Prosecution Insights
Last updated: November 05, 2026
Application No. 17/199,389

ACCELERATED K-FOLD CROSS-VALIDATION

Final Rejection §101§103§112
Filed
Mar 11, 2021
Examiner
KWON, JUN
Art Unit
2127
Tech Center
2100 — Computer Architecture & Software
Assignee
Minitab LLC
OA Round
6 (Final)
43%
Grant Probability
Moderate
7-8
OA Rounds
0m
Est. Remaining
83%
With Interview

Examiner Intelligence

Grants 43% of resolved cases
43%
Career Allowance Rate
35 granted / 82 resolved
-12.3% vs TC avg
Strong +40% interview lift
Without
With
+40.0%
Interview Lift
resolved cases with interview
Typical timeline
4y 9m
Avg Prosecution
31 currently pending
Career history
126
Total Applications
across all art units

Statute-Specific Performance

§101
29.0%
-11.0% vs TC avg
§103
45.7%
+5.7% vs TC avg
§102
9.4%
-30.6% vs TC avg
§112
14.8%
-25.2% vs TC avg
Black line = Tech Center average estimate • Based on career data from 82 resolved cases

Office Action

§101 §103 §112
Detailed Action This Office Action is in response to the remarks entered on 08/03/2026. Amended claims 22, 31 and 35 have been entered herein. New claim 43 has been entered herein. Claims 22-23, 25-32, 34-36, 38-43 are currently pending. Notice of Pre-AIA or AIA Status The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . Claim Objections Amended claim 31 has been entered. Claim Objection has been withdrawn. Claim Rejections - 35 USC § 112 The following is a quotation of the first paragraph of 35 U.S.C. 112(a): (a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention. The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112: The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention. Claims 22-23, 25-32, 34-36, 38-43 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention. Claim 22 recites “(c) partitioning, by the at least one processor, the labeled dataset into K folds, each fold having a fold size;” Nowhere in the specification explicitly explains that each fold has ‘fold size’ value. The limitation seems to recite “each fold having a ‘fold size’ value”, for example, each fold has [n1, n2, … nk, ‘fold size’]. Specification [0070] explains that the input data 302 is partitioned into k folds with sizes ni … nk and does not explain that each fold has ‘fold size’. Accordingly, the examiner concluded that the claim fails to comply with the written description requirement. Claims 23, 25-30 and 43 depend from claim 22 and inherit the same deficiency. Claim 31 is a machine claim which recites the same feature as claim 22 and rejected for at least the same reasons. Claims 32 and 34 depend from claim 31 and inherit the same deficiency. Claim 35 is a machine claim which recites the same feature as claim 22 and rejected for at least the same reasons. Claims 36, 38-42 depend from claim 35 and inherit the same deficiency. Claim Rejections - 35 USC § 101 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 22-23, 25-32, 34-36, and 38-43 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Regarding claim 22, Step 1: Claim 22 recites a computer-implemented method. Therefore, it is directed to the statutory category of Processes. 2A Prong 1: (b), the single fit computing (i) a vector of estimated model coefficients; and (ii) a variance-covariance matrix X T X - 1 of the estimated model coefficients; (mathematical concept, in light of the spec [0074]) (c) partitioning, , each fold having a fold size; (mental process of observation – dividing dataset into subsets can be performed in one’s mind and does not require a computer component) for each of the K folds, deriving a covariance submatrix corresponding to the fold from the stored variance-covariance matrix without recomputing the variance-covariance matrix and without re-fitting the predictive model (mathematical calculation – See spec [0074]) (e) for each fold of the K folds, without retraining the predictive model: (i) forming, from the stored variance-covariance matrix and rows X ( j ) of the design matrix corresponding to the fold, a fold matrix ( I ( j )   -   H ( j ) ) , where H ( j )   = X j X T X - 1 X j T ; (ii) responsive to the number p of predictor variables being less than the fold size, computing an inverse of the fold matrix by inverting a p x p matrix ( X T X   -   X j T X ( j ) ) in accordance with a Sherman-Morrison-Woodbury transformation, thereby reducing a dimension of the matrix that is inverted from the fold size to the number p of predictor variables; (iii) computing predicted residuals for observations in the fold by applying the computed inverse to ordinary residuals of the fold, the ordinary residuals being differences between output values of the observations in the fold and estimated values obtained from the single fit; and (mathematical concept, in light of the spec [0073-0074]) (f) aggregating the predicted residuals stored across the K folds to compute a cross-validation performance statistic for the predictive model; and (mathematical concept. The k-fold R-squared statistics is a mathematical concept in lights of the spec [0068]) (g) selecting the predictive model based on the computed cross-validation performance statistic, wherein the K-fold cross-validation is performed using the single fit of the predictive model and reuse of the stored variance-covariance matrix across the K folds in place of K separate re-fittings of the predictive model, so that a number of processor operations for the cross-validation does not scale with the number K of folds. (mental process of evaluation – selecting the best model based on the performance metric can be done in one’s mind and does not require a computer component) 2A Prong 2: A computer-implemented method for selecting a predictive model using K-fold cross-validation. comprising: (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) (a) receiving, by at least one processor, a labeled dataset comprising a plurality of observations and a plurality of p predictor variables arranged as a design matrix X; (insignificant extra-solution activity MPEP 2106.05(g)(iii) of gathering statistics) (b) training, by the at least one processor, a predictive model in a single fit on all n observations of the labeled dataset, (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f) – generic machine learning model training process) (c) partitioning, by the at least one processor (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) (d) storing, the variance-covariance matrix X T X - 1 as a computational artifact in a memory accessible to the at least one processor after the single fit, (insignificant extra-solution activity MPEP 2106.05(g) of mere data gathering. adding a final step of storing data to a process that only recites a mathematical concept does not add a meaningful limitation to the process of computing the area) (ii) storing the predicted residuals for the fold; (insignificant extra-solution activity MPEP 2106.05(g) of mere data gathering. adding a final step of storing data to a process that only recites a mathematical concept does not add a meaningful limitation to the process of computing the area) The additional elements as disclosed above alone or in combination do not integrate the judicial exception into practical application as they are mere insignificant extra solution activity and generic computer functions that are implemented to perform the disclosed abstract idea above. 2B: A computer-implemented method for selecting a predictive model using K-fold cross-validation. comprising: (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) (a) receiving, by at least one processor, a labeled dataset comprising a plurality of observations and a plurality of p predictor variables arranged as a design matrix X; (indicated as an insignificant extra-solution activity MPEP 2106.05(g) in Step 2A Prong 2. Therefore, the limitation is re-evaluated in Step 2B as well understood, routine and conventional activity MPEP 2106.05(d)(i) of receiving or transmitting data over a network) (b) training, by the at least one processor, a predictive model in a single fit on all n observations of the labeled dataset, (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f) – generic machine learning model training process) (c) partitioning, by the at least one processor (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) (d) storing, the variance-covariance matrix X T X - 1 as a computational artifact in a memory accessible to the at least one processor after the single fit, (indicated as an insignificant extra-solution activity MPEP 2106.05(g) in Step 2A Prong 2. Therefore, the limitation is re-evaluated in Step 2B as well understood, routine and conventional activity MPEP 2106.05(d)(iv) of Storing and retrieving information in memory) (ii) storing the predicted residuals for the fold; (indicated as an insignificant extra-solution activity MPEP 2106.05(g) in Step 2A Prong 2. Therefore, the limitation is re-evaluated in Step 2B as well understood, routine and conventional activity MPEP 2106.05(d)(iv) of Storing and retrieving information in memory) The additional elements as disclosed above in combination of the abstract idea are not sufficient to amount to significantly more than the judicial exception as they are well, understood, routine and conventional activity and generic computer functions that are implemented to perform the disclosed abstract idea above. Regarding claim 23, Step 1: Processes, as above. 2A Prong 1: Incorporates the rejection of claim 22. 2A Prong 2: wherein the predictive model is a machine learning model. (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) 2B: wherein the predictive model is a machine learning model. (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) Regarding claim 25, Step 1: Processes, as above. 2A Prong 1: Incorporates the rejection of claim 22. 2A Prong 2: wherein the selected predictive model is used to predict outcomes for patients in a coronary heart disease prediction system. (field of use and technological environment MPEP 2106.05(h). The claim merely limits the field of use of the predictive model) 2B: wherein the selected predictive model is used to predict outcomes for patients in a coronary heart disease prediction system. (field of use and technological environment MPEP 2106.05(h). The claim merely limits the field of use of the predictive model) Regarding claim 26, Step 1: Processes, as above. 2A Prong 1: The method of claim 22, wherein the predictive model is trained based on an ordinary least squares (OLS) method and the model is a linear model. (mathematical concept, in light of the claim language itself. OLS method is a math calculation) 2A Prong 2: This judicial exception is not integrated into a practical application. 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Regarding claim 27, Step 1: Processes, as above. 2A Prong 1: The method of claim 22, wherein the predictive model is trained based on an iterative reweighted least squares (IRLS) method and the model is a generalized linear model. (mathematical concept, in light of the claim language itself. OLS method is a math calculation) 2A Prong 2: This judicial exception is not integrated into a practical application. 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Regarding claim 28, Step 1: Processes, as above. 2A Prong 1: The method of claim 22, wherein an output variable of the labeled dataset has a distribution selected from the group consisting of a binomial distribution, a Poisson distribution, a gamma distribution, and a negative binomial distribution. (mathematical concept, in light of the claim language itself) 2A Prong 2: This judicial exception is not integrated into a practical application. 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Regarding claim 29, Step 1: Processes, as above. 2A Prong 1: The method of claim 22, wherein each of the k folds comprises a substantially equal number of labeled cases. (mathematical concept, in light of the claim language itself) 2A Prong 2: This judicial exception is not integrated into a practical application. 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Regarding claim 30, Step 1: Processes, as above. 2A Prong 1: The method of claim 22, wherein the predicted residuals for a fold are computed using a one-step approximation of model parameters derived from the model trained on the labelled dataset. (mathematical concept, in light of the claim language itself) 2A Prong 2: This judicial exception is not integrated into a practical application. 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Regarding claim 31, Step 1: Claim 31 recites a computing system comprising: one or more processors; and a memory storing instructions. Therefore, it is directed to the statutory category of a machine. 2A Prong 1: Claim 31 is an apparatus claim which recites the same feature as the method claim 22, and is rejected for at least the same reasons. 2A Prong 2: A computing system comprising: one or more processors; and a memory storing instructions that, when executed by the one or more processors, cause the computing system to: (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) 2B: A computing system comprising: one or more processors; and a memory storing instructions that, when executed by the one or more processors, cause the computing system to: (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) Claim 32 is an apparatus claim which recites the same feature as the method claim 23, and is rejected for at least the same reasons. Claim 34 is an apparatus claim which recites the same feature as the method claim 25, and is rejected for at least the same reasons. Regarding claim 35, Step 1: Claim 31 recites a non-transitory computer readable storage medium storing program instructions. Therefore, it is directed to the statutory category of a machine. 2A Prong 1: Claim 35 is an apparatus claim which recites the same feature as the method claim 25, and is rejected for at least the same reasons. 2A Prong 2: A non-transitory computer readable storage medium storing program instructions, when executed by a processor, causes the processor to: (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) 2B: A non-transitory computer readable storage medium storing program instructions, when executed by a processor, causes the processor to: (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) Regarding claim 36, Step 1: A machine, as above. 2A Prong 1: Incorporates the rejection of claim 35. 2A Prong 2: The non-transitory computer readable storage medium of claim 35, wherein the predictive modeling is performed to train one or more machine learning models. (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) 2B: The non-transitory computer readable storage medium of claim 35, wherein the predictive modeling is performed to train one or more machine learning models. (mere instructions to apply an exception using a generic computer component MPEP 2106.05(f)) Claim 38 is an apparatus claim which recites the same feature as the method claim 25, and is rejected for at least the same reasons. Claim 39 is an apparatus claim which recites the same feature as the method claim 26, and is rejected for at least the same reasons. Claim 40 is an apparatus claim which recites the same feature as the method claim 27, and is rejected for at least the same reasons. Claim 41 is an apparatus claim which recites the same feature as the method claim 28, and is rejected for at least the same reasons. Claim 42 is an apparatus claim which recites the same feature as the method claim 29, and is rejected for at least the same reasons. Regarding claim 43, Step 1: Processes, as above. 2A Prong 1: The method of claim 22, wherein the variance-covariance matrix is computed a single time by the single fit and is retained in the memory and accessed for each of the K folds to derive the corresponding covariance submatrix without recomputing the variance-covariance matrix, such that a number of model-fitting operations performed for the K folds is independent of the number K of folds, wherein the labeled dataset is big data comprising at least 10,000 labeled cases. (mathematical calculation, See spec [0074]) 2A Prong 2: This judicial exception is not integrated into a practical application. 2B: The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. Claim Rejections - 35 USC § 103 In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status. The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. Claims 22-23, 25-26, 29, 31-32, 34-36, 38-39, and 42 are rejected under 35 U.S.C. 103 as being unpatentable over Gianola et al. (“Cross-Validation Without Doing Cross-Validation in Genome-Enabled Prediction”, hereafter ‘Gianola’) in view of Cheng et al. (“Cross- validation of best linear unbiased predictions of breeding values using an efficient leave-one-out strategy”, 2020, hereinafter ‘Cheng’) and further in view of Mueller et al. (US 20090132443 A1, hereafter ‘Mueller’). Regarding Claim 22, Gianola teaches: (a) receiving, by at least one processor, a labeled dataset comprising a plurality of observations and a plurality of p predictor variables arranged as a design matrix X; (Page 3107, Right Column, Paragraph 2, Sentences 1-3; Page 3108, Left Column, Paragraph 5, Sentences 1-2; Page 3118, Right Column, Paragraph 1, Sentences 3-5; Appendix C; Shows that labeled data is inputted into a model as it learns beta from measurements of x and labels of y, [page 3108, left col, Setting, lines 3-5] The marker matrix X is the n x p marker matrix that contains variables) (b) training, by the at least one processor, a predictive model in a single fit on all n observations of the labeled dataset, the single fit computing (Page 3107, Right Column, Paragraph 2, Sentences 1-3; Page 3108, Left Column, Paragraph 5, Sentences 1-2; Page 3118, Right Column, Paragraph 1, Sentences 3-5; Page 3120, Left Column, Paragraph 3, Sentence 4; Appendix C; Shows that the model is trained & parametrized upon the labelled data, [page 3107, ABSTRACT, lines 12-13] shows that the model retraining is not required for the cross-validation method and [page 3108, left col, Setting, line 1 – right col, the last line] shows no retraining process which means that the training is done in a single fit) (i) a vector of estimated model coefficients and ([Page 3108, Left Column, Setting, line 1 – Right Column, line 7] The data is obtained based on the equation y = X β + e , wherein y denotes the vector of phenotypic measurements. The phenotypic measurements are the vector of estimated model coefficient) (ii) a variance-covariance matrix X T X - 1 of the estimated model coefficients: ([Page 3108, Left Column, Setting, line 1 – Right Column, line 15] The OLS estimator of β is β = X ' X - 1 X ' y . The X ' X - 1 is the variance-covariance matrix. The OLS estimator is the estimated model coefficient) (c) partitioning, by the at least one processor, the labeled dataset into K folds, each fold having a fold size; (Page 3107, Right Column, Paragraph 2, Sentences 1-3; Page 3118, Right Column, Paragraph 1, Sentences 3-5; Appendix C; Shows that the labelled data is distributed into a number of folds) (d) storing, the variance-covariance matrix X T X - 1 as a computational artifact in a memory accessible to the at least one processor after the single fit, ; ([Page 3108, Left Column, Setting, line 1 – Right Column, line 15] The OLS estimator of β is β = X ' X - 1 X ' y . The X ' X - 1 is the variance-covariance matrix. The OLS estimator is the estimated model coefficient. [Page 3111, Left Column, below equation (30), line 13-20], [Page 3113, Right Column, 4th paragraph, line 1-4] and [Page 3119, Figure 13] indicates that the algorithms are executed using a computer. It is inherent that storing is performed to implement the algorithm) (e) for each fold of the K folds, and without retraining the predictive model for the fold: (Page 3108, Right Column, Paragraph Leave-one-out (LOO) cross-validation, line 1-5; Eqn. 2; The Xs are the training folds that are received by the LOO cross validation method. [Page 3107, ABSTRACT, line 12-13] shows that the method is performed without model retraining. [Page 3111, Left Column, below equation (30), line 13-20], [Page 3113, Right Column, 4th paragraph, line 1-4] and [Page 3119, Figure 13] indicate that the algorithms are executed using a computer. It is inherent that storing is performed to implement the algorithm) (i) forming, from the stored variance-covariance matrix and rows X ( j ) of the design matrix corresponding to the fold, a fold matrix ( I ( j )   -   H ( j ) ) , where H ( j )   = X j X T X - 1 X j T ; ([page 3109, left col, lines 1-3] The matrix h i i is computed x i ' X ' X - 1 x i wherein the ‘ denotes ‘transformation’, x i ' denotes X j T . The fold matrix is calculated at the denominator of Eqn (6) y i - x i ' β 1 - h i i ) (ii) responsive to the number p of predictor variables being less than the fold size, computing an inverse of the fold matrix by inverting a p x p matrix ( X T X   -   X j T X ( j ) ) in accordance with a Sherman-Morrison-Woodbury transformation, thereby reducing a dimension of the matrix that is inverted from the fold size to the number p of predictor variables; ([page 3108, right col, Eqn (4) and corresponding paragraphs] and [page 3123, Appendix B] Making use of the Sherman-Morrison-Woodbury transformation in Appendix B eqn (81), the [ X ' X - 1 + X ' X - 1 x i x i ' X ' X - 1 1 - x i ' X ' X - 1 x i ] is an inversion of X ' X - x i x i ' . The matrix before the inversion is the p x p matrix. The final result β - i is formed by deleting i t h observations from the model so reduced dimension) (iii) computing predicted residuals for observations in the fold by applying the computed inverse ordinary residuals of the fold, the ordinary residuals being differences between output values of the observations in the fold and estimated values obtained from the single fit; and ([Page 3109, Eqn. 6 & 11; Page 3109 Left Column, Paragraph 1, Sentences 1-2; Appendix C; Appendix E;] Shows that residuals in a fold is calculated by an inverse equation (i.e., “transformation) to compute a prediction error for the folds. Eqn. 9 and following Sentences 1-4 shows that the inverted matrix has reduced size as d of its rows were removed) (iv) storing the predicted residuals for the fold: and ([Page 3111, Left Column, below equation (30), line 13-20], [Page 3113, Right Column, 4th paragraph, line 1-4] and [Page 3119, Figure 13] indicates that the algorithms are executed using a computer. It is inherent that storing is performed to implement the algorithm) (f) aggregating the predicted residuals stored across the K folds to compute a cross-validation performance statistic for the predictive model: and ([Page 3108, Right Column, Leave-one-out (LOO) cross-validation, line 1-3] The n is the number of training folds. [Page 3109, Left Column, Eqn 8; and Page 3109, left column, line 7-17] The equation 8, 1 n ∑ i = 1 n y i - x i ' β 1 - h i i 2 , denotes that outputs from n number of folds are aggregated to calculate the expected mean squared error of prediction. The expected mean squared error of prediction is the cross-validation performance statistics for the predictive model) However, Gianola does not specifically disclose: A computer-implemented method for selecting a predictive model using K-fold cross-validation. comprising: for each of the K folds, deriving a covariance submatrix corresponding to the fold from the stored variance-covariance matrix without recomputing the variance-covariance matrix and without re-fitting the predictive model (g) selecting the predictive model based on the computed cross-validation performance statistic, wherein the K-fold cross-validation is performed using the single fit of the predictive model and reuse of the stored variance-covariance matrix across the K folds in place of K separate re-fittings of the predictive model, so that a number of processor operations for the cross-validation does not scale with the number K of folds. Cheng teaches: … and, for each of the K folds, deriving a covariance submatrix corresponding to the fold from the stored variance-covariance matrix without recomputing the variance-covariance matrix and without re-fitting the predictive model ([Cheng, page 521, left col, line 35 – right col, line 29] For individual i, the Leave-one-out cross-validation Estimates of Breeding Values can be calculated as u ^ i = g i , t r a W 22 - 1 y - i where g i , t r a is the vector of covariances between individual i and individuals in the training data, W 22 - 1 is the inverse of the covariance matrix for individuals in the training data. The calculation u ^ i does not require recomputing the entire variance-covariance matrix W11, W12, W21 and W22. This process is performed without re-fitting (re-training) the predictive model) wherein the K-fold cross-validation is performed using the single fit of the predictive model and reuse of the stored variance-covariance matrix across the K folds in place of K separate re-fittings of the predictive model, so that a number of processor operations for the cross-validation does not scale with the number K of folds. ([Cheng, page 521, left col, line 35 – right col, line 29] discloses generating covariance submatrices using the pre-stored covariance matrix W. The generation of the covariance matrix does not involve re-fitting (re-training) of the machine learning model, and performed by algebraic operations such as transformation, inversion, and matrix multiplication. Therefore, the number of training does not scale with the number of folds) It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to have combined Cheng’s teaching of storing and reusing variance-covariance matrix across the K-folds and deriving a covariance submatrix from the stored matrix without recomputing the variance-covariance matrix with Gianola’s teaching of calculating predicted residuals for k fold cross-validation of a model, with a reasonable expectation of success, to create a method to calculate predicted residuals to determine the cross-validation error of a model. The motivation/suggestion for doing so is to improve the efficiency of the K-fold cross validation by reducing the number of calculations needed when recomputing variance-covariance matrix and to calculate covariance matrix efficiently based on an algebraic calculations as shown in [Cheng, page 521, left col, line 28 – right col, line 15]. However, Gianola in view of Cheng do not specifically disclose: A computer-implemented method for selecting a predictive model using K-fold cross-validation. comprising: (g) selecting the predictive model based on the computed cross-validation performance statistic Mueller teaches: A computer-implemented method for selecting a predictive model using K-fold cross-validation. comprising: (Paragraph 0094; The classifiers (i.e., models) are trained in N-fold cross validation (303) such as a 5-fold cross validation loop. Each sample is in a validation group once and the likelihood of the sample belonging to the risk group is computed by the trained classifier. The N-fold cross validation results provide for classifier evaluation, analysis of generalization, and the receiver operator characteristic (ROC), which are interpreted as the k-fold R-square statistics. A plurality of models is generated and a model is selected for varying numbers of input features and degrees of complexity (Schroeder et al., BMC Molecular Biology, 7(3) (2006))) (g) selecting the predictive model based on the computed cross-validation performance statistic, (Paragraph 0094; The classifiers (i.e., models) are trained in N-fold cross validation (303) such as a 5-fold cross validation loop. Each sample is in a validation group once and the likelihood of the sample belonging to the risk group is computed by the trained classifier. The N-fold cross validation results provide for classifier evaluation, analysis of generalization, and the receiver operator characteristic (ROC), which are interpreted as the k-fold R-square statistics. A plurality of models is generated and a model is selected for varying numbers of input features and degrees of complexity (Schroeder et al., BMC Molecular Biology, 7(3) (2006)). Each model is assigned a ranking and the model with the highest rank is selected. The selected model is evaluated by measuring the area under the ROC curve (AUC) which provides a balanced measure of the generalization performance) It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to have combined Mueller’s teaching of selecting a model with the highest k-fold R-square statistics with Gianola’s teaching of calculating predicted residuals for k fold cross-validation of a model, with a reasonable expectation of success, to create a method to calculate predicted residuals to determine the cross-validation error of a model, as in Gianola, where a model is tuned and selected based of the cross-validation error, as in Mueller. The motivation/suggestion to make this combination to formulate cross-validation algorithms for model selection and model assessment in classification and regression settings which avoids the pitfalls, and then show results. Regarding claim 23, Gianola teaches: The method of claim 22, wherein the predictive model is a machine learning model. (Page 3107, Right Column, Paragraph 2, Sentences 1-3; Page 3108, Left Column, Paragraph 5, Sentences 1-2; Page 3118, Right Column, Paragraph 1, Sentences 3-5; Appendix C; Shows that labeled data is inputted into a model (i.e., “predictive analytics engine”) as it learns beta from measurements of x and labels of y) Regarding claim 25, Gianola in view of Cheng and further in view of Mueller teaches: The method of claim 22, wherein the selected predictive model is used to predict outcomes for patients in a coronary heart disease prediction system. (Paragraph 0096; Probability borders can be determined by the model for coronary heart disease) Regarding claim 26, Gianola teaches: The method of claim 22, wherein the predictive model is trained based on an ordinary least squares (OLS) method and the model is a linear model. (Page 3108, Left Column, Paragraph 5, Sentences 1-2; Shows that a linear model is used; Page 3108, Right Column, Paragraph 2, Sentences 1-2; Appendix C; Shows that procedure uses the ordinary least squares method) Regarding claim 29, Gianola teaches: The method of claim 22, wherein each of the k folds comprises a substantially equal number of labeled cases. ([Page 3107, right column, line 17-19] shows that most often, Cross Validation consists of dividing a data set with n cases, and each fold contains the phenotypic measurement and a vector of genomic covariables of approximately equal size) Regarding claim 31, Gianola in view of Cheng and further in view of Mueller teaches: A computing system comprising: one or more processors; and a memory storing instructions that, when executed by the one or more processors, cause the computing system to: ([Page 3111, Left Column, below equation (30), line 13-20], [Page 3113, Right Column, 4th paragraph, line 1-4] and [Page 3119, Figure 13] indicates that the algorithms are executed using a computer. It is inherent that storing is performed to implement the algorithm) Claim 31 is an apparatus claim which recites the same feature as the method claim 22, and is rejected for at least the same reasons. Claim 32 is an apparatus claim which recites the same feature as the method claim 23, and is rejected for at least the same reasons. Claim 34 is an apparatus claim which recites the same feature as the method claim 25, and is rejected for at least the same reasons. Regarding claim 35, Gianola in view of Cheng and further in view of Mueller teaches: A non-transitory computer readable storage medium storing program instructions that, when executed by a processor, cause the processor to: ([Page 3111, Left Column, below equation (30), line 13-20], [Page 3113, Right Column, 4th paragraph, line 1-4] and [Page 3119, Figure 13] indicates that the algorithms are executed using a computer. It is inherent that storing is performed to implement the algorithm) Claim 35 is an apparatus claim which recites the same feature as the method claim 22, and is rejected for at least the same reasons. Regarding claim 36, Gianola teaches: The non-transitory computer readable storage medium of claim 35, wherein the predictive modeling is performed to train one or more machine learning models. (Page 3108, Left Column, Paragraph 5, Sentences 1-2; Shows that a linear model is used; Page 3108, Right Column, Paragraph 2, Sentences 1-2; Appendix C; Shows that procedure uses the ordinary least squares method) Claim 38 is an apparatus claim which recites the same feature as the method claim 25, and is rejected for at least the same reasons. Claim 39 is an apparatus claim which recites the same feature as the method claim 26, and is rejected for at least the same reasons. Claim 42 is an apparatus claim which recites the same feature as the method claim 29, and is rejected for at least the same reasons. Regarding claim 43, Gianola teaches: a number of model-fitting operations performed for the K folds is independent of the number K of folds, wherein the labeled dataset is big data comprising at least 10,000 labeled cases. ([page 3118, Figure 12] The samples for LOO Bayesian GBLUP can be 15,000 independent samples) Cheng teaches: The method of claim 22, wherein the variance-covariance matrix is computed a single time by the single fit and is retained in the memory and accessed for each of the K folds to derive the corresponding covariance submatrix without recomputing the variance-covariance matrix, such that a number of model-fitting operations performed for the K folds is independent of the number K of folds, [Cheng, page 521, left col, line 35 – right col, line 29] For individual i, the Leave-one-out cross-validation Estimates of Breeding Values can be calculated as u ^ i = g i , t r a W 22 - 1 y - i where g i , t r a is the vector of covariances between individual i and individuals in the training data, W 22 - 1 is the inverse of the covariance matrix for individuals in the training data. The calculation u ^ i does not require recomputing the entire variance-covariance matrix W11, W12, W21 and W22. This process is performed without re-fitting (re-training) the predictive model) Claims 27 and 40 are rejected under 35 U.S.C. 103 as being unpatentable over Gianola in view of Cheng in view of Mueller and further in view of Valdora et al. (VALDORA et al. " Robust Estimation in High Dimensional Generalized Linear Models", hereinafter ‘Valdora’). Regarding claim 27, Gianola in view of Cheng and further in view of Mueller do not specifically disclose: The method of claim 22, wherein the predictive model is trained based on an iterative reweighted least squares (IRLS) method and the model is a generalized linear model. However, Valdora teaches: The method of claim 22, wherein the predictive model is trained based on an iterative reweighted least squares (IRLS) method and the model is a generalized linear model. (Page 3, Paragraph 3, Sentences 1-3; Page 7, Paragraph 1, Sentences 1-3; Shows that the estimation process contains IRLS method and a generalized linear model) It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to have combined VALDORA’s teaching of generalized linear model in combination of the IRLS method with GIANOLA, Cheng and MUELLER’s teaching of calculating predicted residuals for cross-validation of a model, with a reasonable expectation of success, to create a method to calculate predicted residuals to determine the cross-validation error of a model, as in GIANOLA, Cheng and MUELLER, where the model is a generalized linear model using the IRLS method, as in VALDORA. A person of ordinary skill would have been motivated to make this combination for “a deterministic robust initial estimate for generalized linear models” (Page 13, Paragraph 1, Sentence 1; VALDORA). Claim 40 is an apparatus claim which recites the same feature as the method claim 27, and is rejected for at least the same reasons. Claim 28 and 41 are rejected under 35 U.S.C. 103 as being unpatentable over Gianola in view of Cheng in view of Mueller and further in view of Szeto et al. (US 20180018590 A1, hereinafter ‘Szeto’) Regarding claim 28, Gianola in view of Cheng and further in view of Mueller do not specifically disclose: The method of claim 22, wherein an output variable of the labeled dataset has a distribution selected from the group consisting of a binomial distribution, a Poisson distribution, a gamma distribution, and a negative binomial distribution. Szeto teaches: The method of claim 22, wherein an output variable of the labeled dataset has a distribution selected from the group consisting of a binomial distribution, a Poisson distribution, a gamma distribution, and a negative binomial distribution. (Paragraph [0040]; Paragraph [0073]; Shows that labelled data contain different inputs to determine an output for data. Also, the data can consist of different distributions (i.e., “binomial distribution, a Poisson distribution, a gamma distribution, and a negative binomial distribution.”)) It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to have combined SZETO’s teaching of an input and output variable with varying distributions with , GIANOLA, Cheng, MUELLER’s teaching of calculating predicted residuals for cross-validation of a generalized linear model, with a reasonable expectation of success, to create a method to calculate predicted residuals to determine the cross-validation error of a model, as in GIANOLA, Cheng and MUELLER, where the cases have one or more variables and contains a distribution, as in SZETO. A person of ordinary skill would have been motivated to make this combination to “Reducing the dimensionality of the data further improves performance of the computing system by decreasing compute time and by decreasing the transmission time.”. (Paragraph [0088], SZETO) Claim 41 is an apparatus claim which recites the same feature as the method claim 28, and is rejected for at least the same reasons. Claim 30 is rejected under 35 U.S.C. 103 as being unpatentable over Gianola in view of Cheng in view of Mueller and further in view of Jung et al. (Jung et al. " Multiple Predicting K-fold Cross-validation for model selection", hereinafter ‘Jung’). Regarding claim 30, Gianola in view of Cheng and further in view of Mueller do not specifically disclose: The method of claim 22, wherein the predicted residuals for a fold are computed using a one-step approximation of model parameters derived from the model trained on the labelled dataset. Jung teaches: The method of claim 22, wherein the predicted residuals for a fold are computed using a one-step approximation of model parameters derived from the model trained on the labelled dataset. ([Page 3, Algorithm 1], [Page 6, Section 2.1, Paragraph 2, Sentences 1-4], [Page 6, Section 2.3, Paragraph 1, Sentences 1-5], [Page 7, Paragraph 2, Sentence 1] Shows that nonlinear model (i.e., “generalized linear model”) is used to approximate the parameters based on the labelled data of the folds. The ‘one-step approximation’ is an approximation method that uses only current ith results. The equation (12) and (13) shows that the method utilizes only current value y_i for the approximation) It would have been obvious to a person having ordinary skill in the art before the effective filing date of the invention to have combined Jung’s teaching of a generalized linear model with parameter estimates with , GIANOLA, Cheng and MUELLER’s teaching of calculating predicted residuals for cross-validation of a generalized linear model, with a reasonable expectation of success, to create a method to calculate predicted residuals to determine the cross-validation error of a model, as in GIANOLA, Cheng and MUELLER, where the parameter estimates of the model can be deducted from the cases, as in Jung. A person of ordinary skill would have been motivated to make this combination to “These values are averaged to produce a final predicted value which may be more accurate compared to the single predicted value by K-fold CV.”. (Page 2, Paragraph 3, Sentence 2; Jung) Response to Arguments Response to Declaration filed on 8/7/2026 The Declaration under 37 CFR 1.132 filed is insufficient to overcome the rejection of claims 1-20 based upon 35 U.S.C. 101 as set forth in the last Office action because: Arguments: The applicant asserts that the claimed invention changes the operation of the computer because: (a) rather than re-fitting the model once per fold, the method fits the model a single time on all of the observations and, from that single fit, computes both the vector of estimated model coefficients and the associated variance-covariance matrix, (b) the variance-covariance matrix from the single fit is then stored and reused for every fold, and (c) using the Sherman-Morrison-Woodbury identity, reducing the size of the matrix that must be inverted for each fold. Examiner’s Response: Examiner respectfully disagrees. First, the examiner notes that the improvement to the mathematical calculation itself is not an improvement in technology. MPEP 2106.05(a) states “It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements.” The applicant did not identified any additional elements in the claim language that support the allegation for a technical improvement. The applicant’s argument in the Declaration, page 3, explains improvements to the mathematical calculation. The features of (a) fitting the model one time rather than k times; (b) reusing a single stored variance-covariance matrix; (c) eliminating the repetitive train-then-test loop; and (d) reducing the size of each matrix inversion, are improvement to the mathematical calculation, which improves the mathematical calculation by reducing the amount of statistical calculation, but are not improvements in technology. The examiner respectfully directs the applicant to MPEP section 2106.05(a), I. IMPROVEMENTS TO COMPUTER FUNCTIONALITY. The cited MPEP section provides examples that the courts have indicated may show an improvement in computer-functionality and may not be sufficient to show an improvement in computer-functionality. The examiner asserts that the claimed method resembles (ii) Accelerating a process of analyzing audit log data when the increased speed comes solely from the capabilities of a general-purpose computer, and (iii) Mere automation of manual processes, because the claimed steps can be performed solely from the capabilities of a general-purpose computer. The examiner reviewed paragraphs 13-16 in pages 3-5 which allegedly provide evidence of the disclosed improvement and concluded that the arguments are not persuasive. The applicant provides the alleged evidence by citing paragraphs [0094]-[0095] for the seeded folds (paragraph 13), [0032] and [0098] for the reuse of single stored variance-covariance matrix (paragraph 14), and Fig. 6B, Fig. 7, [0008], and [0098]-[0099] for evidence of the faster processing (paragraph 15). As discussed above, alleged improvements appear to be improvements to mathematical concepts, not an improvement in technology. ‘Seeded folds so that the reported results are reproducible’, ‘reusing a single stored variance-covariance matrix’, and ‘evidences for faster processing’ are the improvement to the mathematical calculation, not the improvement in technology. Additionally, the applicant’s argument in paragraph 15 “Though we are not merely claiming a faster process, we report these execution-time measurements only as an observable, quantifiable consequence of underlying changes in how the computer performs the cross-validation, and not as the improvement itself” is not persuasive, because as discussed above, the improvement that leads to the faster process is improvement to the mathematical calculation and not improvement in technology. Furthermore, the examiner characterized “a computer-implemented method for selecting a predictive model using K-fold cross-validation. comprising:”, “(a) receiving, by at least one processor, a labeled dataset comprising a plurality of observations and a plurality of p predictor variables arranged as a design matrix X;”, “(b) training, by the at least one processor, a predictive model in a single fit on all n observations of the labeled dataset”, “(c) partitioning, by the at least one processor (d) storing, the variance-covariance matrix X T X - 1 as a computational artifact in a memory accessible to the at least one processor after the single fit”, and “(ii) storing the predicted residuals for the fold;” as the additional elements in the claim language beyond the identified abstract ideas. These additional elements do not reflect a technical improvement, integrate the abstract idea into a practical application, or provide significantly more than abstract ideas of the claim. Accordingly, the claim is not patent eligible. Response to Arguments filed on 08/03/2026 under 35 U.S.C. 101 Arguments: Applicant asserts that (a) the specification sets for the improvement with particularity, not conclusory [Remarks, page 9], (b) the claimed limitations do not merely invoke a computer as a tool to perform cross-validation more quickly and recites a specific technique: deriving each fold’s predicted residuals from a single model fit through an inverse transformation of the ordinary residuals that improves the accuracy of the analysis [Remarks, pages 9-10], (c) the claims reflect the disclosed improvement by showing a single fit on all observations expressly, the inverse-transformation relationship; and the reduced-dimension inversion all leading to the resource reduction [Remarks, page 10], (d) based on rules like McRO, Enfish, and BASCOM (especially Enfish), the computation improvement discussed improves how the computer performs and provides a technical solution to a technical problem [Remarks, pages 10-11]. Examiner’s Response: The examiner respectfully disagrees with the argument (a), (b) and (c). First, in light of the limitation of claim 22 itself, steps (b), (c), (e), (f) and (g) are directed to the abstract idea of mathematical concepts under Step 2A Prong 1 and steps (a) and (d) are directed to additional elements. Spec [0024], [0027] and [0067] explain shortcomings of traditional k-fold cross-validation method but does not explain how the present invention improves the machine learning method. Spec [0031] only explains the conclusion that the present invention accelerates k-fold cross-validation but without any details. Spec [0068] and [0073]-[0074] explain the folded residuals, also referred to as predicted residuals, are deduced from the single fit and are combined to obtain the desired k-fold R-squared statistics. However, claim does not recites the process explained in [0073]-[0074] of deleting residuals by calculating e ( j ) which is calculated by inverting ( I ( j )   -   H ( j ) ) and multiplying the inverted matrix with ( Y j - Y ^ j ) , and then obtaining the R-squared statistics by calculating S S E ( F j ) . The cited paragraphs [0008], [0032], [0087] and [0098]-[0101] merely explain the conclusions that the performance of the model is improved and the improvements can be quantified. The additional elements such as Steps (a), (b), (d) are recited as a tool to perform the K-fold cross-validation because step (a) merely recites receiving a labeled dataset which is directed to mere data gathering, step (b) is generic machine learning model training process, and (d) recites mere data gathering of storing the calculated variance-covariance matrix in a memory. Regarding assertion (d), the examiner notes that the claim fails to recite details about how the precomputed matrix-based representation functions constitute a new kind of data structure, and how it is structured, reusable, and allows access to directly generated predicted residuals for each fold. The independent claim 22 does not seem to recite a new kind of data structure, rather it recites generic K-fold cross validation of a trained predictive model using mathematical calculations. Enfish and the present invention are not similar because Enfish explicitly recites a data storage and retrieval system for a computer memory comprising: a plurality of logical rows including an OID, a plurality of logical columns intersecting the plurality of logical rows to define a plurality of logical cells, and indexing data stored in table. In contrast, the present invention recites performing K-fold cross-validation and storing the calculated variance-covariance matrix in memory. Additionally, the examiner notes that the improvement of “internal restricting of how model validation is performed” is not an improvement to the act of storing; rather, it is an improvement to the data used for storing, the data being a result of various mathematical processes applied to the data, e.g. an improvement to math. Accordingly, arguments regarding claim 1 are not persuasive. Similarly, arguments regarding claims 31 and 35 are not persuasive. Therefore, arguments regarding dependent claims 23, 25-30, 32, 34, 36, and 38-42 are not persuasive. Response to Arguments filed on 08/03/2026 under 35 U.S.C. 103 Arguments: Applicant asserts that (a) The cited disclosure of an error covariance is not the claimed variance-covariance matrix of the estimated coefficients [Remarks, page 12], (b) the claimed Sherman-Morrison-Woodbury p x p reduction, conditioned on p being less than the fold size, is not taught [Remarks, pages 12-13], (c) Muller teaches conventional retrain-per-fold cross validation and does not cure the deficiencies, (d) the stated motivation is conclusory, (e) Claims 27 and 40 further requires the predictive model to be generalized linear model trained by iteratively reweighted least squares, and claim 30 further requires computing the predicted residuals for a fold using a one-step approximation of model parameters derived from the single fit [Remarks, page 13]. Examiner’s response: Regarding (a), the examiner respectfully disagrees with the applicant’s assertion, because the previous version of the limitation “a variance-covariance matrix associated with the estimated model coefficients” does not specify how the matrix is being calculated and allows the broad interpretation of the variance-covariance matrix. The applicant amended the limitation to clarify how the variance-covariance matrix is being calculated. Even though the variance-covariance matrix limitation is narrowed, Gianola still teaches the limitation because it explicitly recites calculating an inverse of X’X in [page 3108, right col, Leave-one-out (LOO) cross-validation, Eqn (4)]. Additionally, a new reference Cheng is introduced to teach the amended Step (d) limitations. The examiner respectfully disagrees with the applicant’s assertion (b). [Gianola, page 3108, right col, Eqn (4) and corresponding paragraphs] and [page 3123, Appendix B] teaches making use of the Sherman-Morrison-Woodbury transformation in Appendix B Eqn (81), the [ X ' X - 1 + X ' X - 1 x i x i ' X ' X - 1 1 - x i ' X ' X - 1 x i ] is an inversion of X ' X - x i x i ' . The matrix before the inversion is the p x p matrix. The final result β - i is formed by deleting ith observations from the model so reduced dimension. Regarding the assertion (c), the examiner notes that Muller is introduced to teach the technical limitation of “selecting the predictive model based on the computed cross-validation performance statistic” and not to teach the limitation of “for each of the K folds, deriving a covariance submatrix corresponding to the fold from the stored variance-covariance matrix without recomputing the variance-covariance matrix and without re-fitting the predictive model.” Regarding (d), the examiner believes that motivation is not conclusory and it is obvious for a person with ordinary skill in the art would introduce the model selection method of Muller to implement the present invention. Selecting and using the model with best performance only improves the prediction accuracy of the trained model. Regarding (e), even though Gianola and Muller do not specifically disclose the limitations of the dependent claims, Valdora and Jung are introduced to remedy the deficiency of Gianola and Muller. Conclusion THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a). A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action. Any inquiry concerning this communication or earlier communications from the examiner should be directed to JUN KWON whose telephone number is (571)272-2072. The examiner can normally be reached Monday – Friday 8:00AM – 4:00PM ET. Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice. If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Abdullah Kawsar can be reached at (571)270-3169. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300. Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000. /JUN KWON/Examiner, Art Unit 2127 /ABDULLAH AL KAWSAR/Supervisory Patent Examiner, Art Unit 2127
Read full office action

Prosecution Timeline

Show 13 earlier events
Aug 19, 2025
Response Filed
Dec 11, 2025
Final Rejection mailed — §101, §103, §112
Jan 26, 2026
Response after Non-Final Action
Jan 26, 2026
Request for Continued Examination
Jan 27, 2026
Response after Non-Final Action
May 01, 2026
Non-Final Rejection mailed — §101, §103, §112
Aug 03, 2026
Response Filed
Oct 23, 2026
Final Rejection mailed — §101, §103, §112 (current)

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12737633
TRAINING A FEDERATED GENERATIVE ADVERSARIAL NETWORK
3y 9m to grant Granted Sep 15, 2026
Patent 12731035
LABEL INFERENCE IN SPLIT LEARNING DEFENSES
3y 8m to grant Granted Sep 08, 2026
Patent 12718063
DEEP LEARNING ARCHITECTURE FOR ADVERSE MEDIA SCREENING
3y 11m to grant Granted Aug 25, 2026
Patent 12711383
ACCURATE ENSEMBLE BY MUTATING NEURAL NETWORK PARAMETERS
7y 3m to grant Granted Aug 18, 2026
Patent 12705504
KNOWLEDGE BASE CONSTRUCTION
8y 5m to grant Granted Aug 11, 2026
Study what changed to get past this examiner. Based on 5 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

7-8
Expected OA Rounds
43%
Grant Probability
83%
With Interview (+40.0%)
4y 9m (~0m remaining)
Median Time to Grant
High
PTA Risk
Based on 82 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month