩模布局数据中选择目标边缘对作为指示制造光刻掩模的制造难度的指标。 掩模布局数据包括多边形,并且每个多边形具有大量边缘。 每个目标边对由一个或多个多边形的边缘的两个边缘限定。 确定了光刻掩模的可制造性,包括制造光刻掩模的制造损失。 确定制造损失是基于所选择的目标边对。 确定平版印刷掩模的可制造性使用在连续刻度上定义的可制造性的连续可微性。 输出光刻掩模的可制造性。 光刻掩模的可制造性取决于制造光刻掩模的制造损失。 版权所有(C)2010,JPO&INPIT

    Device, method and program for extracting invisible information
    10.
    发明专利
    Device, method and program for extracting invisible information 有权
    用于提取不可见信息的设备,方法和程序

    公开(公告)号:JP2007164715A

    公开(公告)日:2007-06-28

    申请号:JP2005363767

    申请日:2005-12-16

    CPC classification number: G06K19/06037 G06K7/14 G06K2019/06225

    Abstract: PROBLEM TO BE SOLVED: To extract invisible information even when visible information is overlapped with the invisible information. SOLUTION: The present invention provides a method for extracting the invisible information from an object with the visible information overlapped with at least a part of the invisible information formed by stealth ink and the method includes steps: for irradiating the object with light which makes the stealth ink emit light: for receiving reflected light from the object; for extracting image information from the received reflected light; for dividing the image information into a plurality of pieces of color channel information; for calculating a correlation function among the pieces of color channel information about the respective pixel values constituting the image information about at least two pieces of color channel information selected from the pieces of color channel information; for reducing and interpolating visible information components about one of the selected color channel information using the correlation function and for extracting the invisible information from one piece of interpolated color channel information. COPYRIGHT: (C)2007,JPO&INPIT

    Abstract translation: 要解决的问题:即使可见信息与不可见信息重叠,也可以提取隐形信息。 解决方案:本发明提供一种用于从具有与隐形墨水形成的隐形信息的至少一部分重叠的可见信息的对象提取不可见信息的方法,该方法包括以下步骤:用光照射物体 使隐形墨水发光:用于接收来自物体的反射光; 用于从所接收的反射光中提取图像信息; 用于将图像信息划分成多条彩色通道信息; 用于计算关于构成关于从所述彩色通道信息中选择的至少两条彩色通道信息的图像信息的各个像素值的色彩​​通道信息之间的相关函数; 用于使用相关函数来减少和内插关于所选颜色通道信息之一的可见信息分量,并从一条内插色彩信道信息中提取不可见信息。 版权所有(C)2007,JPO&INPIT

Patent Agency Ranking