WO1998010352A1 - Verfahren und vorrichtung zur rechnergestützten generierung mindestens eines künstlichen trainingsdatenvektors für ein neuronales netz - Google Patents
Verfahren und vorrichtung zur rechnergestützten generierung mindestens eines künstlichen trainingsdatenvektors für ein neuronales netz Download PDFInfo
- Publication number
- WO1998010352A1 WO1998010352A1 PCT/DE1997/001497 DE9701497W WO9810352A1 WO 1998010352 A1 WO1998010352 A1 WO 1998010352A1 DE 9701497 W DE9701497 W DE 9701497W WO 9810352 A1 WO9810352 A1 WO 9810352A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- training data
- neural network
- distribution
- rfj
- error
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Ceased
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/214—Generating training patterns; Bootstrap methods, e.g. bagging or boosting
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/24—Classification techniques
- G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
- G06F18/2413—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches based on distances to training or reference patterns
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/09—Supervised learning
-
- G—PHYSICS
- G06—COMPUTING OR CALCULATING; COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/82—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using neural networks
Definitions
- the training data is very noisy in many areas of application, for example when modeling financial data such as stock or currency rates.
- the training data thus receive random disturbances that have nothing to do with the system dynamics that are actually to be modeled.
- training data vectors are generated anew, which none at all Contain a statement about the system to be modeled.
- the training data vectors also contain no information about the actual noise on which the system is based.
- the amount of training data is increased, but this does not have to support the learning process, since a predetermined noise, which has nothing to do with the actual system dynamics, is used to train the neural network. This can still lead to overtraining.
- the invention is therefore based on the problem of artificially forming new training data vectors for a neural network, but avoiding overtraining the neural network.
- a residual error is determined after training the neural network with available training data vectors from a training data set.
- An input-related backward error is determined from the residual error, for example using a gradient descent method.
- the determination of the backward error corresponds to the usual procedure during the training of a neural network for the adaptation of the individual weights of the neural network. If the input-related backward error has been determined, a statistical distribution assigned to the respective input is generated taking into account the respective backward error, and the artificial training data vector is generated taking into account the respective statistical distribution at the inputs of the neural network.
- the artificially generated training data vectors are dependent on the backward error that still exists after the training of the neural network and thus on depend on the performance of the neural network.
- the training data vectors generated contain information about the system to be modeled. In this way, overtraining of the neural network by the additional training data vectors can be avoided.
- the device according to claim 9 has a computing unit which is set up in such a way that the method steps described above are carried out.
- Fig. 1 is a flow chart in which the individual process steps of the method are shown;
- Fig. 2 is a sketch showing a computer arrangement with which the method can be carried out.
- a residual error RE is determined by a computer R.
- the residual error RE is, for example, in the case of m training data vectors TDV with the target values tj_ assigned to the training data vectors TDV, the training data vector TDVi being clearly identified in each case by an index i, the following
- any other rule can of course be used to form the residual error RE after the training of the neural network NN.
- a backward error RFj is determined (102).
- the backward error RF can be determined, for example, using a gradient descent method based on the input signals of the neural networks NN.
- Training data vector TDVi (x ⁇ , ..., Xj, ..., x n )
- Noise term vector (e ⁇ _, ..., e--., ..., e n )
- Input data vector z. (x ⁇ + e 1 ( .. -, Xj + e, ..., x n + e n )
- N denotes a number of components of the training data vectors TDVi, the noise term vectors e and the input data vectors z.
- the backward error RFj is determined individually for each input j of the neural network NN, an index j being used to uniquely identify a component of the input data vector z and an input of the neural network NN, respectively.
- the backward error RF of the input data vector z thus results from the respective partial derivatives of the residual error RE according to the individual input signals ZJ.
- the symbol t uniquely identifies a point in time at which the backward error RFj is determined.
- a statistical distribution Sj is now formed in a third step (103) for each input j of the neural network NN.
- Any statistical distribution for describing a random process can generally be used as the statistical distribution Sj.
- the noise term ej t which is used to form at least one artificial training data vector KTDV, lies in the range of the interval:
- the noise width s results. according to the following rule:
- the at least one artificial training data vector KTDV is generated (104). This clearly means that the artificial training data vector KTDV is generated by the random process described in each case by the statistical distribution Sj.
- the process can be clearly described as follows.
- the input data represented by the originally available training data vectors TDVi, are noisy with a random process that is dependent on the respective gradient error function for each input after the input signals.
- Noise on inputs with a large noise width s "stiffens" the neural network NN, thereby preventing overtraining, which leads to a better generalization ability of the neural network NN.
- This method can be used particularly advantageously in application situations in which only a relatively small number of training data vectors TDVi are available for training the neural network NN.
- the training data set TDM can be artificially extended to an expanded training data set KTDM without distorting the actual system dynamics, since the statistical properties of the training data set TDM are also contained in the artificial training data vectors KTDV.
- a typical application situation of the method lies in the analysis of financial markets, for example of stock markets or also securities markets.
- the following adaptation rule can be used for a Gaussian statistical distribution S j :
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Artificial Intelligence (AREA)
- General Physics & Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Mathematical Physics (AREA)
- Molecular Biology (AREA)
- Computational Linguistics (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Multimedia (AREA)
- Evolutionary Biology (AREA)
- Bioinformatics & Computational Biology (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Medical Informatics (AREA)
- Databases & Information Systems (AREA)
- Complex Calculations (AREA)
- Image Analysis (AREA)
- Financial Or Insurance-Related Operations Such As Payment And Settlement (AREA)
Abstract
Description
Claims
Priority Applications (4)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| JP10512095A JP2000517447A (ja) | 1996-09-03 | 1997-07-15 | ニューラルネットワーク用の少なくとも1つの人工的トレーニングデータベクトルをコンピュータ支援により生成するための方法及び装置 |
| EP97935429A EP0925541B1 (de) | 1996-09-03 | 1997-07-15 | Verfahren und vorrichtung zur rechnergestützten generierung mindestens eines künstlichen trainingsdatenvektors für ein neuronales netz |
| DE59703304T DE59703304D1 (de) | 1996-09-03 | 1997-07-15 | Verfahren und vorrichtung zur rechnergestützten generierung mindestens eines künstlichen trainingsdatenvektors für ein neuronales netz |
| US09/254,298 US6282529B1 (en) | 1996-09-03 | 1997-07-15 | Method and apparatus for computer-supported generation of at least one artificial training data vector for a neural network |
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| DE19635758A DE19635758C1 (de) | 1996-09-03 | 1996-09-03 | Verfahren und Vorrichtung zur rechnergestützten Generierung mindestens eines künstlichen Trainingsdatenvektors für ein neuronales Netz |
| DE19635758.6 | 1996-09-03 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO1998010352A1 true WO1998010352A1 (de) | 1998-03-12 |
Family
ID=7804522
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/DE1997/001497 Ceased WO1998010352A1 (de) | 1996-09-03 | 1997-07-15 | Verfahren und vorrichtung zur rechnergestützten generierung mindestens eines künstlichen trainingsdatenvektors für ein neuronales netz |
Country Status (5)
| Country | Link |
|---|---|
| US (1) | US6282529B1 (de) |
| EP (1) | EP0925541B1 (de) |
| JP (1) | JP2000517447A (de) |
| DE (2) | DE19635758C1 (de) |
| WO (1) | WO1998010352A1 (de) |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1388812A1 (de) * | 2002-07-04 | 2004-02-11 | Ronald E. Dr. Kates | Trainingsverfahren für lernfähiges System |
| WO2004090807A3 (de) * | 2003-04-10 | 2005-12-22 | Bayer Technology Services Gmbh | Verfahren zum trainieren von neuronalen netzen |
Families Citing this family (12)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| JP2004354251A (ja) * | 2003-05-29 | 2004-12-16 | Nidek Co Ltd | 欠陥検査装置 |
| JP4635220B2 (ja) * | 2005-06-07 | 2011-02-23 | 学校法人慶應義塾 | 複数の化学物質の測定方法 |
| JP5307996B2 (ja) * | 2006-09-06 | 2013-10-02 | 株式会社Dnaチップ研究所 | 判別因子セットを特定する方法、システム及びコンピュータソフトウェアプログラム |
| US20180089581A1 (en) * | 2016-09-27 | 2018-03-29 | Futurewei Technologies, Inc. | Apparatus and method for dataset model fitting using a classifying engine |
| US10944440B2 (en) * | 2018-04-11 | 2021-03-09 | Booz Allen Hamilton Inc. | System and method of processing a radio frequency signal with a neural network |
| DE102018222347A1 (de) * | 2018-12-19 | 2020-06-25 | Robert Bosch Gmbh | Verfahren zum Trainieren eines neuronalen Netzes |
| WO2020237011A1 (en) * | 2019-05-23 | 2020-11-26 | Cognizant Technology Solutions U.S. Corporation | Quantifying the predictive uncertainty of neural networks via residual estimation with i/o kernel |
| KR20210130048A (ko) | 2020-04-21 | 2021-10-29 | 삼성에스디에스 주식회사 | 분석 모델 학습 방법 및 장치 |
| US11715037B2 (en) * | 2020-09-11 | 2023-08-01 | International Business Machines Corporation | Validation of AI models using holdout sets |
| DE102021202289A1 (de) | 2021-03-09 | 2022-09-15 | Robert Bosch Gesellschaft mit beschränkter Haftung | Computerimplementiertes Verfahren und Vorrichtung zum Trainieren eines datenbasierten Zeitpunktbestimmungsmodells für die Bestimmung eines Öffnungs- oder Schließzeitpunkts eines Einspritzventils mithilfe maschineller Lernverfahren |
| DE102021202287B4 (de) | 2021-03-09 | 2023-03-23 | Robert Bosch Gesellschaft mit beschränkter Haftung | Verfahren und Vorrichtung zum Betreiben eines Kraftstoffeinspritzventils mithilfe maschineller Lernverfahren |
| US11822887B2 (en) * | 2021-03-12 | 2023-11-21 | Adobe, Inc. | Robust name matching with regularized embeddings |
Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5359699A (en) * | 1991-12-02 | 1994-10-25 | General Electric Company | Method for using a feed forward neural network to perform classification with highly biased data |
| WO1995011486A1 (en) * | 1993-10-18 | 1995-04-27 | Miles, Inc. | Method for unsupervised neural network classification with back propagation |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| DE19531967C2 (de) * | 1995-08-30 | 1997-09-11 | Siemens Ag | Verfahren zum Training eines neuronalen Netzes mit dem nicht deterministischen Verhalten eines technischen Systems |
-
1996
- 1996-09-03 DE DE19635758A patent/DE19635758C1/de not_active Expired - Fee Related
-
1997
- 1997-07-15 WO PCT/DE1997/001497 patent/WO1998010352A1/de not_active Ceased
- 1997-07-15 JP JP10512095A patent/JP2000517447A/ja active Pending
- 1997-07-15 DE DE59703304T patent/DE59703304D1/de not_active Expired - Fee Related
- 1997-07-15 EP EP97935429A patent/EP0925541B1/de not_active Expired - Lifetime
- 1997-07-15 US US09/254,298 patent/US6282529B1/en not_active Expired - Fee Related
Patent Citations (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US5359699A (en) * | 1991-12-02 | 1994-10-25 | General Electric Company | Method for using a feed forward neural network to perform classification with highly biased data |
| WO1995011486A1 (en) * | 1993-10-18 | 1995-04-27 | Miles, Inc. | Method for unsupervised neural network classification with back propagation |
Non-Patent Citations (3)
| Title |
|---|
| DATUM M S ET AL: "An artificial neural network for sound localization using binaural cues", JOURNAL OF THE ACOUSTICAL SOCIETY OF AMERICA, JULY 1996, ACOUST. SOC. AMERICA THROUGH AIP, USA, vol. 100, no. 1, ISSN 0001-4966, pages 372 - 383, XP002048951 * |
| SLICHER A W R ET AL: "An innovative approach to training neural networks for strategic management of construction firms", DEVELOPMENTS IN NEURAL NETWORKS AND EVOLUTIONARY COMPUTING FOR CIVIL AND STRUCTURAL ENGINEERING, DEVELOPMENTS IN NEURAL NETWORKS AND EVOLUTIONARY COMPUTING FOR CIVIL AND STRUCTURAL ENGINEERING, CAMBRIDGE, UK, 28-30 AUG. 1995, ISBN 0-948749-36-9, 1995, EDINBURGH, UK, CIVIL-COMP PRESS, UK, pages 87 - 93, XP002048952 * |
| SUNGZOON CHO ET AL: "Evolution of neural network training set through addition of virtual samples", PROCEEDINGS OF 1996 IEEE INTERNATIONAL CONFERENCE ON EVOLUTIONARY COMPUTATION (ICEC'96) (CAT. NO.96TH8114), PROCEEDINGS OF IEEE INTERNATIONAL CONFERENCE ON EVOLUTIONARY COMPUTATION, NAGOYA, JAPAN, 20-22 MAY 1996, ISBN 0-7803-2902-3, 1996, NEW YORK, NY, USA, IEEE, USA, pages 685 - 688, XP002048950 * |
Cited By (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP1388812A1 (de) * | 2002-07-04 | 2004-02-11 | Ronald E. Dr. Kates | Trainingsverfahren für lernfähiges System |
| WO2004006041A3 (en) * | 2002-07-04 | 2004-10-14 | Ronald E Kates | Method for training a learning-capable system |
| US7801839B2 (en) | 2002-07-04 | 2010-09-21 | Kates Ronald E | Method for training a learning-capable system |
| WO2004090807A3 (de) * | 2003-04-10 | 2005-12-22 | Bayer Technology Services Gmbh | Verfahren zum trainieren von neuronalen netzen |
| US7406451B2 (en) | 2003-04-10 | 2008-07-29 | Bayer Technology Services Gmbh | Method for training neural networks |
Also Published As
| Publication number | Publication date |
|---|---|
| US6282529B1 (en) | 2001-08-28 |
| EP0925541B1 (de) | 2001-04-04 |
| EP0925541A1 (de) | 1999-06-30 |
| JP2000517447A (ja) | 2000-12-26 |
| DE19635758C1 (de) | 1997-11-20 |
| DE59703304D1 (de) | 2001-05-10 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| EP2649567B1 (de) | Verfahren zur rechnergestützten modellierung eines technischen systems | |
| DE212020000731U1 (de) | Kontrastives Vortraining für Sprachaufgaben | |
| DE68929102T2 (de) | Lernverarbeitungssystem | |
| DE102007017259B4 (de) | Verfahren zur rechnergestützten Steuerung und/oder Regelung eines technischen Systems | |
| DE69423228T2 (de) | Unüberwachtes klassifizierungsverfahren mit neuronalem netzwerk mit back propagation | |
| EP0925541B1 (de) | Verfahren und vorrichtung zur rechnergestützten generierung mindestens eines künstlichen trainingsdatenvektors für ein neuronales netz | |
| DE69719058T2 (de) | Verbessertes neuronales netzwerk | |
| DE102019116305A1 (de) | Pipelining zur verbesserung der inferenzgenauigkeit neuronaler netze | |
| DE69115488T2 (de) | Neuronale prozessoren für realwerteingangbearbeitung | |
| DE102008020379A1 (de) | Verfahren zur rechnergestützten Steuerung und/oder Regelung eines technischen Systems | |
| DE69223447T2 (de) | Lernverfahren für neuronales Netzwerk und Klassifizieranlage zum Anwenden dieses Verfahrens | |
| DE4217832A1 (de) | Mustererkennungs- und musterlerngeraet | |
| EP0901658B1 (de) | Verfahren zur optimierung eines fuzzy-regelsatzes durch einen rechner | |
| DE69333247T2 (de) | Trainingsverfahren und Gerät zur Erzeugung eines neues Neurons | |
| DE19504664C2 (de) | Chaotisches rekurrentes neuronales Netz und Lernverfahren dafür | |
| DE102018212155A1 (de) | Verfahren und eine Vorrichtung zum aktiven Einregeln eines Prädiktors auf ein Eingangssignal | |
| DE69809402T2 (de) | Assoziativneuron in einem künstlichen neuralen netzwerk | |
| DE4330847A1 (de) | Vorrichtung und Verfahren zur Datenverarbeitung | |
| WO1998034176A1 (de) | Verfahren zur transformation einer zur nachbildung eines technischen prozesses dienenden fuzzy-logik in ein neuronales netz | |
| WO1998007100A1 (de) | Rechnergestütztes verfahren zur auswahl von trainingsdaten für ein neuronales netz | |
| EP0591259B1 (de) | Verfahren zur verarbeitung von unsicherheiten von eingangsdaten in neuronalen netzwerken | |
| EP1093639A2 (de) | Neuronales netz und verfahren und anordnung zum trainieren eines neuronalen netzes | |
| EP0548127A1 (de) | Neuronales Netzwerk und Schaltungsanordnung zur Bool'schen Realisierung neuronaler Netze vom ADALINE-Typ. | |
| DE19641286C1 (de) | Lernverfahren für selbstorganisierende neuronale Netzwerke und Vorrichtung zur Durchführung des Verfahrens | |
| DE69023034T2 (de) | Datenverarbeitungssystem. |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| AK | Designated states |
Kind code of ref document: A1 Designated state(s): JP US |
|
| AL | Designated countries for regional patents |
Kind code of ref document: A1 Designated state(s): AT BE CH DE DK ES FI FR GB GR IE IT LU MC NL PT SE |
|
| DFPE | Request for preliminary examination filed prior to expiration of 19th month from priority date (pct application filed before 20040101) | ||
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application | ||
| WWE | Wipo information: entry into national phase |
Ref document number: 1997935429 Country of ref document: EP |
|
| ENP | Entry into the national phase |
Ref country code: JP Ref document number: 1998 512095 Kind code of ref document: A Format of ref document f/p: F |
|
| WWE | Wipo information: entry into national phase |
Ref document number: 09254298 Country of ref document: US |
|
| WWP | Wipo information: published in national office |
Ref document number: 1997935429 Country of ref document: EP |
|
| WWG | Wipo information: grant in national office |
Ref document number: 1997935429 Country of ref document: EP |


