Statistical impact analysis machine

Fornell, Claes ; Cha, Jaesung ; et al.

2014

Online Patent

Zugriff:

View record in USPTO Patent Grants (Volltext)

A computer-implemented initial run module processes manifest variable data using computer-defined model specification parameters stored in a database to provide initial estimates of weights that are associated with latent variables. The initial run module employs a unique value-based weighting partial least squares computer-implemented process. A final run module then operates upon the manifest variable data to determine the importance of the predictor values that are then used to control the industrial, manufacturing or commercial process. The final run module implements a unique patient partial least squares regression model utilizing a boosting learning technique.

Titel:	Statistical impact analysis machine
Autor/in / Beteiligte Person:	Fornell, Claes ; Cha, Jaesung ; Doriot, Philip Debard
Link:	View record in USPTO Patent Grants (Volltext)
Veröffentlichung:	2014
Medientyp:	Patent
Sonstiges:	Nachgewiesen in: USPTO Patent Grants Sprachen: English Patent Number: 8,666,515 Publication Date: March 04, 2014 Appl. No: 13/143422 Application Filed: January 07, 2009 Assignees: CFI Group USA, LLC (Ann Arbor, MI, US) Claim: 1. A computer-implemented apparatus for controlling a process based on measured physical attributes, comprising: a non-transitory computer-readable storage medium having data structures for storing: (a) manifest variables based on said measured physical attributes; (b) latent variables representing causally related attributes associated with said manifest variables, the latent variables including predictor latent variables and dependent latent variables, wherein the dependent latent variables may be expressed as a linear combination of at least one predictor latent variables; (c) model specification parameters including a path structure parameter indicating causal path attributes to express the causal relationship between the predictor latent variables and the dependent latent variables and a value-based parameter indicating prediction priorities associated with weights of the dependent latent variables with respect to the predictor latent variables; a non-transitory computer-readable storage medium having encoded therein an initial run module that operates upon said data structures to provide estimates of weights associated with said latent variables; said initial run module providing estimates of weights of manifest variables with respect to latent variables by employing a computer-implemented value-based weighting partial least squares process employing an inside approximation weighting scheme utilizing the value based parameter thereby allowing optimization of each latent variable according to its own prediction priorities; the non-transitory computer-readable storage medium having encoded therein a final run module using said estimated weights and said manifest variables to calculate latent variable scores, wherein the latent variable scores are defined as weighted averages of manifest variables; said final run module employing a computer implemented patient partial least squares regression process that operates to calculate the path coefficients between predictor latent variables and dependent latent variables using the latent variable scores and path structure parameter; said final run module providing control parameter selections for controlling said process, where the control parameters are selected using the path coefficients associated with said latent variables. Claim: 2. The apparatus of claim 1 wherein said patient partial least squares regression employed by said final run module utilizes a boosting machine learning algorithm and a forward stagewise learning algorithm with shrinking to slow the regression process. Claim: 3. The apparatus of claim 2 wherein the boosting machine learning algorithm utilizes PLS operator as a weak learner. Claim: 4. The apparatus of claim 1 wherein the computer implemented value-based weighting patient least squares process employs an inside approximation weighting scheme utilizing a matrix of correlation coefficients of the latent variables to estimate the values of the latent variables when the prediction priorities of a latent variable are undefined. Claim: 5. The apparatus of claim 1 wherein the data structure for storing model specification parameters includes a measurement model specification matrix representing causal relationship between the manifest variables and the latent variables. Claim: 6. The apparatus of claim 1 wherein the model specification parameters are initially defined by a user of said apparatus. Claim: 7. The apparatus of claim 6 wherein the model specification parameters are automatically updated by the initial run module based on the initial estimated weights. Claim: 8. The apparatus of claim 7 wherein the initial run module re-estimates the weights of the manifest variables upon determining that the model specification parameters were automatically updated. Claim: 9. The apparatus of claim 1 wherein the final run module is operable to determine a correlation of a first latent variable to a second latent variable. Claim: 10. The apparatus of claim 1 wherein the final run module is operable to iteratively calculate a prediction error sum of squares and to cease iterations when the prediction error sum of squares is minimized. Claim: 11. A computer-implemented apparatus for controlling a process based on measured physical attributes, comprising: a non-transitory computer-readable storage medium storing a computer-defined model having data structures for storing: (a) manifest variables based on said measured physical attributes; (b) latent variables representing causally related attributes associated with said manifest variables, the latent variables including predictor latent variables and dependent latent variables, wherein the dependent latent variables may be expressed as a linear combination of at least one predictor latent variables; (c) model specification parameters including a path structure parameter indicating causal path attributes to express the causal relationship between the predictor latent variables and the dependent latent variables and a value-based parameter indicating prediction priorities associated with weights of the dependent latent variables with respect to the predictor latent variables; an initial run module that operates upon said model to provide initial estimates of weights of manifest variables associated with said latent variables; said initial run module employing a computer-implemented value-based weighting partial least squares process to calculate inside approximations of the latent variables, wherein said initial run module is configured to rotate the latent variables based upon the prediction priorities which are used for said inside approximations of the latent variables; and a final run module that operates upon said model after said initial run module to provide estimates of path coefficients associated with said latent variables; said final run module providing control parameter selections for controlling said process, where the control parameters are selected using the estimates of path coefficients associated with said latent variables. Claim: 12. The apparatus of claim 11 wherein said final run module employs a patient least squares regression model that utilizes a boosting machine learning algorithm, a forward stagewise learning algorithm with shrinking to slow the learning. Claim: 13. The apparatus of claim 12 wherein the boosting machine learning algorithm utilizes PLS operator as a weak learner. Claim: 14. The apparatus of claim 12 wherein the final run module further comprises a latent variable calculator receiving manifest variables from the data structure storing the manifest variables and the estimated weights from the initial run module and calculating latent variable scores by calculating the weighted average of the manifest variables. Claim: 15. The apparatus of claim 11 wherein the non-transitory computer-readable storage medium has a data structure for storing model specification parameters including a matrix representing relationship between the manifest variables and the latent variables. Claim: 16. The apparatus of claim 11 wherein the initial run module updates the model specification parameters after calculating the weights of manifest variables. Claim: 17. A computer-implemented apparatus for controlling a process based on measured physical attributes, comprising: a non-transitory computer-readable storage medium having stored therein data structures storing: (a) manifest variables based on said measured physical attributes; (b) latent variables representing causally related attributes associated with said manifest variables, the latent variables including predictor latent variables and dependent latent variables, wherein the dependent latent variables may be expressed as a linear combination of at least one predictor latent variables; (c) model specification parameters including a path structure parameter indicating causal path attributes to express the causal relationship between the predictor latent variables and the dependent latent variables and a value-based parameter indicating prediction priorities associated with weights of the dependent latent variables with respect to the predictor latent variables; a non-transitory computer-readable storage medium having encoded therein a partial least squares calculator employing a computer-implemented value-based weighting partial least squares process to estimate weights of manifest variables with respect to latent variables; the value-based weighting partial least squares process utilizing the value-based parameter to estimate the values of the latent variables; the estimated values of the latent variables being used by the partial least squares calculator to estimate the weights of manifest variables; the non-transitory computer-readable storage medium having encoded therein a latent variable calculator that receives manifest variables from the data structure storing said manifest variables and estimated weights from the partial least squares calculator and calculates a latent variable score by determining a weighted average of the manifest variables; the non-transitory computer-readable storage medium having encoded therein a regression calculator employing a computer-implemented patient partial least squares regression process to calculate path coefficients between predictor latent variables and dependent latent variables using the latent variable scores and path structure parameter; the patient partial least squares regression process utilizes a boosting machine learning algorithm and a forward stagewise learning algorithm with shrinking to slow the regression process; and the non-transitory computer-readable storage medium having a data structure for storing control parameter selections for controlling said process, where the control parameters are based on the correlation coefficients associated with said latent variables. Claim: 18. A computer-implemented method for controlling a process based on measured physical attributes, comprising: receiving raw data representing the measured physical attributes; initializing a manifest variable data structure residing on a computer memory for storing manifest variable data, deriving manifest variable data from the raw data and storing said manifest variables in the manifest variable data structure; initializing a latent variable data structure residing on a computer memory for storing latent variables, latent variables including predictor latent variables and dependent latent variables, wherein the dependent latent variables may be expressed as a linear combination of at least one predictor latent variables; initializing a model specification parameter data structure residing on a computer memory for storing model specification parameters, the model specification parameters including a path structure parameter indicating causal path attributes to express the causal relationship between the predictor latent variables and the dependent latent variables and a value-based parameter indicating prediction priorities associated with weights of the dependent latent variables with respect to the predictor latent variables; estimating weights of manifest variables with respect to the latent variables by employing a value-based weighting partial least squares algorithm which utilizes the value-based parameter to estimate latent variables scores and utilizes the latent variable scores to estimate the weights of manifest variables; calculating latent variable scores by calculating the weighted averages of manifest variables using said estimated weights and said manifest variables; storing said latent variable scores in the latent variable data structure; calculating path coefficients of predictor latent variables in relation to dependent latent variables by employing a patient partial least squares regression process that utilizes a boosting machine learning algorithm and a forward stagewise learning algorithm with shrinking to slow the regression process, wherein the patient partial least squares regression uses the latent variable scores and the path structure matrix to calculate the path coefficients; and controlling said process using control parameters, where the control parameters are based on the correlation coefficients associated with said latent variables. Claim: 19. The method of claim 18 further comprising: updating the model specification parameters residing in the model specification parameter data structure according to the estimated weights. Claim: 20. The method of claim 19 further comprising: re-estimating weights of manifest variables with respect to the latent variables by employing a value-based weighting partial least squares algorithm using the updated model specification parameters. Claim: 21. A computer-implemented apparatus for controlling a process based on measured physical attributes, comprising: a non-transitory computer-readable storage medium storing a computer-defined model having data structures for storing: (a) manifest variables based on said measured physical attributes; (b) latent variables representing causally related attributes associated with said manifest variables; (c) model specification parameters including a path structure parameter indicating causal path attributes to express the causal relationship between the predictor latent variables and the dependent latent variables and a value-based parameter indicating prediction priorities associated with weights of the dependent latent variables with respect to the predictor latent variables; an initial run module that operates upon said model to provide initial estimates of weights associated with said latent variables; a final run module that operates upon said model after said initial run module to provide final estimates of weights associated with said latent variables; said final run module having a patient partial least squares regression module employing a machine learning algorithm that uses a forward stagewise technique with shrinkage to slow down the learning rate to thereby reduce overfitting and further employs a machine learning algorithm that uses a boosting technique; said final run module providing path coefficients between dependent latent variables and predictor latent variables; said final run module providing control parameter selections for controlling said process, where the control parameters are selected using the final estimates of weights associated with said latent variables. Claim: 22. A non-transitory computer-readable storage having encoded therein computer readable instructions for determining the statistical impact of a plurality of physical attributes expressed as manifest variables comprising: an input module operable to receive raw data and to store said raw data in a raw data database; a user interface module operable to receive inputs from a user to initialize a computer memory having data structures for storing; (a) manifest variables based on said measured physical attributes; (b) latent variables representing causally related attributes associated with said manifest variables, the latent variables including predictor latent variables and dependent latent variables, wherein the dependent latent variables may be expressed as a linear combination of at least one predictor latent variables; (c) model specification parameters including a path structure parameter indicating causal path attributes to express the causal relationship between the predictor latent variables and the dependent latent variables and a value-based parameter indicating prediction priorities associated with weights of the dependent latent variables with respect to the predictor latent variables; an initial run module operable to receive manifest variables and model specification parameters and to estimate weights of manifest variables using a value-based weighting patient least squares algorithm that employs an inside approximation weighting scheme utilizing the value-based parameter wherein estimated weights are based on the prediction priorities of the latent variables; the initial run module further operable to update model specification parameters residing in the model specification parameter database based on results of the value-weighting patient least squares algorithm a final run module operable to receive the manifest variable data and the model specification parameters and to re-estimate weights of the manifest variables using the value-weighting patient least squares algorithm and updated model specification parameters; the final run module further operable to calculate latent variable scores for each latent variable using the re-estimated weights and the manifest variable data, wherein the calculated latent variable scores are the weighted averages of the manifest variables, and wherein the latent variable scores are stored in said data structure storing latent variables; the final run module further operable to calculate the path coefficients of a dependent latent variable utilizing a patient partial least squares regression algorithm that uses a forward stagewise technique with shrinkage to slow down the learning rate to thereby reduce overfitting and using a boosting machine learning algorithm; and the final run module providing control parameter selections for controlling said process, where the control parameters are selected using the final estimates of weights associated with said latent variables. Claim: 23. The non-transitory computer-readable storage medium of claim 22 wherein the initial run module is further operable to estimate weights using an inside approximation scheme based on a correlation of latent variables when the prediction priorities of the latent variables are undefined. Claim: 24. The non-transitory computer-readable storage medium of claim 22 wherein the final run module is operable to determine a correlation of a first latent variable to a second latent variable. Claim: 25. A computer-implemented apparatus for controlling a process based on measured physical attributes, comprising: a non-transitory computer-readable storage medium having stored therein data structures storing: (a) manifest variables based on said measured physical attributes; (b) latent variables representing causally related attributes associated with said manifest variables, the latent variables including predictor latent variables and dependent latent variables, wherein the dependent latent variables may be expressed as a linear combination of at least one predictor latent variables; (c) model specification parameters including a path structure parameter indicating causal path attributes to express the causal relationship between the predictor latent variables and the dependent latent variables, a value-based parameter indicating prediction priorities associated with weights of the dependent latent variables with respect to the predictor latent variables, and a measurement model parameter indicating causal relationships between the manifest variables and the latent variables; a non-transitory computer-readable storage medium having encoded therein an initial run module receiving manifest variables from the data structure storing the manifest variables and model specification parameters from the data structure storing model parameters, the model specification parameters including the value-based parameter and the measurement model parameter; initial run module having a means for estimating weights using a value-based partial least squares algorithm; a non-transitory computer-readable storage medium having encoded therein a final run module receiving manifest variables from the data structure storing the manifest variables and model specification parameters from the data structure storing model parameters, the model specification parameters including the path structure parameter; the final run module calculating latent variable scores based on the manifest variable and the estimated weights; the final run module having a means for calculating path coefficients using a patient partial least squares algorithm; the final run module providing control parameter selections for controlling said process, where the control parameters are selected using the final estimates of weights associated with said latent variables. Current U.S. Class: 700/28 Patent References Cited: 6192319 February 2001 Simonson et al. ; 2002/0169658 November 2002 Adler ; 2002/0177909 November 2002 Fu et al. ; 2008/0082195 April 2008 Samardzija ; 2008/0215386 September 2008 Eder Primary Examiner: Hartman, Jr., Ronald Attorney, Agent or Firm: Harness, Dickey & Pierce, PLC

Klicken Sie ein Format an und speichern Sie dann die Daten oder geben Sie eine Empfänger-Adresse ein und lassen Sie sich per Email zusenden.

BibTeX Citavi, JabRef, u.a.
(Literaturverwaltung)

PDF kein Volltext!
(Merkzettel, Notizen)

RIS Endnote, Citavi u.a.
(Literaturverwaltung)

MODS
(XML zur Weiterverarbeitung)

oder

Wählen Sie das für Sie passende Zitationsformat und kopieren Sie es dann in die Zwischenablage, lassen es sich per Mail zusenden oder speichern es als PDF-Datei.

Gewünschter Zitations-Stil:

oder

Bitte prüfen Sie, ob die Zitation formal korrekt ist, bevor Sie sie in einer Arbeit verwenden. Benutzen Sie gegebenenfalls den "Exportieren"-Dialog, wenn Sie ein Literaturverwaltungsprogramm verwenden und die Zitat-Angaben selbst formatieren wollen.