DETAILED ACTION
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 .
Responsive to communications on 06/01/2026
Claims 1-20 have been amended
Claims 1-20 pending
Claims 1-20 rejected
Final Action
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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.
Response to Arguments
Response to Interview
Examiner confirms and thanks applicant for interview conducted on 03/30/2026. Examiner confirms discussions regarding 112(b), 101, and 103 rejections and agrees with applicants interview summary form.
Response to 35 USC § 112(b) arguments
Claims 1-20 were previously rejected for being indefinite due to the term “the new time series model” not providing clarity to what model was being referred to. Applicant has amended to claims to contain the terms “first”, “second” , and “third” which provides antecedent basis and clarity to the models. Examiner confirms the amendments overcome the previous 112(b) rejection and the examiner withdraws the previous 112(b) rejection of the claims.
Response to 35 USC § 101 arguments
Claims 1-6 and 11-15 were previously rejected as being directed towards an abstract idea.
Issue: Applicant argues that the claims are not directed to an abstract idea. Applicant argues that characterizing the steps of the claim as a mathematic processes oversimplifies the claim limitations and does not consider the claim as a whole. Applicant argues that the “cumulative contribution threshold” as mentioned over specifically an “ordered set of variables” in the amended claim, when combined with the iterative steps of the mechanism, to make the claim directed to a computer-centric optimization process for model construction rather than a mathematic calculation.
Rule: The MPEP section 2106.04(a)(2) gives example of a mathematic calculation “performing a resampled statistical analysis to generate a resampled distribution, “
Analysis: Examiner notes that arguments pertain to amended portions of the claim which require further consideration. Examiner also notes that an iterative process occurring over ordered set of variables does not make the claim separate from a mathematic calculation if the steps recited themselves, are math. For example, the term “aggregate contribution” is a textual replacement for “calculate contribution of the variables.” The examiner believes the mathematic calculations performed in the claimed invention to be similar to the example given in the MPEP, which pertains to generating resampled analysis and distributions of values.
Conclusion: Examiner disagrees that the claimed invention is not directed to an abstract idea. However, the 101 rejections has been withdrawn in light of the amendments as the examiner believes the claimed invention improves upon the functioning of a computer, please see issue 3 where this is discussed.
Issue: Applicant argues that the claims do not recite a mental process. That is because the claim method requires processing potentially large sets of variables, computing contribution metrics, maintain ordered contributions, iterative generation and validation of models, and dynamically modifying thresholds and models. Applicant argues that these steps are data intensive and are performed with potentially high dimensional datasets, which cannot be reasonably performed by an individual in the human mind.
Rule: The MPEP 2111 states “During patent examination, the pending claims must be "given their broadest reasonable interpretation consistent with the specification.”
Analysis: Examiner notes that arguments pertain to amended portions of the claim which require further consideration. The examiner notes that the claimed invention, while understood to apply to models with large sets of variables as intended, is not required in the claim language. The claim language does not differentiate or require a high number of variables. Under broadest reasonable interpretation, the model may only have 2 variables, which can reasonably be performed in the mind of an individual skilled in the art. Furthermore, the examiner disagrees with the implied argument that, because the data is too difficult, that it no longer recites an abstract idea. Under mathematic calculations, the MPEP does not provide a basis for suggesting that the difficulty of the mathematic equation impacts patentability. As for mental processes, the mental processes identified were “excluding variables…” and “if the new time series model is valid…” These steps in the original claim set could be performed by an individual reasonably skilled in the art. Conclusion: Examiner disagrees that the claimed invention is not directed to a mental process. However, the 101 rejections has been withdrawn in light of the amendments as the examiner believes the claimed invention improves upon the functioning of a computer, please see issue 3 where this is discussed.
Issue: Applicant argues that the claims integrate any alleged exception into a practical application. The applicant states that the claim invention applies the alleged judicial exception to improving the functioning of a computer system by reducing computational load of the system, controlling model complexity, minimizing memory usage, and improving efficiency of the model. Applicant argues that the output of the claim is not a number, but a computationally efficient forecasting model with a reduced variable set that directly impacts the performance of the system.
Rule: The USPTO December 5, 2025 Memorandum subject “Advance notice of change to the MPEP in light of Ex Parte Desjardins” states in pages 2-3 “See, e.g., Ex Parte Desjardins, Appeal No. 2024-000567 (PTAB September 26, 2025, Appeals Review Panel Decision) (precedential), in which the specification identified the improvement to machine learning technology by explaining how the machine learning model is trained to learn new tasks while protecting knowledge about 2 previous tasks to overcome the problem of “catastrophic forgetting,” and that the claims reflected the improvement identified in the specification. Indeed, enumerated improvements identified in the Desjardins specification included disclosures of the effective learning of new tasks in succession in connection with specifically protecting knowledge concerning previously accomplished tasks; allowing the system to reduce use of storage capacity; and the enablement of reduced complexity in the system. Such improvements were tantamount to how the machine learning model itself would function in operation and therefore not subsumed in the identified mathematical calculation.”
Analysis: The specification identifies an improvement in technology. See par 31: “Validation of the new model ensures its continued accuracy and therefore justifies reliance upon the new model to accurately forecast behavior while consuming fewer system resources (e.g., processing, memory, and/or bandwidth).” The claims have been amended to reflect the improvement to technology. The claims now recite “consuming fewer system resources.” The Ex Parte Desjardins decision referenced above pertained to the training of a machine learning model, which is then improved to reduce complexity in the system. This invention is similarly directed to improving a machine learning model to reduce complexity in a system, which is the goal of the application and the improvement is reflected in the specification and the claims.
Conclusion: In light of the amendments and improvement to technology, examiner withdraws previous 101 rejection.
Issue: Applicant argues that the claims recite significantly more than any alleged exception. Applicant argues that the recited steps in the claim provide a non-conventional and non-generic combination of steps that is not merely routine or well understood. Specifically, the applicant argues that the feedback controlled optimization is a specific processing step that alters how a computer operates to build a forecasting model, rather than a usage of a computer in its ordinary capacity.
Rule: The MPEP 2106.05 states : An inventive concept "cannot be furnished by the unpatentable law of nature (or natural phenomenon or abstract idea) itself." Genetic Techs. Ltd. v. Merial LLC, 818 F.3d 1369, 1376, 118 USPQ2d 1541, 1546 (Fed. Cir. 2016). See also Alice Corp., 573 U.S. at 21-18, 110 USPQ2d at 1981 (citing Mayo, 566 U.S. at 78, 101 USPQ2d at 1968 (after determining that a claim is directed to a judicial exception, "we then ask, ‘[w]hat else is there in the claims before us?") (emphasis added)); RecogniCorp, LLC v. Nintendo Co., 855 F.3d 1322, 1327, 122 USPQ2d 1377 (Fed. Cir. 2017) ("Adding one abstract idea (math) to another abstract idea (encoding and decoding) does not render the claim non-abstract"). Instead, an "inventive concept" is furnished by an element or combination of elements that is recited in the claim in addition to (beyond) the judicial exception, and is sufficient to ensure that the claim as a whole amounts to significantly more than the judicial exception itself. Alice Corp., 573 U.S. at 27-18, 110 USPQ2d at 1981 (citing Mayo, 566 U.S. at 72-73, 101 USPQ2d at 1966). “
Analysis: In order for the claim elements to provide a non-conventional and non-generic combination of steps that is not merely routine or well understood, the claimed steps would be required to be additional elements rather than parts of the judicial exception. As outlined in the previous rejection, the claimed steps were pertaining to mathematic calculations and mental processes, and therefore, are not considered additional elements which apply the judicial exception.
Conclusion: Examiner disagrees that the claimed invention amounts to significantly more under section 2B by being non-routine activity. However, the 101 rejections has been withdrawn in light of the amendments as the examiner believes the claimed invention improves upon the functioning of a computer, please see issue 3 where this is discussed.
Response to 35 USC § 103 arguments
Claims 1-20 were previously rejected under 103 combination references.
Issue: Applicant argues the cited references do not teach the “Cumulative Contribution Threshold” over the ordered set. Applicant states that amended claim requires ordering variables based on contribution, computing an aggregated (cumulative) contribution across that ordered set, and applying a cumulative contribution threshold that defines a boundary such that variables falling below the boundary (their collective contribution below 5% threshold for example) are excluded from the subsequent model. Applicants have amended the claim to define the cumulative contribution threshold and provides more specific arguments for why each limitation is not covered by the prior art.
Issue: Applicant argues that Ishiguro does not teach or suggest a cumulative aggregation over an ordered set, but instead on importance of individual features. Applicants argue that Ishiguro does not run a cumulative contribution across the ordered list of variables, define a global boundary, or exclude variables based on the cumulative contribution boundary threshold condition. Applicant states that Ishiguro discusses excluding a feature based on whether it improves model evaluation metric, not how it contributes to a collective cumulative contribution profile.
Rule: The MPEP 2111 states “During patent examination, the pending claims must be "given their broadest reasonable interpretation consistent with the specification.”
Analysis: The terms recited in the claim are interpreted with broadest reasonable interpretation consistent with the specification. Ishiguro clearly illustrates an ordered set see abstract “ (S106) of updating the order of the feature amounts” where the cumulative contribution is measured (see figure 2, where the AIC is cumulative of all the features in the subset), where the global boundary is the threshold where increasing the feature does not improve the AIC (see again figure 2). As understood by one ordinarily skilled in the art, a cumulative contribution profile reads on a model evaluation metric, where the cumulative contribution profile is a profile which evaluates how well the features represent the model. Ishiguro does the same thing, where the features in a subset are measured by the AIC to see how well they map to the model. Furthermore, Ishiguro removes multiple features at once, see figure 3, this is not necessarily purely an individual feature classification, this is measuring through multiple features combined.
Conclusion: As the scope of the claim has been amended, new art may be introduced to reject the claim. While the examiner maintains the rejection with Ishiguro, believing that Ishiguro still teaches the above limitations as recited above, In order to advance compact prosecution, the examiner has also added new prior art Koehrsen_2018 which maps closer to how is outlined in the specifications.
Issue: Applicant argues that Chowdhury does not teach or suggest cumulative contribution-based boundary selection. Applicant argues that Chowdhury teaches adding or removing variables in a stepwise manner, not constructing ordered set of variables with a cumulative contribution across them, with a global cutoff point.;
Rule: The MPEP 2111 states “During patent examination, the pending claims must be "given their broadest reasonable interpretation consistent with the specification.”
Analysis: Examiner agrees that Chowdhury does not discuss cumulative contribution boundary selection. Chowdhury was a reference used for suggesting adding back features after they have been removed.
Conclusion: As the scope of the claim has been amended, new art may be introduced to reject the claim.
Issue: Applicant argues that the cumulative contribution threshold is a distinct selection method than the cited references, in that it treats variables are an ordered contribution distribution, not isolated candidates. It uses cumulative aggregation and global single boundary to prune candidates.
Rule: The MPEP 2111 states “During patent examination, the pending claims must be "given their broadest reasonable interpretation consistent with the specification.”
Analysis: As stated above in response A, the examiner believes the scoring criteria of Ishiguro represent a cumulative aggregation. See figure 3, where the AIC gives a score based on the cumulative number of features in the subset, rather than giving a score for each feature separately.
Conclusion: As the scope of the claim has been amended, new art may be introduced to reject the claim. While the examiner maintains the rejection with Ishiguro, believing that Ishiguro still teaches the above limitations as recited above, In order to advance compact prosecution, the examiner has also added new prior art Koehrsen_2018 which maps closer to how is outlined in the specifications.
Issue: applicant further reiterates that the combination of Ishiguro and Chowdhury does not remedy the deficiencies outlined in a-c.
Conclusion: Examiner has outlined a-c above. As the scope of the claim has been amended, new art may be introduced to reject the claim. While the examiner maintains the rejection with Ishiguro, believing that Ishiguro still teaches the above limitations as recited above, In order to advance compact prosecution, the examiner has also added new prior art Koehrsen_2018 which maps closer to how is outlined in the specifications.
Issue: applicant further reiterates that the combination of Ishiguro and Chowdhury do not teach the limitations of “cumulative contribution threshold over an ordered set of variables” and “aggregated contribution relative to a boundary defined by the threshold”
Conclusion: Examiner has outlined a-c above. As the scope of the claim has been amended, new art may be introduced to reject the claim. While the examiner maintains the rejection with Ishiguro, believing that Ishiguro still teaches the above limitations as recited above, In order to advance compact prosecution, the examiner has also added new prior art Koehrsen_2018 which maps closer to how is outlined in the specifications.
Issue: Applicant argues that the claimed adaptive threshold adjustment based on model validity is not taught or suggested. Applicant argues that the claim recites iteratively generating models and lowering the cumulative contribution threshold when the model fails validation to include additional variables. Applicant argues that the claimed inventions of Ishiguro and Chowdhury do not include the cumulative threshold, and do not adjust that threshold.
Rule: The MPEP 2144.01 states “"[I]n considering the disclosure of a reference, it is proper to take into account not only specific teachings of the reference but also the inferences which one skilled in the art would reasonably be expected to draw therefrom."
Analysis: The previous claim set did not define a meaning for cumulative contribution threshold. The claim was previously rejected under 103 analysis where it was determined that it would be obvious to adjust the threshold by adding features, where the number of features under the AIC where the model is improved is the threshold. The examiner believes that this would be a conclusion one ordinarily skilled in the art would reasonably expect to draw, that if the AIC is based on a number of features, that adding a feature would be tantamount to increasing the AIC threshold. Since the claim has been amended to clarify the contribution threshold more specifically new prior art may be introduced. Specifically, the prior art of Huu teaches a threshold removal of features from a model, with a re-addition of features if they are deemed important. Par 60: “On a first retrain, in some example embodiments, the current subset of features is the entirety of the set of available features. On later retrains, as features are removed from or restored to the current subset, the features in the current subset of features changes.” One reasonably skilled in the art would understand that when following a “widely used” machine learning principle such as stepwise selection as cited by Chowdhury, that if the features were removed based on a threshold as outlined in Huu, that the features would also be added back based on that same threshold.
Conclusion: As the scope of the claim has been amended, new art may be introduced to reject the claim.
Issue: Applicant argues the references fail to teach the “consuming fewer system resources limitation” While both references discuss reducing variables to improve model performance, they do not disclose a mechanism which balances predictive performance against computational resource consumption and a process which converges on that balance using adaptive threshold control.
Rule: The MPEP 2145 (II) states “Mere recognition of latent properties in the prior art does not render nonobvious an otherwise known invention. In re Wiseman, 596 F.2d 1019, 201 USPQ 658 (CCPA 1979) (Claims were directed to grooved carbon disc brakes wherein the grooves were provided to vent steam or vapor during a braking action. A prior art reference taught noncarbon disc brakes which were grooved for the purpose of cooling the faces of the braking members and eliminating dust. The court held the prior art references when combined would overcome the problems of dust and overheating solved by the prior art and would inherently overcome the steam or vapor cause of the problem relied upon for patentability by applicants. Granting a patent on the discovery of an unknown but inherent function (here venting steam or vapor) "would remove from the public that which is in the public domain by virtue of its inclusion in, or obviousness from, the prior art."
Analysis: The examiner believes that the limitation “consuming fewer system resources” is an inherent benefit in the prior art of feature extraction. Furthermore, since the scope of the claim has been amended, new art may be introduced to reject the claim. The prior art of Huu explicitly states that the feature reduction workflow is done to consume fewer system resources, See par 18: “Without the unimportant features as input, computation and storage resources consumed by the modeling layer are reduced.” As the other references also remove features as inputs, they also will inherently contain the benefit as outlined in Huu.
Conclusion: As the scope of the claim has been amended, new art may be introduced to reject the claim.
Issue: Applicant argues against the motivation to combine Ishiguro and Chowdhury. Applicant states that both Ishiguro and Chowdhury approaches rely on individual variable evaluation, not cumulative aggregation. Applicant does not specific reasons against the combination itself, applicant argues against the content of the references.
Rule: The MPEP 2145 (IV) states “One cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references.”
Analysis: As the scope of the claim has been amended, new art may be introduced to reject the claim. As outlined above, the examiner believes the prior art of Ishiguro to teach and make obvious the recited steps of the claim. Please see the arguments in section 1 as well as the rejection below which addresses the amended limitations. As the scope of “cumulative contribution threshold” has been changed, new prior art may be introduced.
Conclusion: As the scope of the claim has been amended, new art may be introduced to reject the claim.
Issue: Applicant argues claimed combination produces a non-obvious technical effect not cited by prior art
Rule: The MPEP 2145 (VI) states “Although the claims are interpreted in light of the specification, limitations from the specification are not read into the claims.”
Analysis: The non-obvious technical effects cited by examiner describe limitations which are not claimed as understood under BRI.
Conclusion: As the scope of the claim has been amended, new art may be introduced to reject the claim. In order to advance compact prosecution, the examiner has also added new prior art Koehrsen_2018 which maps closer to how is outlined in the specifications, which the examiner believes to also recite the non-obvious technical effects of the claim.
End Response to Arguments
Claim Rejections - 35 USC § 103
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.
Claim 1 is rejected under 35 U.S.C. 103 as being unpatentable over Ishiguro_2021(US 20210027108 A1), Koehrsen_2018 (“A Feature Selection Tool for Machine Learning in Python”), and Chowdhury_2020 (“in view of “Variable selection strategies and its importance in clinical prediction modelling”) in further view of Huu_2021 (US 20210350273 A1).
Claim 1:Ishiguro_2021 makes obvious A computer implemented method (par 1: “The present invention relates to a data processing apparatus, a data processing method,”) of selecting influencer variables (abstract: “the feature amount selection unit performs feature amount selection including:”) for a time series forecasting model, the method comprising: (par 3: “ time-series”)
receiving an original time series forecasting model configured to predict a future of a target variable based on known data at a time of forecasting (par 26: “In the present embodiment, more specifically, a chemical mechanical polishing (CMP) apparatus, which is one of the semiconductor manufacturing apparatuses, will be described as the manufacturing apparatus. Here, the data serving as the explanatory variable is a monitor value of a processing condition during the processing, such as a rotation speed of a wafer and a slurry amount, and data related to a state of the apparatus itself, such as a time of using a polishing pad. On the other hand, the data serving as the objective variable is a polishing amount (removal rate: RR) of the wafer by processing in the CMP apparatus. A model formula to be created is a regression formula for predicting the RR based on the above apparatus data.”) Examiner note: Where the regression formula is the original time series model. ) see also par 3: “An average value, a standard deviation value, or the like of the time-series signal is one of the feature amounts.” Where the model takes in time-series signal data. )and an original set of variables; (par 27: “The apparatus data recorded in the recording unit is computed by the computing unit, and is subjected to pre-analysis processing such as elimination of apparently abnormal data and extraction of the feature amounts.”) Examiner note: Where the extraction of feature amounts from the time series variable is the original set of variables. See also par 3: “In normal data processing, a time-series signal obtained from measurement performed by a sensor is not used as it is, and a feature amount that well represents a feature of the signal or a value referred to as a feature is often used”
calculating contributions of the original set of variables to the original time series forecasting model; (See fig . 1 step S101 “create feature amount ranking”. Par 29: “ In the flowchart of FIG. 1, in the first step in the feature amount selection unit, firstly, significance of the feature amount, (Examiner note: a measure of contributions) that is, a feature amount ranking is created by using Fisher criterion which is one of filter methods (S101).” Examiner note: Where from the flowchart in Fig. 1 it can be seen that this step occurs before any iterations, and is therefore understood to be in respect to the original variables and model.
excluding variables falling below a cumulative contribution threshold (See fig. 1 step S105 “Search optimal value of AIC (o(i)) of evolution index and delete feature amount based on optimal value” … Par 33: “FIG. 2 shows an example of a graph in which numbers of the respective subsets are plotted on a horizontal axis and the corresponding AICs are plotted on a vertical axis. In the fourth step of the present embodiment, unnecessary feature amounts are determined and deleted based on the AIC calculated in the third step (S105). (Examiner note: a measure of contribution) In the graph 201 of FIG. 2, the AIC has a smallest AIC at a No. 84 subset. This indicates that feature amounts whose ranking is from the first to the 84th contribute to improving the prediction performance, but the 85th and subsequent feature amounts do not contribute to improving the prediction performance. Therefore, in the fourth step, the feature amount corresponding to a ranking of the subset number following the subset number with the smallest AIC is deleted. That is, in the present embodiment, the feature amounts in the 85th and higher rankings are deleted.” Examiner note: a exclusion of variables 85 – 100 falling below the minimal contribution threshold.) , wherein the cumulative contribution threshold defines a boundary over an ordered set of variables, wherein variables whose aggregated contribution up to the boundary does not exceed the cumulative contribution threshold are excluded from a subsequent model (Par 33: “FIG. 2 shows an example of a graph in which numbers of the respective subsets (examiner note: ordered set of variables) are plotted on a horizontal axis and the corresponding AICs are plotted on a vertical axis. In the fourth step of the present embodiment, unnecessary feature amounts are determined and deleted based on the AIC calculated in the third step (S105). (Examiner note: a measure of contribution) In the graph 201 of FIG. 2, the AIC has a smallest AIC at a No. 84 subset. This indicates that feature amounts whose ranking is from the first to the 84th contribute to improving the prediction performance, but the 85th and subsequent feature amounts do not contribute to improving the prediction performance. (Examiner note: Where the threshold/boundary is where the model stops improving. See fig 2 boundary line) Therefore, in the fourth step, the feature amount corresponding to a ranking of the subset number following the subset number with the smallest AIC is deleted. That is, in the present embodiment, the feature amounts in the 85th and higher rankings are deleted.” Examiner note: a exclusion of variables 85 – 100 falling below the minimal contribution threshold.) ,
Examiner note: Please see arguments towards 103, where the examiner outlines in argument 1 why Ishiguro still reads on the amended claim limitation. In order to advance compact prosecution, the examiner will also reject the above claim limitation with a new reference Koehrsen_2018, which maps clearly to how the examiner understands the limitation based on the specifications.
creating a first new time series forecasting model from remaining variables; par 32: “ In the third step in the feature amount selection unit, for all the subsets created in the second step, an evaluation index, which is a value serving as an index for evaluating prediction performance in a regression or classification problem, that is, a model formula evaluation index, is calculated and created (S104). The present embodiment involves a regression problem of estimating the RR, and the above-described AIC is adopted as an index for evaluating the prediction performance. The third step is to calculate the respective AIC for all subsets.” Examiner note: See Fig 1 steps S104 – S107. Where S104 “create model formula evaluation index for each subset” implies that each subset (remaining variables after each iteration) is used to create a new model which is being evaluated in each iteration. Where while the deleting step S105 occurs after S104, it is understood that each subset is necessary excluding variables from the original set of variables, as well as that this process is occurring literately and therefore repeats.
storing the first new time series forecasting model in a non-transitory computer readable storage medium; (par 24: “The data processing apparatus and the processing apparatus method according to the first embodiment will be described with reference to FIGS. 1 to 5. The data processing apparatus of the present embodiment includes, although not shown, a recording unit that records electronic data and a computing unit that computes the recorded electronic data. Since such a data processing apparatus can be implemented by a general computer that is represented by a personal computer (PC) and includes a central processing unit (CPU) for computing the electronic data, a storage unit for storing the electronic data and various processing programs, an input/output unit including a keyboard and a display, and a communication interface, the data processing apparatus is not shown.”
Par 39: “Using the subset including the selected feature amount in the above procedure, the computing unit creates a regression model for estimating the RR. Information on the regression model and the subset including the feature amount is stored in the recording unit. As described above, a step of acquiring both the apparatus data and the RR data for creating a model for a desired period or a desired amount and selecting the feature amount to create the model is generally referred to as a training step.”) Examiner note: Where although not expressed in the workflow of Fig.1, it is understood and obvious to one ordinarily skilled in the art that the regression model is stored. Where the presence of a “storage unit” makes obvious non-transitory computer readable storage medium
if the first new time series forecasting model is valid based upon a performance horizon, ( (Par 38: “ The iteration from the second step to the fifth step is performed until a minimum value of an AIC obtained in the third step in an m-th iteration is equal to a minimum value of an AIC obtained in the third step in an (m−1)th iteration. After the iteration is ended, a subset having a lowest AIC becomes a subset including the selected feature amount.” (Examiner note: Where a determination of an AIC is a check of the validity of the model based on an AIC performance horizon.) iterating to further reduce a number of variables and generate a second new time series forecasting model; (par 37: After the fifth step is ended, in the present embodiment, the second to fifth steps are iterated (S102). FIG. 3 shows graphs of AIC values in respective subsets obtained in the third step of respective iterations by iteration. It can be seen that the feature amounts selected by the iteration decrease, and a minimum value of the AIC also decreases, that is, the prediction performance improves.) Examiner note: Where as stated step S104 in the iteration implies a model being generated.
if the second new time series forecasting model is not valid based upon the performance horizon, ( (Par 38: “ The iteration from the second step to the fifth step is performed until a minimum value of an AIC obtained in the third step in an m-th iteration is equal to a minimum value of an AIC obtained in the third step in an (m−1)th iteration. After the iteration is ended, a subset having a lowest AIC becomes a subset including the selected feature amount.” (Examiner note: Where a determination of an AIC is a check of the validity of the model based on an AIC performance horizon.) lowering the cumulative contribution threshold a third new time series forecasting model; and (par 37: After the fifth step is ended, in the present embodiment, the second to fifth steps are iterated (S102). FIG. 3 shows graphs of AIC values in respective subsets obtained in the third step of respective iterations by iteration. It can be seen that the feature amounts selected by the iteration decrease, and a minimum value of the AIC also decreases, that is, the prediction performance improves.) Examiner note: Whereas stated step S104 in the iteration implies a model being generated. Please see also Fig. 2. Which depicts the AIC of 50 subsets to be too low, and increasing the threshold number of samples to 84.
outputting a selected set of influencer variables (Par 39: “Using the subset including the selected feature amount in the above procedure, the computing unit creates a regression model for estimating the RR. Information on the regression model and the subset including the feature amount is stored in the recording unit.”) for the third new time series forecasting model that accurately forecasts behavior (Par 39-40: ”As described above, a step of acquiring both the apparatus data and the RR data for creating a model for a desired period or a desired amount and selecting the feature amount to create the model is generally referred to as a training step. On the other hand, an operation step of acquiring only the apparatus data and predicting the RR based on the data is referred to as a testing step. ”) Examiner note: The feature amount is used for a another new model in the testing step.
Ishiguro_2021 does not expressly recite
to exclude fewer of the original set of variables
while consuming fewer system resources.
Koehrsen_2018 further makes obvious wherein the cumulative contribution threshold defines a boundary over an ordered set of variables, wherein variables whose aggregated contribution up to the boundary does not exceed the cumulative contribution threshold are excluded from a subsequent model;
(see sections “Zero Importance Features” and “Low importance features”:
PNG
media_image1.png
591
1467
media_image1.png
Greyscale
“On the right we have the cumulative importance versus the number of features. The vertical line is drawn at threshold of the cumulative importance, in this case 99%.” … see section “Removing Features; “Once we’ve identified the features to discard, we have two options for removing them. All of the features to remove are stored in the ops dict of the FeatureSelector and we can use the lists to remove features manually. Another option is to use the remove built-in function.” (Examiner note: Where this removes features identified with zero and low importance thresholds)”
Ishiguro_2021 and Koehrsen_2018 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning and are both pertinent to the same problem called feature extraction for machine learning.
Before the effective filing date, it would have been obvious to a person ordinarily skilled in the art to combine Ishiguro_2021 and Koehrsen_2018.
The rational for doing so would have been a simple substitution of one known element for another to obtain predictable results. The prior art of Ishiguro_2021 and Koehrsen_2018 both rank features for feature extraction to improve a machine learning model. The prior art of Ishiguro_2021 using a contribution based on AIC, but implies that any other evaluation index may be used. Par 11 states: “As mentioned above, the AIC may be used for this evaluation index. However, the AIC is effective as an index for securing the generalization performance of the model and preventing the over-learning, but there are cases where the over-learning occurs even when optimization is performed by evaluation using the AIC. “Koehrsen_2018 supplies an alternative evaluation index, being a cumulative contribution threshold. Koehrsen_2018 also implies that this issue of finding low ranked features is well known in the art, see par 2: “Frustrated by the ad-hoc feature selection methods I found myself applying over and over again for machine learning problems, I built a class for feature selection in Python available on GitHub. The Feature Selector includes some of the most common feature selection methods:
Features with a high percentage of missing values
Collinear (highly correlated) features
Features with zero importance in a tree-based model
Features with low importance
Features with a single unique value” Where Koehrsen_2018 demonstrates that determining zero/low importance features is a known issue in the art where Koerhsen_2018 identifies a method of performing the features.
Therefore, it would have been obvious to substitute the feature extraction workflow using AIC of Ishiguro_2021 with the usage of cumulative contribution thresholds by Koerhsen_2018 to solve for the common issue of determining low impact features using interchangeable known methods in the art to obtain the invention as specified in the claims.
Koehrsen_2018 does not expressly recite exclude fewer of the original set of variables
while consuming fewer system resources.
Chowdhury_2019 in further view of Huu_2021 makes obvious exclude fewer of the original set of variables (page 4 col 2 par 3: “Stepwise selection methods are a widely used variable selection technique, particularly in medical applications. This method is a combination of forward and backward selection procedures that allows moving in both directions, adding and removing variables at different steps. … if stepwise selection starts with backward elimination, the variables are deleted from the full model based on statistical significance and then added back if they later appear significant. ”
Ishiguro_2021 and Chowdhury_2019 are analogous art to the claimed invention because they are from the same field of endeavor called variable selection. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021 and Chowdhury_2019. The rationale for doing so would have been to follow a teaching and motivation proposed by Chowdhury_2019.
Ishiguro_2021 performs feature selection by creating subsets, and then deleting feature amounts that do not improve those subsets. This is similar to the method of backward elimination as taught by Chowdhury_2019 where page 3 col 2 par 3: “This method starts with a full model that considers all of the variables to be included in the model. Variables then are deleted one by one from the full model until all remaining variables are considered to have some significant contribution to the outcome.” Where backward elimination is being done to each subset. One benefit of using stepwise selection, as stated by Chowdhury_2019 is page 5 col 1 par 3: “This method (examiner note: stepwise) allows researchers to examine models with different combinations of variables that otherwise may be over looked.6 The method is also comparatively objective as the same variables are generally selected from the same data set even though different persons are conducting the analysis. This helps reproduce the results and validate in model.” Also, as stated by Chowdhury_2019page 5 col 1 par 3: “The stepwise selection method is perhaps the most widely used method of variable selection.”
Therefore, it would have been obvious to combine the feature elimination and subset workflow of Ishiguro_2021 with the method of stepwise selection of Chowdhury_2019 for the benefit of having an objective model and allowing researches to examine different combinations to obtain the invention as specified in the claims. One reasonably skilled in the art would know to use the concepts of stepwise selection, as they are “widely used.”
While Chowdhury_2019 makes obvious excluding fewer of the original set of variables, Chowdhury_2019 does not expressly recite that this is done through [a threshold]
while consuming fewer system resources.
Huu_2021 however makes obvious [using a threshold to exclude fewer of the original set of variables] (par 51: “ In operation 630, the feature reduction module 260 determines, based on the importance measures for each feature, whether to remove the feature from the set of features. For example, a predetermined number or percentage of least important features may be removed, all or a predetermined number or percentage of features with an importance measure below a predetermined threshold may be removed, all or a predetermined number of percentage of features considered “unimportant” in operation 620 may be removed, or any suitable combination thereof. After one or more features are removed in operation 630, the method 600 returns to operation 610. In the next retrain, another set of ML models is generated using the modified set of features. In some example embodiments, the removed features are returned back to the set of features after a predetermined number of retrains (e.g., one retrain, two retrains, or four retrains) in order to re-evaluate their respective effective importance on most recent training dataset. Thus, the number of ML models in the set of ML models may be one or a greater than one.”
while consuming fewer system resources. (par 53: “The models that are generated with the reduced set of features are “leaner” than the original models because they make use of fewer features. Operations of accessing or generating the removed features are saved. Compared with methods that simply repeatedly generate the models without performing operations 620 and 630, this approach saves substantial network, computational, and memory resources, as well as time.”) Examiner note: Where this implies that models with fewer features (such as the other references of Ishiguro_2021, Koehrsen_2018, Chowdhury_2019) all inherently save memory resources.
Ishiguro_2021, Koehrsen_2018, Chowdhury_2019, and Huu_2021 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning and specifically feature extraction. Before the effective filing date, it would have been obvious to a person ordinarily skilled in the art to combine Ishiguro_2021, Koehrsen_2018, Chowdhury_2019, and Huu_2021. The rational for doing so would have been to follow a teaching and motivation proposed in the prior art.
Ishiguro_2021 provides an iterative method of model improvement using feature ranking based on an AIC criteria. Koehrsen_2018 provides an alternative method of ranking features. Chowdhury_2019 makes obvious and routine the fact that features are re-added into models as a widely used practice. However, Chowdhury_2019 does not reflect this addition in the context of a threshold. Huu_2021 however makes obvious re-adding variables based on their measured importance when they were removed based on a threshold. One reasonably skilled in the art would understand that when following a “widely used” machine learning principle such as stepwise selection as cited by Chowdhury, that if the features were removed based on a threshold as outlined in Huu, that the features would also be added back based on that same threshold. This provides a benefit as proposed by Huu_2021 that par 64: “Compared with the use of the methods 600 and 700 without restoring features after a delay, the method 800 obtains better results over time by re-testing the features and detecting if the removed features regain predictive power.”
Therefore it would have been obvious to combine the feature extraction methods of Ishiguro_2021 and Koehrsen_2018, with the re-addition of features by lowering a contribution of Chowdhury_2019 and Huu_2021 for the benefit of obtaining better results over time through re-testing of features to obtain the invention as specified in the claims.
Claims 2, 4-6, and 11-14 are rejected under 35 U.S.C. 103 as being unpatentable over Ishiguro_2021, Koehrsen_2018, Chowdhury_2020, Huu_2021, and further in view of US Sarwat_2021 (20210124089 A1)
Claim 2:
Ishiguro_2021 makes obvious The computer implemented method of claim 1 wherein the original time series forecasting model See claim 1)
Sarwat_2021, however, makes obvious is elastic-net linear regression Sarwat_2021 par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection. In datasets with high correlation between the predictors, LASSO's variable selection performs poorly. Also, for datasets where the dimensionality, p, is much less when compared to the number of observations n, ridge regularization outperforms LASSO. An elastic net is a combination of ridge and LASSO regularization techniques that applies an elastic net penalty. Given the tuning parameter λ that controls the penalty's magnitude, the model solves the objective function, F(⋅) defined in Equation (2.1) over its entire grid space. Let f(y, η) denote the negative log-likelihood function for the i.sup.th record. If the response is of type Gaussian, then f(y, η)=.sup.(y-η).sup.2. The variable α controls the elastic net penalty, with α=0 denoting ridge, α=1 denoting LASSO, and α∈(0, 1) denoting elastic net.
Ishiguro_2021, and Sarwat_2021 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, and Sarwat_2021. The rationale for doing so would have been to follow a teaching and motivation proposed in the prior art.
Sarwat_2021 teaches that 2021 par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection. In datasets with high correlation between the predictors, LASSO's variable selection performs poorly. Also, for datasets where the dimensionality, p, is much less when compared to the number of observations n, ridge regularization outperforms LASSO. An elastic net is a combination of ridge and LASSO regularization techniques that applies an elastic net penalty
Ishiguro_2021 wants to also decrease overfitting, where the focus of Ishiguro_2021 is to reduce it. See par 11: “As mentioned above, the AIC may be used for this evaluation index. However, the AIC is effective as an index for securing the generalization performance of the model and preventing the over-learning, but there are cases where the over-learning occurs even when optimization is performed by evaluation using the AIC.”
Therefore, it would have been obvious to combine the feature extraction workflow of Ishiguro_2021 and with an elastic-net model of Sarwat_2021 for the benefit of reducing overfitting to have a more accurate model to obtain the invention as specified in the claims.
Claim 4:
Ishiguro_2021 makes obvious The computer implemented method of claim see claim 1)
Sarwat_2021, however, makes obvious further comprising subjecting variables to regularization (par 35: “ To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection. )
Ishiguro_2021, and Sarwat_2021 are analogous art to the claimed invention because they are from the same field of endeavor called regression and time series data forecasting. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021 and Sarwat_2021. The rationale for doing so would have been to follow a teaching and motivation proposed in the prior art.
Sarwat_2021 teaches that 2021 par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection.”
Ishiguro_2021 wants to also decrease overfitting, where the focus of Ishiguro_2021 is to reduce it. See par 11: “As mentioned above, the AIC may be used for this evaluation index. However, the AIC is effective as an index for securing the generalization performance of the model and preventing the over-learning, but there are cases where the over-learning occurs even when optimization is performed by evaluation using the AIC.”
Therefore, it would have been obvious to combine the feature extraction workflow of Ishiguro_2021 with the regularization of variables of Sarwat_2021 for the benefit of reducing overfitting to have a more accurate model to obtain the invention as specified in the claims.
Claim 5:
The computer implemented method of claim 4
Ishiguro_2021 does not expressly recite wherein the regularization is lasso.
Sarwat_2021 however, makes obvious wherein the regularization is lasso. (par 35: “ To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection)
As stated in claim 4, it would have been obvious to combine the feature extraction workflow of Ishiguro_2021 with the lasso regularization of variables of Sarwat_2021 for the same benefit of reducing overfitting to have a more accurate model to obtain the invention as specified in the claims.
Claim 6:
The computer implemented method of claim 4
Ishiguro_2021 does not expressly recite wherein the regularization is ridge.
Sarwat_2021 makes obvious wherein the regularization is ridge. (par 35: “ To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection)
As stated in claim 4, it would have been obvious to combine the feature extraction workflow of Ishiguro_2021 with the ridge regularization of variables of Sarwat_2021 for the same benefit of reducing overfitting to have a more accurate model to obtain the invention as specified in the claims.
Claim 11:The limitations of claim 11 are substantially the same as those of claim 4 and are therefore rejected due to the same reasons as outlined above for claim 4. Additionally, Ishiguro_2021 makes obvious the additional limitations of A non-transitory computer readable storage medium embodying a computer program for performing a method (par 24: “The data processing apparatus and the processing apparatus method according to the first embodiment will be described with reference to FIGS. 1 to 5. The data processing apparatus of the present embodiment includes, although not shown, a recording unit that records electronic data and a computing unit that computes the recorded electronic data. Since such a data processing apparatus can be implemented by a general computer that is represented by a personal computer (PC) and includes a central processing unit (CPU) for computing the electronic data, a storage unit for storing the electronic data and various processing programs, an input/output unit including a keyboard and a display, and a communication interface, the data processing apparatus is not shown.”)
Claim 12:
The limitations of claim 12 are substantially the same as those of claim 5 except that it depends from claim 11 and is therefore rejected due to the same reasons as outlined above for claims 5 and 11.
Claim 13:The limitations of claim 13 are substantially the same as those of claim 6 except that it depends from claim 11 and is therefore rejected due to the same reasons as outlined above for claims 6 and 11.
Claim 14:
Ishiguro_2021 and Chowdhury_2019 make obvious The non-transitory computer readable storage medium of claim 11 wherein the time series forecasting model (see claim 11)
Ishiguro_2021 does not expressly recite is elastic-net linear regression
Sarwat_2021, however, makes obvious is elastic-net linear regression Sarwat_2021 par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection. In datasets with high correlation between the predictors, LASSO's variable selection performs poorly. Also, for datasets where the dimensionality, p, is much less when compared to the number of observations n, ridge regularization outperforms LASSO. An elastic net is a combination of ridge and LASSO regularization techniques that applies an elastic net penalty. Given the tuning parameter λ that controls the penalty's magnitude, the model solves the objective function, F(⋅) defined in Equation (2.1) over its entire grid space. Let f(y, η) denote the negative log-likelihood function for the i.sup.th record. If the response is of type Gaussian, then f(y, η)=.sup.(y-η).sup.2. The variable α controls the elastic net penalty, with α=0 denoting ridge, α=1 denoting LASSO, and α∈(0, 1) denoting elastic net.
Ishiguro_2021, and Sarwat_2021 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, and Sarwat_2021. The rationale for doing so would have been to follow a teaching and motivation proposed in the prior art.
Sarwat_2021 teaches that 2021 par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection. In datasets with high correlation between the predictors, LASSO's variable selection performs poorly. Also, for datasets where the dimensionality, p, is much less when compared to the number of observations n, ridge regularization outperforms LASSO. An elastic net is a combination of ridge and LASSO regularization techniques that applies an elastic net penalty
Ishiguro_2021 wants to also decrease overfitting, where the focus of Ishiguro_2021 is to reduce it. See par 11: “As mentioned above, the AIC may be used for this evaluation index. However, the AIC is effective as an index for securing the generalization performance of the model and preventing the over-learning, but there are cases where the over-learning occurs even when optimization is performed by evaluation using the AIC.”
Therefore, it would have been obvious to combine the feature extraction workflow of Ishiguro_2021 and with an elastic-net model of Sarwat_2021 for the benefit of reducing overfitting to have a more accurate model to obtain the invention as specified in the claims.
Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over Ishiguro_2021, Koehrsen_2018, Chowdhury_2020, Huu_2021, and further in view of Cheng_2021 (US 20210064998 A1)
Claim 3:
Ishiguro_2021 makes obvious The computer implemented method of claim 1 wherein the original time series forecasting model(See claim 1)
Cheng_2021, however, makes obvious is L1 trend filtering. Par 28: “ In one example of single time series trend detection, the exemplary embodiments detect a trend in each time period. For each local trend, exemplary embodiments need to detect a time length and a slope. The exemplary embodiments can have threshold on length and slope to maintain only a subset of trends. The multivariate time series in the same group, e.g., stocks in the same sector or vehicle speed in the same road segment during a period time, usually has similar trend patterns. The challenge is how to detect the trend of the group as a whole characteristic for group behavior analysis. To address such issue, the exemplary embodiments use an l.sub.1 trend filtering method on the whole multi-variate time series. The exemplary embodiments learn the piecewise linear trends for all the time series jointly using the following equation:“
Ishiguro_2021, and Cheng_2021 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, and Cheng_2021.The rationale for doing so would have been to follow a motivation proposed in the art by Cheng_2021.
Cheng_2021 par 28 states: “The multivariate time series in the same group, e.g., stocks in the same sector or vehicle speed in the same road segment during a period time, usually has similar trend patterns. The challenge is how to detect the trend of the group as a whole characteristic for group behavior analysis. To address such issue, the exemplary embodiments use an l.sub.1 trend filtering method on the whole multi-variate time series. The exemplary embodiments learn the piecewise linear trends for all the time series jointly using the following equation:“ the inventor of Ishiguro_2021 discusses time series signal measurement from semiconductor manufacturing apparatus. This time series signal measurements may differ based on trends across time i.e.: increased production for seasonal demand. In such a scenario, the inventor of Ishiguro_2021 would be motivated to use L1 trend filtering In order to solve similar multivariate time series problems to detect the trend of a group as a whole, as it is a known time series model for that use.
Therefore, it would have been obvious to combine the time series model workflow of Ishiguro_2021 with the L1 trend filtering model of Cheng_2021 for the benefit of trend detection to obtain the invention as specified in the claims.
Claims 8-10 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Ishiguro_2021, Koehrsen_2018, Chowdhury_2020, Huu_2021 and O’Neil_2011 (“Impact of Managed Acceleration on In-Memory Database Analytic Workloads”)
Claim 8:
Ishiguro_2021 makes obvious The computer implemented method of claim 1 wherein: the non-transitory computer readable storage medium comprises par 24: “The data processing apparatus and the processing apparatus method according to the first embodiment will be described with reference to FIGS. 1 to 5. The data processing apparatus of the present embodiment includes, although not shown, a recording unit that records electronic data and a computing unit that computes the recorded electronic data. Since such a data processing apparatus can be implemented by a general computer that is represented by a personal computer (PC) and includes a central processing unit (CPU) for computing the electronic data, a storage unit for storing the electronic data and various processing programs, an input/output unit including a keyboard and a display, and a communication interface, the data processing apparatus is not shown.” )and first new time series forecasting model is valid. (Par 38: “ The iteration from the second step to the fifth step is performed until a minimum value of an AIC obtained in the third step in an m-th iteration is equal to a minimum value of an AIC obtained in the third step in an (m−1)th iteration. After the iteration is ended, a subset having a lowest AIC becomes a subset including the selected feature amount.” Examiner note: Where a determination of an AIC is a check of the validity of the model. )
Ishiguro_2021 does not expressly recite in-memory database … an in-memory database engine of the in-memory database
O’Neill_2011, however, makes obvious in-memory database … an in-memory database engine of the in-memory database abstract: “In-Memory Databases, such as SAP HANA, enable new levels of database performance by removing the disk bottleneck and by compressing data in memory. The consequence of this improved performance means that reports and analytic queries can now be processed on demand.”)
Ishiguro_2021, and O’Neill_2011 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Where Neill_2011 focuses on systems to perform the workflows outlined in Ishiguro_2021. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, and O’Neill_2011. The rationale for doing so would have been to follow a teaching proposed by Neill_2011. O’Neill_2011 page 1 par 1 states “The emergence of In-Memory Databases (IMDBs) has helped to erode the bottle necks of disk access latency and thus speed-up query processing times. For analytic queries, having data in-memory avoids the need to read tables from disk and the data is always available for immediate processing.” The inventor of Ishiguro_2021 would recognize that they had memory functionality, and would likely have used an in-memory database to gain the benefit to allow them to process the regression analysis results and AIC for each subset quicker. Therefore, it would have been obvious to combine the workflow and variable extraction techniques of Ishiguro_2021 with the in-memory databases of O’O’Neill_2011 for the benefit of faster processing to obtain the invention as specified in the claims.
Claim 9:
Ishiguro_2021 makes obvious The computer implemented method of claim 1 wherein: the non-transitory computer readable storage medium comprises an in-memory par 24: “The data processing apparatus and the processing apparatus method according to the first embodiment will be described with reference to FIGS. 1 to 5. The data processing apparatus of the present embodiment includes, although not shown, a recording unit that records electronic data and a computing unit that computes the recorded electronic data. Since such a data processing apparatus can be implemented by a general computer that is represented by a personal computer (PC) and includes a central processing unit (CPU) for computing the electronic data, a storage unit for storing the electronic data and various processing programs, an input/output unit including a keyboard and a display, and a communication interface, the data processing apparatus is not shown.” a contribution. See fig . 1 step S101 “create feature amount ranking”. Par 29: “ In the flowchart of FIG. 1, in the first step in the feature amount selection unit, firstly, significance of the feature amount, (Examiner note: a measure of contributions) that is, a feature amount ranking is created by using Fisher criterion which is one of filter methods (S101).”
Ishiguro_2021 does not expressly recite in-memory database … an in-memory database engine of the in-memory database
O’Neill_2011, however, makes obvious in-memory database … an in-memory database engine of the in-memory database abstract: “In-Memory Databases, such as SAP HANA, enable new levels of database performance by removing the disk bottleneck and by compressing data in memory. The consequence of this improved performance means that reports and analytic queries can now be processed on demand.”)
Ishiguro_2021, and O’Neill_2011 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Where O’Neill_2011 focuses on systems to perform the workflows outlined in Ishiguro_202 .Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, and O’Neill_2011. The rationale for doing so would have been to follow a teaching proposed by O’Neill_2011. O’Neill_2011 page 1 par 1 states “The emergence of In-Memory Databases (IMDBs) has helped to erode the bottle necks of disk access latency and thus speed-up query processing times. For analytic queries, having data in-memory avoids the need to read tables from disk and the data is always available for immediate processing.” The inventor of Ishiguro_2021 and Chowdhury_2019 , would recognize that they had memory functionality, and would likely have used an in-memory database to gain the benefit to allow them to process the regression analysis results and AIC for each subset quicker. Therefore, it would have been obvious to combine the workflow and variable extraction techniques of Ishiguro_2021, with the in-memory databases of O’Neill_2011 for the benefit of faster processing to obtain the invention as specified in the claims.
Claim 10:
Ishiguro_2021 makes obvious The computer implemented method of claim 1 wherein: the non-transitory computer readable storage medium comprises an in-memory par 24: “The data processing apparatus and the processing apparatus method according to the first embodiment will be described with reference to FIGS. 1 to 5. The data processing apparatus of the present embodiment includes, although not shown, a recording unit that records electronic data and a computing unit that computes the recorded electronic data. Since such a data processing apparatus can be implemented by a general computer that is represented by a personal computer (PC) and includes a central processing unit (CPU) for computing the electronic data, a storage unit for storing the electronic data and various processing programs, an input/output unit including a keyboard and a display, and a communication interface, the data processing apparatus is not shown.” ) first new time series forecasting model. par 32: “ In the third step in the feature amount selection unit, for all the subsets created in the second step, an evaluation index, which is a value serving as an index for evaluating prediction performance in a regression or classification problem, that is, a model formula evaluation index, is calculated and created (S104). The present embodiment involves a regression problem of estimating the RR, and the above-described AIC is adopted as an index for evaluating the prediction performance. The third step is to calculate the respective AIC for all subsets.” Examiner note: See Fig 1 steps S104 – S107. Where S104 “create model formula evaluation index for each subset” implies that each subset (remaining variables after each iteration) is used to create a new model which is being evaluated in each iteration. Where while the deleting step S105 occurs after S104, it is understood that each subset is necessary excluding variables from the original set of variables, as well as that this process is occurring literately and therefore repeats.
Ishiguro_2021 does not expressly recite in-memory database … an in-memory database engine of the in-memory database
O’Neill_2011, however, makes obvious in-memory database … an in-memory database engine of the in-memory database abstract: “In-Memory Databases, such as SAP HANA, enable new levels of database performance by removing the disk bottleneck and by compressing data in memory. The consequence of this improved performance means that reports and analytic queries can now be processed on demand.”)
Ishiguro_2021, and O’Neill_2011 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Where O’Neill_2011 focuses on systems to perform the workflows outlined in Ishiguro_2021 .Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, and O’Neill_2011. The rationale for doing so would have been to follow a teaching proposed by O’Neill_2011. O’Neill_2011 page 1 par 1 states “The emergence of In-Memory Databases (IMDBs) has helped to erode the bottle necks of disk access latency and thus speed-up query processing times. For analytic queries, having data in-memory avoids the need to read tables from disk and the data is always available for immediate processing.” The inventor of Ishiguro_2021 and Chowdhury_2019 , would recognize that they had memory functionality, and would likely have used an in-memory database to gain the benefit to allow them to process the regression analysis results and AIC for each subset quicker. Therefore, it would have been obvious to combine the workflow and variable extraction techniques of Ishiguro_2021, with the in-memory databases of O’Neill_2011 for the benefit of faster processing to obtain the invention as specified in the claims.
Claim 16:
The limitations of claim 16 are substantially the same as those of claim 1 and are therefore rejected due to the same reasons as outlined above for claim 1. Additionally, Ishiguro_2021 makes obvious the additional limitations of A computer system comprising: one or more processors :a software program, executable on said computer system, the software program configured to cause an in-memory par 24: “The data processing apparatus and the processing apparatus method according to the first embodiment will be described with reference to FIGS. 1 to 5. The data processing apparatus of the present embodiment includes, although not shown, a recording unit that records electronic data and a computing unit that computes the recorded electronic data. Since such a data processing apparatus can be implemented by a general computer that is represented by a personal computer (PC) and includes a central processing unit (CPU) for computing the electronic data, a storage unit for storing the electronic data and various processing programs, an input/output unit including a keyboard and a display, and a communication interface, the data processing apparatus is not shown.”)
Ishiguro_2021 does not expressly recite database engine of an in-memory database
O’Neill_2011, however, makes obvious database engine of an in-memory database
abstract: “In-Memory Databases, such as SAP HANA, enable new levels of database performance by removing the disk bottleneck and by compressing data in memory. The consequence of this improved performance means that reports and analytic queries can now be processed on demand.”)
Ishiguro_2021, and O’Neill_2011 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Where O’Neill_2011 focuses on systems to perform the workflows outlined in Ishiguro_2021 .Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, and O’Neill_2011. The rationale for doing so would have been to follow a teaching proposed by O’Neill_2011. O’Neill_2011 page 1 par 1 states “The emergence of In-Memory Databases (IMDBs) has helped to erode the bottle necks of disk access latency and thus speed-up query processing times. For analytic queries, having data in-memory avoids the need to read tables from disk and the data is always available for immediate processing.” The inventor of Ishiguro_2021 and Chowdhury_2019 , would recognize that they had memory functionality, and would likely have used an in-memory database to gain the benefit to allow them to process the regression analysis results and AIC for each subset quicker. Therefore, it would have been obvious to combine the workflow and variable extraction techniques of Ishiguro_2021, with the in-memory databases of O’Neill_2011 for the benefit of faster processing to obtain the invention as specified in the claims.
Claims 7, 17, 18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Ishiguro_2021, Koehrsen_2018, Chowdhury_2020, Huu_2021, Sarwat_2021 and O’Neil_2011
Claim 7:
The computer implemented method of claim 4 wherein: the non-transitory computer readable storage medium (see claim 4)
Ishiguro_2021 does not expressly recite comprises an in-memory database; and an in-memory database engine of the in-memory database performs the regularization.
Sarwat_2021, however, makes obvious the non-transitory computer readable storage medium comprises an in-memory par 126: “The methods and processes described herein can be embodied as code and/or data. The software code and data described herein can be stored on one or more machine-readable media (e.g., computer-readable media), which may include any device or medium that can store code and/or data for use by a computer system. When a computer system and/or processor reads and executes the code and/or data stored on a computer-readable medium, the computer system and/or processor performs the methods and processes embodied as data structures and code stored within the computer-readable storage medium. It should be appreciated by those skilled in the art that computer-readable media include removable and non-removable structures/devices that can be used for storage of information, such as computer-readable instructions, data structures, program modules, and other data used by a computing system/environment. A computer-readable medium includes, but is not limited to, volatile memory such as random access memories (RAM, DRAM, SRAM); and non-volatile memory such as flash memory, various read-only-memories (ROM, PROM, EPROM, EEPROM), magnetic and ferromagnetic/ferroelectric memories (MRAM, FeRAM), and magnetic and optical storage devices (hard drives, magnetic tape, CDs, DVDs); network devices; or other media now known or later developed that are capable of storing computer-readable information/data. Computer-readable media and machine-readable media should not be construed or interpreted to include any propagating signals. A computer-readable medium of the subject invention can be, for example, a compact disc (CD), digital video disc (DVD), flash memory device, volatile memory, or a hard disk drive (HDD), such as an external HDD or the HDD of a computing device, though embodiments are not limited thereto. A computing device can be, for example, a laptop computer, desktop computer, server, cell phone, or tablet, though embodiments are not limited thereto.”
and an in-memory par 35: “ To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection.”
Ishiguro_2021, and Sarwat_2021 do not expressly recite
O’Neill_2011 however, makes obvious abstract: “In-Memory Databases, such as SAP HANA, enable new levels of database performance by removing the disk bottleneck and by compressing data in memory. The consequence of this improved performance means that reports and analytic queries can now be processed on demand.”)
Ishiguro_2021, Sarwat_2021, and O’Neill_2011 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Where O’Neill_2011 focuses on systems to perform the workflows outlined in Ishiguro_2021, and Sarwat_2021. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, Sarwat_2021, and O’Neill_2011. The rationale for doing so would have been to follow a teaching proposed by O’Neill_2011. O’Neill_2011 page 1 par 1 states “The emergence of In-Memory Databases (IMDBs) has helped to erode the bottle necks of disk access latency and thus speed-up query processing times. For analytic queries, having data in-memory avoids the need to read tables from disk and the data is always available for immediate processing.” The inventor of Ishiguro_2021, Sarwat_2021 would recognize that they had memory functionality, and would likely have used an in-memory database to gain the benefit to allow them to process the regression analysis results and AIC for each subset quicker. Therefore, it would have been obvious to combine the workflow and variable extraction techniques of Ishiguro_2021, , Sarwat_2021 with the in-memory databases of O’Neill_2011 for the benefit of faster processing to obtain the invention as specified in the claims.
Claim 17:
Ishiguro_2021, and O’Neill_2011 make obvious The computer system as in claim 16 wherein the in-memory database engine is further configured to see claim 16)
Sarwat_2021, however, makes obvious further comprising subjecting variables to regularization (par 35: “ To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection. )
Ishiguro_2021, O’Neill_2011 and Sarwat_2021 are analogous art to the claimed invention because they are from the same field of endeavor called regression and time series data forecasting. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, O’Neill_2011 and Sarwat_2021. The rationale for doing so would have been to follow a teaching and motivation proposed in the prior art.
Sarwat_2021 teaches that 2021 par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection.”
Ishiguro_2021 wants to also decrease overfitting, where the focus of Ishiguro_2021 is to reduce it. See par 11: “As mentioned above, the AIC may be used for this evaluation index. However, the AIC is effective as an index for securing the generalization performance of the model and preventing the over-learning, but there are cases where the over-learning occurs even when optimization is performed by evaluation using the AIC.”
Therefore, it would have been obvious to combine the feature extraction workflow and database engine of Ishiguro_2021, and O’Neill_2011 with the regularization of variables of Sarwat_2021 for the benefit of reducing overfitting to have a more accurate model to obtain the invention as specified in the claims.
Claim 18:
Ishiguro_2021, Sarwat_2021 and O’Neill_2011 make obvious The computer system of claim 17 wherein the regularization
Ishiguro_2021 does not expressly recite comprises lasso or ridge.
Sarwat_2021 makes obvious comprises lasso or ridge. (par 35: “ To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection)
Ishiguro_2021, , O’Neill_2011 and Sarwat_2021 are analogous art to the claimed invention because they are from the same field of endeavor called regression and time series data forecasting. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, , O’Neill_2011 and Sarwat_2021. The rationale for doing so would have been to follow a teaching and motivation proposed in the prior art.
Sarwat_2021 teaches that 2021 par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection.”
Ishiguro_2021 wants to also decrease overfitting, where the focus of Ishiguro_2021 is to reduce it. See par 11: “As mentioned above, the AIC may be used for this evaluation index. However, the AIC is effective as an index for securing the generalization performance of the model and preventing the over-learning, but there are cases where the over-learning occurs even when optimization is performed by evaluation using the AIC.”
Therefore, it would have been obvious to combine the feature extraction workflow and database engine of Ishiguro_2021, O’Neill_2011 with the regularization of variables using lasso or ridge of Sarwat_2021 for the benefit of reducing overfitting to have a more accurate model to obtain the invention as specified in the claims.
Claim 20:
Ishiguro_2021, Chowdhury_2019 , and O’Neill_2011 make obvious The computer system of claim 16 wherein the third new time series forecasting model (see claim 16)
Ishiguro_2021 does not expressly recite is elastic-net linear regression.
Sarwat_2021 however makes obvious is elastic-net linear regression.
par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection. In datasets with high correlation between the predictors, LASSO's variable selection performs poorly. Also, for datasets where the dimensionality, p, is much less when compared to the number of observations n, ridge regularization outperforms LASSO. An elastic net is a combination of ridge and LASSO regularization techniques that applies an elastic net penalty. Given the tuning parameter λ that controls the penalty's magnitude, the model solves the objective function, F(⋅) defined in Equation (2.1) over its entire grid space. Let f(y, η) denote the negative log-likelihood function for the i.sup.th record. If the response is of type Gaussian, then f(y, η)=.sup.(y-η).sup.2. The variable α controls the elastic net penalty, with α=0 denoting ridge, α=1 denoting LASSO, and α∈(0, 1) denoting elastic net.
Ishiguro_2021, , O’Neill_2011 and Sarwat_2021 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, O’Neill_2011 and Sarwat_2021. The rationale for doing so would have been to follow a teaching and motivation proposed in the prior art.
Sarwat_2021 teaches that 2021 par 35-36: “To reduce overfitting and improve the ordinary least squares estimates of a linear regression model, two types of penalization techniques are used: ridge regularization to minimize the residual sum of squares with respect to the L.sub.2 norm of the coefficients that keeps all predictors in the model; and least absolute shrinkage and selection operator (LASSO) regularization to minimize the residual sum of squares contingent on the L.sub.1 norm of the coefficients through continuous shrinkage and automatic variable selection. In datasets with high correlation between the predictors, LASSO's variable selection performs poorly. Also, for datasets where the dimensionality, p, is much less when compared to the number of observations n, ridge regularization outperforms LASSO. An elastic net is a combination of ridge and LASSO regularization techniques that applies an elastic net penalty
Ishiguro_2021 wants to also decrease overfitting, where the focus of Ishiguro_2021 is to reduce it. See par 11: “As mentioned above, the AIC may be used for this evaluation index. However, the AIC is effective as an index for securing the generalization performance of the model and preventing the over-learning, but there are cases where the over-learning occurs even when optimization is performed by evaluation using the AIC.”
Therefore, it would have been obvious to combine the feature extraction workflow and database engine of Ishiguro_2021 and O’Neill_2011 with an elastic-net model of Sarwat_2021 for the benefit of reducing overfitting to have a more accurate model to obtain the invention as specified in the claims.
Claim 15 is rejected under 35 U.S.C. 103 as being unpatentable over Ishiguro_2021, Koehrsen_2018, Chowdhury_2020, Huu_2021, Sarwat_2021, and Cheng_2021
Claim 15:Ishiguro_2021, Chowdhury_2019 and Sarwat_2021 make obvious The non-transitory computer readable storage medium as in of claim 11 wherein the third new time series forecasting model (see claim 11)
Cheng_2021, however, makes obvious is L1 trend filtering. Par 28: “ In one example of single time series trend detection, the exemplary embodiments detect a trend in each time period. For each local trend, exemplary embodiments need to detect a time length and a slope. The exemplary embodiments can have threshold on length and slope to maintain only a subset of trends. The multivariate time series in the same group, e.g., stocks in the same sector or vehicle speed in the same road segment during a period time, usually has similar trend patterns. The challenge is how to detect the trend of the group as a whole characteristic for group behavior analysis. To address such issue, the exemplary embodiments use an l.sub.1 trend filtering method on the whole multi-variate time series. The exemplary embodiments learn the piecewise linear trends for all the time series jointly using the following equation:“
Ishiguro_2021 , Sarwat_2021, and Cheng_2021 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, Sarwat_2021 and Cheng_2021.The rationale for doing so would have been to follow a motivation proposed in the art by Cheng_2021.
Cheng_2021 par 28 states: “The multivariate time series in the same group, e.g., stocks in the same sector or vehicle speed in the same road segment during a period time, usually has similar trend patterns. The challenge is how to detect the trend of the group as a whole characteristic for group behavior analysis. To address such issue, the exemplary embodiments use an l.sub.1 trend filtering method on the whole multi-variate time series. The exemplary embodiments learn the piecewise linear trends for all the time series jointly using the following equation:“ the inventor of Ishiguro_2021 discusses time series signal measurement from semiconductor manufacturing apparatus. This time series signal measurements may differ based on trends across time i.e.: increased production for seasonal demand. In such a scenario, the inventor of Ishiguro_2021 would be motivated to use L1 trend filtering In order to solve similar multivariate time series problems to detect the trend of a group as a whole, as it is a known time series model for that use.
Therefore, it would have been obvious to combine the time series model workflow of Ishiguro_2021, and Sarwat_2021 with the L1 trend filtering model of Cheng_2021 for the benefit of trend detection to obtain the invention as specified in the claims.
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Ishiguro_2021, Koehrsen_2018, Chowdhury_2020, Huu_2021, Neil_2011, and Cheng_2021
Claim 19:
Ishiguro_2021, and O’Neill_2011 make obvious The computer system of claim 16 wherein the time third new series forecasting model (see claim 16)
Ishiguro_2021, and O’Neill_2011 do not expressly recite is L1 trend filtering.
Cheng_2021, however, makes obvious is L1 trend filtering. Par 28: “ In one example of single time series trend detection, the exemplary embodiments detect a trend in each time period. For each local trend, exemplary embodiments need to detect a time length and a slope. The exemplary embodiments can have threshold on length and slope to maintain only a subset of trends. The multivariate time series in the same group, e.g., stocks in the same sector or vehicle speed in the same road segment during a period time, usually has similar trend patterns. The challenge is how to detect the trend of the group as a whole characteristic for group behavior analysis. To address such issue, the exemplary embodiments use an l.sub.1 trend filtering method on the whole multi-variate time series. The exemplary embodiments learn the piecewise linear trends for all the time series jointly using the following equation:“
Ishiguro_2021, O’Neill_2011 and Cheng_2021 are analogous art to the claimed invention because they are from the same field of endeavor called machine learning. Before the effective filing date, it would have been obvious to a person of ordinary skill in the art to combine Ishiguro_2021, O’Neill_2011 and Cheng_2021.The rationale for doing so would have been to follow a motivation proposed in the art by Cheng_2021.
Cheng_2021 par 28 states: “The multivariate time series in the same group, e.g., stocks in the same sector or vehicle speed in the same road segment during a period time, usually has similar trend patterns. The challenge is how to detect the trend of the group as a whole characteristic for group behavior analysis. To address such issue, the exemplary embodiments use an l.sub.1 trend filtering method on the whole multi-variate time series. The exemplary embodiments learn the piecewise linear trends for all the time series jointly using the following equation:“ the inventor of Ishiguro_2021 discusses time series signal measurement from semiconductor manufacturing apparatus. This time series signal measurements may differ based on trends across time i.e.: increased production for seasonal demand. In such a scenario, the inventor of Ishiguro_2021 would be motivated to use L1 trend filtering In order to solve similar multivariate time series problems to detect the trend of a group as a whole, as it is a known time series model for that use.
Therefore, it would have been obvious to combine the time series model workflow and engine of Ishiguro_2021, and O’Neill_2011 with the L1 trend filtering model of Cheng_2021 for the benefit of trend detection to obtain the invention as specified in the claims.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to AHMAD HUSSAM SHALABY whose telephone number is (571)272-7414. The examiner can normally be reached Mon-Fri 7:30am - 5pm.
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, Emerson Puente can be reached at 5712723652. 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.
/A.H.S./Examiner, Art Unit 2187
/EMERSON C PUENTE/Supervisory Patent Examiner, Art Unit 2187
Prosecution Insights