CTNF 18/498,406 CTNF 101938 DETAILED ACTION Notice of Pre-AIA or AIA Status 07-03-aia AIA 15-10-aia The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA. Claim Rejections - 35 USC § 101 07-04-01 AIA 07-04 35 U.S.C. 101 reads as follows: Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title. Claims 1–20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. Regarding independent claims 1, 2, and 15 Step 1 – whether the claim falls within a statutory category. See MPEP 2106.03. Claim 1 is drawn to a system (machine) claim; claim 2 is drawn to a method (process) claim; and claim 15 is drawn to one or more non-transitory, computer-readable mediums (manufacture) claim. Each independent claim therefore falls within one of the four statutory categories of 35 U.S.C. 101. Step 1 = Yes. Step 2A Prong One – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II. Independent claims 1, 2, and 15 recite the following limitations: “generating a first feature input based on the first dataset”; “inputting the first feature input into a first plurality of statistical routines to determine a first plurality of respective outputs, wherein the first plurality of statistical routines performs a respective first statistical analysis of the first feature input, and wherein each of the first plurality of statistical routines is based on a first respective algorithm”; “determining a first aggregate statistical profile for the first dataset based on the first plurality of respective outputs” (claim 1 further reciting that the profile “comprises a series of values corresponding to the first plurality of respective outputs, wherein the series of values is based on a respective effectiveness of a plurality of model types for generating predictions based on the one or more categories of data trends”); “selecting, based on the [respective effectiveness of the plurality of model types / first aggregate statistical profile], a first untuned hyperparameter, for a first plurality of untrained models, for tuning to a specific value”; and “tuning the first untuned hyperparameter to the specific value”. These limitations recite a judicial exception, namely a combination of mathematical concepts and mental processes . Mathematical concepts (MPEP § 2106.04(a)(2), subsection I). Under the broadest reasonable interpretation, the recited “statistical routines” each perform a “statistical analysis” that, as set forth in the specification, comprises mathematical calculations and relationships--e.g., descriptive statistics such as mean, median, variance, and standard deviation; inferential statistics; hypothesis testing; regression analysis; analysis of variance (ANOVA); chi-square testing; and principal component analysis (see specification ¶¶ [0036]–[0039], [0082]). The “first aggregate statistical profile” is recited as a “series of values”--disclosed as an array or profile matrix (¶¶ [0048], [0083])--and thus organizes and manipulates information through mathematical correlations. “Tuning the first untuned hyperparameter to the specific value” and “selecting…for tuning to a specific value” likewise express mathematical relationships between the statistical outputs and the selected value (¶ [0101]). Limitations directed to mathematical relationships, formulas, equations, and calculations fall within the mathematical-concepts grouping. See MPEP § 2106.04(a)(2), subsection I. Mental processes (MPEP § 2106.04(a)(2), subsection III). The steps of “generating a first feature input,” “determining a first aggregate statistical profile,” and “selecting…a first untuned hyperparameter” recite observations, evaluations, judgments, and opinions of the type that can be practically performed in the human mind, or by a human using pen and paper. Selecting a feature, evaluating and aggregating statistical outputs into a profile, and selecting a hyperparameter on the basis of that profile are evaluative and judgment steps. Apart from the generic computer components addressed in Prong Two, nothing in these limitations forecloses their performance as mental steps. Such concepts fall within the mental-processes grouping. See MPEP § 2106.04(a)(2), subsection III. The recitation of generic “statistical routines” and “untrained models” does not remove these limitations from the abstract-idea groupings, because the routines and models are claimed at a high level of generality and merely carry out the recited calculations and evaluations. This treatment is consistent with the 2024 AI SME Update, Example 47 (claim 2), in which “training” recited as a backpropagation/gradient-descent calculation was found to recite a mathematical concept and “detecting/analyzing” were found to recite mental processes; and Example 48 (claim 1), in which a mathematical formula evaluated with a deep neural network was found to recite a mathematical concept. The claim therefore recites a judicial exception. Step 2A Prong One = Yes. Step 2A Prong Two – whether the claim as a whole integrates the recited judicial exception into a practical application. See MPEP 2106.04(d). The additional elements recited beyond the judicial exception are evaluated individually and in combination. Claim 1 recites “one or more processors,” “one or more non-transitory, computer-readable mediums comprising instructions that when executed by the one or more processors cause operations,” “receiving a first dataset,” and “generating for display, on a user interface, a recommendation for using the tuned first model for time-series forecasting.” Claim 15 recites “one or more non-transitory, computer-readable mediums comprising instructions that when executed by one or more processors causes operations” and “receiving a first dataset.” Claim 2 recites no hardware beyond the generic “statistical routines” and “untrained models” used as tools, together with “receiving a first dataset.” The recited “processors” and “non-transitory, computer-readable mediums” are described at a high level of generality (see specification ¶¶ [0065]–[0067]) and amount to mere instructions to implement the abstract idea on a generic computer, or merely use a computer as a tool to perform the abstract idea. See MPEP § 2106.05(f). The “receiving a first dataset” limitation is mere data gathering recited at a high level of generality and constitutes insignificant extra-solution activity. See MPEP § 2106.05(g). The claim 1 limitation “generating for display, on a user interface, a recommendation for using the tuned first model” is likewise insignificant extra-solution activity, because it merely outputs and presents the result of the abstract idea (see MPEP § 2106.05(g)), and additionally does no more than generally link the abstract idea to the particular field of time-series forecasting (see MPEP § 2106.05(h)). The claim does not improve the functioning of a computer or any other technology or technical field. See MPEP § 2106.05(a). The benefit asserted in the specification--reducing redundant model training, fitting, and tuning time and lowering the barrier to entry for forecasting (¶¶ [0004]–[0011])--is an improvement to the abstract idea itself (a more efficient model-selection and hyperparameter-selection process), not an improvement to computer capabilities or to a technological process. The recited “untrained models compris[ing] respective algorithms for time-series forecasting” and “default hyperparameter tuning” merely confine the exception to a field of use and invoke generic software. The claims recite no concrete technological action that applies the result of the analysis to effect a particular technological outcome. This is consistent with the 2024 AI SME Update: Example 47 (claim 2) and Example 48 (claim 1) were ineligible where a generic computer or neural network merely applied the exception, in contrast to Example 47 (claim 3), which was eligible only because additional steps used the analysis to drop malicious packets and block traffic in real time. The instant claims contain no analogous limitation. Considered individually and in combination, the additional elements do not integrate the exception into a practical application. Step 2A = Yes; the claims are directed to the abstract idea. Step 2B – whether the claim recites additional elements that amount to significantly more than the judicial exception. See MPEP 2106.05. The additional elements are reconsidered individually and as an ordered combination. The “processors” and “non-transitory, computer-readable mediums” remain no more than mere instructions to apply the exception using a generic computer and do not supply an inventive concept. See MPEP § 2106.05(f). The “receiving a first dataset” step and the claim 1 “generating for display, on a user interface, a recommendation” step, reconsidered at Step 2B, remain insignificant extra-solution activity and additionally recite functions that are well-understood, routine, and conventional--namely receiving or storing data and presenting results on a generic user interface. See MPEP § 2106.05(d), subsection II. The “statistical routines” and “untrained models” are invoked at a high level of generality as tools that carry out the recited exception and add nothing beyond the abstract idea. No additional element, alone or in combination, amounts to significantly more than the judicial exception. Step 2B = No. Claims 1, 2, and 15 are ineligible. Regarding dependent claims 3–14 and 16–20 Claims 3–14 depend from claim 2 and claims 16–20 depend from claim 15. Each dependent claim merely narrows the previously identified abstract idea and does not add any additional element--beyond those addressed above with respect to the independent claims--that integrates the exception into a practical application or amounts to significantly more. For the reasons given for claims 1, 2, and 15, the dependent claims are likewise rejected. The step-by-step analysis follows. Step 1 – statutory category. See MPEP 2106.03. Claims 3–14 are method (process) claims and claims 16–20 are manufacture claims. Each falls within a statutory category. Step 1 = Yes. Step 2A Prong One – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II. Regarding claim 3 (and analogously claim 16), the claim recites “determining a first time period for a first model of the first plurality of statistical routines,” “determining a first statistical variation for the first model over the first time period,” and “determining a respective output…based on the first statistical variation” . The recited statistical variation is a mathematical calculation (e.g., variance or standard deviation, see ¶ [0082]) and the recited determinations are evaluations practically performed in the human mind. The limitations fall within the mathematical-concepts and mental-processes groupings. See MPEP § 2106.04(a)(2), subsections I and III. Regarding claim 4 (and analogously claim 17), the claim recites “comparing a first respective output…to a threshold value” and “determining a difference between the first respective output and the threshold value, wherein selecting the first untuned hyperparameter is based on the difference” . Comparing a value to a threshold and computing a difference are evaluations and mathematical calculations that fall within the mental-processes and mathematical-concepts groupings. See MPEP § 2106.04(a)(2), subsections I and III. Regarding claim 5 (and analogously claim 18), the claim recites “filtering a first plurality of untuned hyperparameters based on the first aggregate statistical profile to generate a filtered subset…and selecting the first untuned hyperparameter from the filtered subset” . Filtering and selecting on the basis of the profile are evaluations and judgments that fall within the mental-processes grouping. See MPEP § 2106.04(a)(2), subsection III. Regarding claim 6 (and analogously claim 19), the claim recites “filtering a first plurality of untuned hyperparameters based on an age of the first dataset to generate a filtered subset…and selecting the first untuned hyperparameter from the filtered subset” . Filtering and selecting based on dataset age are evaluations and judgments within the mental-processes grouping. See MPEP § 2106.04(a)(2), subsection III. Regarding claim 7, the claim recites “filtering a first plurality of untuned hyperparameters based on a reliability of the first dataset to generate a filtered subset…and selecting the first untuned hyperparameter from the filtered subset” . Filtering and selecting based on dataset reliability are evaluations and judgments within the mental-processes grouping. See MPEP § 2106.04(a)(2), subsection III. Regarding claim 8, the claim recites “ranking a first plurality of untuned hyperparameters based on the first aggregate statistical profile to generate a ranked order…and selecting the first untuned hyperparameter based on the ranked order” . Ranking and selecting are evaluations and judgments within the mental-processes grouping, and the ordering reflects mathematical relationships within the mathematical-concepts grouping. See MPEP § 2106.04(a)(2), subsections I and III. Regarding claim 9, the claim recites “determining respective training time predictions for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile…and selecting the first untuned hyperparameter based on the respective training time predictions” . Predicting and selecting are evaluations and mathematical calculations within the mental-processes and mathematical-concepts groupings. See MPEP § 2106.04(a)(2), subsections I and III. Regarding claim 10, the claim recites “determining respective performance predictions for each of a first plurality of untuned hyperparameters…and selecting the first untuned hyperparameter based on the respective performance predictions” . Predicting and selecting are evaluations and mathematical calculations within the mental-processes and mathematical-concepts groupings. See MPEP § 2106.04(a)(2), subsections I and III. Regarding claim 11, the claim recites “determining respective predictions for a number of hyperparameters requiring training…and selecting the first untuned hyperparameter based on the respective predictions for the number of hyperparameters requiring training” . Predicting and selecting are evaluations and mathematical calculations within the mental-processes and mathematical-concepts groupings. See MPEP § 2106.04(a)(2), subsections I and III. Regarding claim 12, the claim recites “determining respective sample size requirements for training…and selecting the first untuned hyperparameter based on the respective sample size requirements for training” . Determining requirements and selecting are evaluations and mathematical calculations within the mental-processes and mathematical-concepts groupings. See MPEP § 2106.04(a)(2), subsections I and III. Regarding claim 13 (and analogously claim 20), the claim recites “determining respective processing power requirements for training…and selecting the first untuned hyperparameter based on the respective processing power requirements for training” . Determining requirements and selecting are evaluations and mathematical calculations within the mental-processes and mathematical-concepts groupings. See MPEP § 2106.04(a)(2), subsections I and III. Regarding claim 14, the claim recites “generating a profile matrix for the first dataset” and “populating values of the profile matrix based on a comparison of the first plurality of respective outputs and respective model requirements for the first plurality of untrained models” . Generating and populating a matrix of values organizes and manipulates information through mathematical correlations (mathematical-concepts grouping), and the recited comparison is an evaluation practically performed in the human mind (mental-processes grouping). See MPEP § 2106.04(a)(2), subsections I and III. Each of claims 3–14 and 16–20 therefore recites a judicial exception. Step 2A Prong One = Yes. Step 2A Prong Two – integration into a practical application. See MPEP 2106.04(d). The dependent claims add no additional element beyond those addressed above for the independent claims from which they depend. Claims 3–14 add no hardware beyond the generic tools already addressed; claims 16–20 inherit only the generic “non-transitory, computer-readable mediums” and “processors” of claim 15. The added limitations merely narrow the abstract idea by specifying how the statistical analysis is performed (claims 3, 14) or how the hyperparameter is selected (claims 4–13, 17–20). Each such limitation is itself part of the abstract idea and does not constitute an additional element. The generic computer components remain mere instructions to apply the exception (MPEP § 2106.05(f)), and any data gathering or output remains insignificant extra-solution activity (MPEP § 2106.05(g)) that merely links the exception to a field of use (MPEP § 2106.05(h)). No dependent claim recites an improvement to computer functionality or to any technology. See MPEP § 2106.05(a). The exception is not integrated into a practical application. Step 2A = Yes. Step 2B – significantly more. See MPEP 2106.05. For the same reasons stated for the independent claims, the additional elements of the dependent claims, considered individually and in combination, do not amount to significantly more than the judicial exception. The narrowing limitations of claims 3–14 and 16–20 recite mathematical relationships and mental processes that are practically capable of being performed in the human mind or with the aid of pen and paper, and the only additional elements are generic computer components and insignificant extra-solution activity that are well-understood, routine, and conventional. See MPEP § 2106.05(d), subsection II. Step 2B = No. Accordingly, claims 1–20 are rejected under 35 U.S.C. 101 as being directed to an abstract idea without significantly more. Claim Rejections - 35 USC § 103 07-06 AIA 15-10-15 In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status. 07-20-aia AIA 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. 07-103 AIA The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action. 07-23-aia AIA The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows: 1. Determining the scope and contents of the prior art. 2. Ascertaining the differences between the prior art and the claims at issue. 3. Resolving the level of ordinary skill in the pertinent art. 4. Considering objective evidence present in the application indicating obviousness or nonobviousness. 07-21-aia AIA Claim s 1–5, 13, 15–18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Ahuja et al. (Ahuja) , U.S. Patent Application Publication No. US 2023/0153394 A1, in view of Rossi et al. (Rossi) , U.S. Patent Application Publication No. US 2024/0152769 A1, and further in view of Hetherington et al. (Hetherington) , U.S. Patent Application Publication No. US 2020/0097810 A1 . As to independent Claim 1 , the preamble recites "A system for minimizing development time in artificial intelligence models by automating hyperparameter selection based on dataset fittings of time-series data, the system comprising:" Ahuja discloses this preamble. Ahuja is directed to an automated machine-learning pipeline that, in a single feed-forward pass, measures temporal statistics of an input timeseries and uses those statistics to select a forecasting algorithm from a plurality of forecasting algorithms and to tune the selected algorithm's hyperparameters, investing computer resources asymmetrically in the most promising algorithm so as to consume "fewer computer resources to achieve a given accuracy" ( Ahuja , Title: "One-pass approach to automated timeseries forecasting"; Ahuja , Abstract). This is the same end pursued by the claimed invention -- minimizing development time by automating hyperparameter selection based on dataset fittings of time-series data. Ahuja carries the bulk of the limitations of Claim 1: "one or more processors; and one or more non-transitory, computer-readable mediums comprising instructions that when executed by the one or more processors cause operations comprising" -- Ahuja discloses a "computer" that "hosts and operates an ML pipeline" which automatically performs the recited measuring, selecting, and tuning operations ( Ahuja , ¶ [0031]). The computer is embodied as "a rack server such as a blade, a personal computer" ( Ahuja , ¶ [0032]) -- generic hardware that inherently comprises one or more processors together with the non-transitory storage that holds the pipeline's executable instructions (FIG. 1, COMPUTER 100); "receiving a first dataset, wherein the first dataset comprises one or more categories of data trends" -- Ahuja teaches "computer 100 stores and operates or processes ML library 130 and original timeseries 111 that is a temporally ordered sequence of tuples" ( Ahuja , ¶ [0034]), where the "original timeseries 111" corresponds to the recited "first dataset." Ahuja further teaches "Each tuple may correspond to a distinct time in the past" ( Ahuja , ¶ [0034]), where the tuples are considered the recited categories; "inputting the first feature input into a first plurality of statistical routines to determine a first plurality of respective outputs, wherein the first plurality of statistical routines performs a respective first statistical analysis of the first feature input" -- Ahuja teaches a first plurality of statistical routines that each perform a respective statistical analysis to produce a first plurality of respective outputs: the pipeline "performs statistical tests on the timeseries" ( Ahuja , ¶ [0021]) and "automatically derives temporal statistics 120 by quantitatively analyzing the tuples in original timeseries 111" ( Ahuja , ¶ [0036]), those derivations including "a mean, mode, maximum, entropy, or variance" together with frequency and seasonality ( Ahuja , ¶¶ [0036], [0039]). The set of statistical tests/derivations reads on the "first plurality of statistical routines," and the resulting computed values are the "first plurality of respective outputs." With respect to the recited "first feature input" that is input into those statistical routines: per the sequence of Claim 1, the first feature input is first generated from the first dataset and is then input into the first plurality of statistical routines. Ahuja analyzes the time-series data directly and does not itself generate a discrete first feature input; that first feature input is generated by Rossi , as set forth below, and in the combination of Ahuja and Rossi the first feature input generated by Rossi is provided as the input to the first plurality of statistical routines taught by Ahuja , thereby satisfying "inputting the first feature input into a first plurality of statistical routines to determine a first plurality of respective outputs"; "selecting … a first untuned hyperparameter, for a first plurality of untrained models, for tuning to a specific value, wherein the first plurality of untrained models comprises respective algorithms for time-series forecasting, and wherein each of the first plurality of untrained models comprises default hyperparameter tuning" -- Ahuja maintains "a set of many forecasting algorithms … available internally for the ML pipeline to choose from" ( Ahuja , ¶ [0017]) -- the recited "first plurality of untrained models" -- and the algorithm-selection step "evaluates each forecasting algorithm quickly for its approximate goodness of fit, using at least some of predefined set of default model hyperparameters for each" ( Ahuja , ¶ [0017]). Those "predefined … default model hyperparameters for each" forecasting algorithm read on "default hyperparameter tuning." Ahuja then selects the hyperparameters of the most accurate forecasting algorithm for tuning ( Ahuja , claim 1; Ahuja , ¶ [0022]: "Inform the hyperparameters tuning stage to reduce tuning range(s)"); and "tuning the first untuned hyperparameter to the specific value to generate a tuned first model" -- Ahuja "tun[es] … hyperparameters of the most accurate ML algorithm" based on the temporal statistic so that "[t]he result from the ML pipeline is a rigorously trained and production ready ML model" ( Ahuja , ¶ [0031]; Ahuja , claim 1). That production-ready model is the recited "tuned first model." Ahuja thus measures temporal statistics directly from the tuples of the timeseries, selects an algorithm on the basis of those statistics, and tunes the selected algorithm's hyperparameters. Accordingly, Ahuja does not teach: "generating a first feature input based on the first dataset" ; "determining a first aggregate statistical profile for the first dataset based on the first plurality of respective outputs, wherein the first aggregate statistical profile comprises a series of values corresponding to the first plurality of respective outputs, wherein the series of values is based on a respective effectiveness of a plurality of model types for generating predictions based on the one or more categories of data trends" ; and "generating for display, on a user interface, a recommendation for using the tuned first model for timeseries forecasting." In the same field of endeavor, Rossi supplies these limitations through a meta-learning forecasting framework that computes, from the input time-series dataset, a meta-feature vector, supplies that vector to a meta-learner which scores each forecasting model in a plurality of forecasting models, and surfaces the selected model to the user. For the recited "generating a first feature input based on the first dataset," Rossi computes "a time-series meta-feature vector based on the time-series dataset" and "extracts meta-features from the time-series dataset" ( Rossi , Abstract; Rossi , ¶ [0075]). The meta-feature vector that Rossi derives from the dataset is the recited first feature input generated from the first dataset. As noted above with respect to the inputting limitation, this generated first feature input is the input that, in the combination of Ahuja and Rossi , is provided to the first plurality of statistical routines taught by Ahuja ; the claimed sequence of generating a first feature input and then inputting that first feature input into the first plurality of statistical routines is therefore satisfied by Rossi generating the first feature input and Ahuja providing the statistical routines that operate upon it. For the recited "determining a first aggregate statistical profile … comprises a series of values corresponding to the first plurality of respective outputs, wherein the series of values is based on a respective effectiveness of a plurality of model types," Rossi 's meta-learner generates, from the meta-feature vector, "a performance score for a forecasting model … [and] select[s] the forecasting model from a plurality of forecasting models based on the performance score" ( Rossi , Abstract). The performance score is produced "for each" forecasting model ( Rossi , ¶ [0070]) by reference to "performance data which measures the performance of a plurality of forecasting models as applied to" the time-series data ( Rossi , ¶ [0055]) and is organized as a tensor of per-model performance values ( Rossi , ¶ [0058]). The collection of per-forecasting-model performance scores is the recited "series of values," and because each score quantifies how well its corresponding forecasting model is predicted to perform on the input timeseries, the series of values is "based on a respective effectiveness of a plurality of model types for generating predictions" exactly as claimed ( Rossi , ¶ [0038]; Rossi , ¶ [0044]). For the recited "generating for display, on a user interface, a recommendation for using the tuned first model for timeseries forecasting," Rossi discloses "User interface 215" via which the meta-learning apparatus "may display the selected model or the forecasted data from the selected model to the user" ( Rossi , ¶ [0037]; FIG. 2). Displaying the selected forecasting model to the user via user interface 215 is the recited recommendation, presented for display on a user interface. Ahuja and Rossi are analogous to the claimed invention because both are directed to automated machine-learning systems that select a time-series forecasting model from a plurality of candidate forecasting models on the basis of statistical characteristics computed from an input time-series dataset. It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine Rossi 's meta-feature-vector computation, per-model performance scoring, and user-interface presentation of the selected model with Ahuja 's one-pass forecasting pipeline. The motivation to combine is supplied by Rossi itself, which teaches that this meta-learning architecture identifies "a forecasting model with the highest expected performance for an input time-series dataset" ( Rossi , ¶ [0038]) and displays the result to the user ( Rossi , ¶ [0037]). Incorporating that architecture into Ahuja , whose pipeline already selects from "a set of many forecasting algorithms" and tunes the selected algorithm's hyperparameters ( Ahuja , ¶¶ [0017], [0031]), yields a system that is more accurate, more transparent to the user, and that "consumes fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract). A skilled artisan would have had a reasonable expectation of success because Ahuja and Rossi operate on the same input -- a time-series dataset characterized by statistical descriptors -- and both produce the same kind of output, a selected forecasting model. The combination of Ahuja and Rossi , however, does not teach: "wherein the first dataset comprises one or more categories of data trends" ; and "wherein each of the first plurality of statistical routines is based on a first respective algorithm." In the same field of endeavor, Hetherington teaches "wherein the first dataset comprises one or more categories of data trends" by describing "automatically generating statistical features describing trends in time-series data that may then become inputs to machine learning models" through "a framework using several automated techniques" ( Hetherington , Abstract; Hetherington , ¶ [0008]), such that the time-series dataset received and processed by Hetherington 's framework comprises one or more categories of data trends. Hetherington further teaches "wherein each of the first plurality of statistical routines is based on a first respective algorithm" because its framework is built around "a set of algorithms for selecting a number and size of window based statistical features" ( Hetherington , Abstract) and applies multiple algorithmically distinct statistical routines, including "Grubb's test, student's t-test, generalized extreme [value]" ( Hetherington , ¶ [0035]) and moving-window-based statistical functions ( Hetherington , ¶ [0048]), each of which is based on its own respective algorithm. Ahuja , Rossi , and Hetherington are analogous to the claimed invention because all three are directed to automated machine-learning systems that derive statistical descriptors from an input time-series dataset for use in downstream model selection and hyperparameter configuration. It would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to further modify the system of Ahuja and Rossi to receive a time-series dataset characterized in terms of its data trends and to apply a first plurality of statistical routines each based on a first respective algorithm, as taught by Hetherington . The motivation to combine is supplied by Hetherington itself, which explains that such automated, algorithmically distinct statistical-feature generation "yield[s] best scores in terms of prediction accuracy" ( Hetherington , Abstract) and that "the benefits of using moving window based statistical functions" have been identified in the time-series analysis literature ( Hetherington , ¶ [0048]). Combining Hetherington 's explicit trend-based feature framework with Ahuja 's temporal-statistics-driven model selection and Rossi 's meta-learner performance scoring produces a system that improves the predictive accuracy of the resulting tuned forecasting model. Ahuja and Hetherington are both assigned to Oracle International Corporation and represent successive stages of the same line of Oracle AutoML-for-forecasting research, providing an express expectation that their teachings would be combined. Claim 2 is a method reciting the same receiving, generating, inputting, determining, selecting, and tuning operations as Claim 1 but containing fewer limitations -- it omits the user-interface display recitation of Claim 1 and recites the aggregate statistical profile and the selecting step in broader terms. Accordingly, Claim 2 is rejected on the same rationale as Claim 1. As to dependent Claim 3 , which recites "wherein determining the first plurality of respective outputs further comprises," the combination of Ahuja and Rossi does not teach: "determining a first time period for a first model of the first plurality of statistical routines" ; "determining a first statistical variation for the first model over the first time period" ; and "determining a respective output, of the first plurality of respective outputs, for the first model based on the first statistical variation." In the same field of endeavor, Hetherington teaches "determining a first time period for a first model of the first plurality of statistical routines" by selecting the size of a window over which a window-based statistical feature is computed ( Hetherington , Abstract: "a set of algorithms for selecting a number and size of window based statistical features to use as input features"; Hetherington , ¶ [0008]: "the framework automatically generates multiple time-series features by constructing window based statistical features (e.g. moving average) of original time-series, as inputs to machine learning or deep learning models"). The size of the window is the recited first time period, and the window-based statistical-feature algorithm to which that window size applies is the recited first model of the first plurality of statistical routines, because the window size defines the temporal span over which that statistical routine operates. Hetherington further teaches "determining a first statistical variation for the first model over the first time period" by computing, over that window, a moving window-based statistical function -- for example, a moving average -- of the time-series data ( Hetherington , ¶ [0008]; Hetherington , ¶ [0048]: "the benefits of using moving window based statistical functions"). The moving window-based statistical function computed over the window is the recited first statistical variation for the first model over the first time period. Hetherington further teaches "determining a respective output, of the first plurality of respective outputs, for the first model based on the first statistical variation" because the value of the window-based statistical feature is the resulting feature produced by that statistical routine, and the framework selects the window-based feature values that "yield best scores in terms of prediction accuracy" ( Hetherington , Abstract). The window-based statistical-feature value computed for the first model is the recited respective output of the first plurality of respective outputs, determined based on the first statistical variation. Ahuja , Rossi , and Hetherington are analogous to the claimed invention as all three are from the same field of endeavor of automated machine learning for time-series forecasting in which statistical descriptors computed from an input time-series dataset are used to inform model selection and hyperparameter configuration. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to determine, for a first model of the first plurality of statistical routines, a first time period and a first statistical variation over that first time period, and to determine the respective output based on the first statistical variation, as taught by Hetherington , within the combined system of Ahuja and Rossi . The motivation to combine Ahuja , Rossi , and Hetherington is as recited by Hetherington , which teaches that selecting particular values for the number and size of window-based statistical features "yield[s] best scores in terms of prediction accuracy" ( Hetherington , Abstract) and that the use of "moving window based statistical functions" provides recognized benefits in time-series analysis ( Hetherington , ¶ [0048]); a person of ordinary skill would therefore have applied Hetherington 's window-based statistical-feature determination within the combined Ahuja – Rossi pipeline to improve the predictive accuracy of the resulting tuned forecasting model. Ahuja and Hetherington are further both assigned to Oracle International Corporation, providing an express expectation that their teachings would be combined. As to dependent Claim 4, Ahuja teaches the additional limitations of (i) "comparing a first respective output of the first plurality of respective outputs to a threshold value," (ii) "determining a difference between the first respective output and the threshold value," and (iii) "selecting the first untuned hyperparameter is based on the difference." Ahuja 's pipeline scores each forecasting algorithm with a fitness score and compares those scores directly against one another to identify the most accurate algorithm: "Appropriately selected cross validation strategy and fitness scoring metric enables different forecasting methods with various hyperparameter configurations to be compared directly (i.e. apples-to-apples comparison)" ( Ahuja , ¶ [0025]), and "Validation provides a respective fitness score that measures the respective accuracy of each of proxy models 150" ( Ahuja , ¶ [0053]). The acceptance/best-performance level against which each algorithm's fitness is measured is the threshold value, and comparing each algorithm's fitness score against that threshold yields the corresponding difference. Ahuja then "reliably select[s] the best ML algorithm and hyperparameters configuration for a given timeseries" ( Ahuja , ¶ [0025]; Ahuja , ¶ [0054]: identifying the proxy model with the highest fitness as the selected algorithm). Selecting the most-accurate algorithm -- i.e., the one whose fitness score reflects the smallest difference from the best -- to proceed to hyperparameter tuning is the recited selection based on the difference. As to dependent Claim 5, Ahuja teaches the additional limitations of (i) "filtering a first plurality of untuned hyperparameters based on the first aggregate statistical profile to generate a filtered subset of the first plurality of untuned hyperparameters" and (ii) "selecting the first untuned hyperparameter from the filtered subset." Ahuja expressly filters the hyperparameter search space on the basis of the temporal statistics: the pipeline "Inform[s] the hyperparameters tuning stage to reduce tuning range(s) for seasonal periodicity and differencing" ( Ahuja , ¶ [0022]). "Reduce tuning range(s)" maps directly to filtering the first plurality of untuned hyperparameters based on the aggregate statistical profile (the temporal statistics), and the reduced ranges constitute the filtered subset. Ahuja 's asymmetric investment of computer resources "in the tuning and training of the most promising ML algorithm(s)" ( Ahuja , Abstract) further confirms that the hyperparameters of the less-promising algorithms are filtered out of the tuning step. Selecting the first untuned hyperparameter from the reduced tuning ranges/most-promising algorithm is the recited selection from the filtered subset. As to dependent Claim 13, Ahuja teaches the additional limitations of (i) "determining respective processing power requirements for training for each of a first plurality of untuned hyperparameters of the first plurality of untrained models based on the first aggregate statistical profile" and (ii) "selecting the first untuned hyperparameter based on the respective processing power requirements for training." Ahuja 's pipeline expressly invests computer resources asymmetrically on the basis of the temporal statistics: "efficiency is achieved by asymmetry of investment of computer resources in the tuning and training of the most promising ML algorithm(s)" ( Ahuja , Abstract), and the per-algorithm goodness-of-fit evaluation that drives this asymmetric investment is itself based on the temporal statistics ( Ahuja , ¶ [0017]). The respective level of computer-resource investment per algorithm/hyperparameter configuration is the recited respective processing-power requirement for training, and it is determined from the dataset's temporal statistics (the aggregate statistical profile). Ahuja then selects the most-promising algorithm's hyperparameters for tuning while spending fewer resources on the others ( Ahuja , Abstract: "consumes fewer computer resources to achieve a given accuracy") -- the recited selection based on the respective processing-power requirements. As to independent Claim 15, Claim 15 is directed to one or more non-transitory, computer-readable mediums comprising instructions that when executed cause the same operations recited in Claim 2. Rossi expressly discloses the non-transitory computer-readable medium implementation: "Some examples of the method, apparatus, non-transitory computer readable medium, and system…" ( Rossi , ¶ [0068]); and its meta-learning apparatus 200 comprises a processor 205 and memory 210 ( Rossi , ¶ [0037]; FIG. 2), the latter being a non-transitory computer-readable medium that stores the executable instructions of the pipeline. Ahuja likewise discloses a computer that hosts and operates the ML pipeline as executable software ( Ahuja , ¶¶ [0031], [0032]). Claim 15 is therefore rejected on the same grounds as Claims 1 and 2, with Rossi further supplying the recited non-transitory computer-readable medium implementation. As to Claim 16, Claim 16 is the non-transitory computer-readable medium counterpart of Claim 3 and is rejected on the same grounds as Claim 3. As to Claim 17, Claim 17 is the non-transitory computer-readable medium counterpart of Claim 4 and is rejected on the same grounds as Claim 4. As to Claim 18, Claim 18 is the non-transitory computer-readable medium counterpart of Claim 5 and is rejected on the same grounds as Claim 5. As to Claim 20, Claim 20 is the non-transitory computer-readable medium counterpart of Claim 13 and is rejected on the same grounds as Claim 13 . 07-21-aia AIA Claim s 6–8, 12, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Ahuja , in view of Rossi , in view of Hetherington , and further in view of Vu et al. (Vu) , U.S. Patent Application Publication No. US 2022/0327058 A1 . As to dependent Claim 6 , which recites "wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises," the combination of Ahuja , Rossi , and Hetherington does not teach: "filtering a first plurality of untuned hyperparameters based on an age of the first dataset to generate a filtered subset of the first plurality of untuned hyperparameters" ; and "selecting the first untuned hyperparameter from the filtered subset." In the same field of endeavor, Vu teaches "filtering a first plurality of untuned hyperparameters based on an age of the first dataset to generate a filtered subset of the first plurality of untuned hyperparameters" by determining a data-allocation size for the time-series data from the dataset's characteristics and allocating the data, in sequential (time/age) order, to candidate machine-learning pipelines ( Vu , claim 1: "determining a data allocation size of time series data based on one or more characteristics of a time series data set"; Vu , Abstract: "A data allocation size of time series data may be determined based on one or more characteristics of a time series data set. The time series data may be allocated for use by candidate machine learning pipelines based on the data allocation size"; Vu , detailed description: "a sequential order of the time series data set is used while allocating the time series data based on the data allocation size"; Vu , detailed description: the time series "is sequential; its order cannot be randomized"). The sequential order of the time-series data is its temporal, i.e., age, ordering, and the data-allocation size determines which age-ordered portion of the first dataset is committed to each candidate machine-learning pipeline. Because each candidate machine-learning pipeline carries its own untuned hyperparameters, allocating only a sequentially (age-)selected subset of the first dataset to each pipeline filters the first plurality of untuned hyperparameters based on the age of the first dataset, thereby generating the recited filtered subset. Vu further teaches "selecting the first untuned hyperparameter from the filtered subset" by identifying "a holdout data set, a test data set, and a training data set" based on the data-allocation size ( Vu , Abstract) and selecting from among the candidate machine-learning pipelines evaluated on that age-allocated data. The untuned hyperparameter of the candidate pipeline selected from the age-filtered allocation is the recited first untuned hyperparameter selected from the filtered subset. Ahuja , Rossi , Hetherington , and Vu are analogous to the claimed invention as all four are from the same field of endeavor of automated machine learning for time-series forecasting in which candidate forecasting pipelines and their hyperparameters are selected on the basis of characteristics computed from an input time-series dataset. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the combined system of Ahuja , Rossi , and Hetherington to filter the first plurality of untuned hyperparameters based on the age of the first dataset, as taught by Vu . The motivation to combine Ahuja , Rossi , Hetherington , and Vu is as recited by Vu , which teaches that determining the data-allocation size from the time-series characteristics and allocating the data in sequential order improves the efficiency of pipeline evaluation by limiting computational expenditure to the portion of the dataset most relevant to the forecasting problem ( Vu , Abstract); a person of ordinary skill would therefore have applied Vu 's age-ordered data-allocation filtering within the combined Ahuja – Rossi – Hetherington pipeline to further reduce development time, directly furthering Ahuja 's express objective of consuming "fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract). As to dependent Claim 7 , which recites "wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises," the combination of Ahuja , Rossi , and Hetherington does not teach: "filtering a first plurality of untuned hyperparameters based on a reliability of the first dataset to generate a filtered subset of the first plurality of untuned hyperparameters" ; and "selecting the first untuned hyperparameter from the filtered subset." In the same field of endeavor, Vu teaches "filtering a first plurality of untuned hyperparameters based on a reliability of the first dataset to generate a filtered subset of the first plurality of untuned hyperparameters" by evaluating the predictions of each candidate machine-learning pipeline against held-out data and automatically generating a ranked list of the pipelines on the basis of those evaluations ( Vu , claim 1: "evaluating predictions of each of the one or more candidate machine learning pipelines using at least the one or more features; and automatically generating a ranked list of machine learning pipelines from the one or more candidate machine learning pipelines for time series forecasting based upon evaluating predictions of each of the one or more candidate machine learning pipelines"; Vu , Abstract: "A holdout data set, a test data set, and a training data set may be identified" based on the data-allocation size). Evaluating a candidate pipeline's predictions against a held-out portion of the first dataset is the mechanism by which the reliability of the first dataset for that candidate hyperparameter configuration is quantified: a pipeline whose predictions hold up against the held-out data is reliable on the first dataset, whereas a pipeline whose predictions degrade against the held-out data is not. Because each candidate machine-learning pipeline carries its own untuned hyperparameters, restricting the candidates to those whose predictions are reliable against the held-out data filters the first plurality of untuned hyperparameters based on the reliability of the first dataset, thereby generating the recited filtered subset. Vu further teaches "selecting the first untuned hyperparameter from the filtered subset" by selecting the highest-ranked candidate pipeline from the automatically generated ranked list ( Vu , claim 1; Vu , Abstract). The untuned hyperparameter of the highest-ranked, reliability-vetted pipeline is the recited first untuned hyperparameter selected from the filtered subset. Ahuja , Rossi , Hetherington , and Vu are analogous to the claimed invention as all four are from the same field of endeavor of automated machine learning for time-series forecasting in which candidate forecasting pipelines and their hyperparameters are selected on the basis of characteristics computed from an input time-series dataset. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the combined system of Ahuja , Rossi , and Hetherington to filter the first plurality of untuned hyperparameters based on the reliability of the first dataset, as taught by Vu . The motivation to combine Ahuja , Rossi , Hetherington , and Vu is as recited by Vu , which teaches that evaluating each candidate pipeline's predictions against held-out data and ranking the pipelines accordingly yields a reliable, accuracy-driven ordering of candidates for time-series forecasting ( Vu , claim 1; Vu , Abstract); a person of ordinary skill would therefore have applied Vu 's holdout-based reliability evaluation within the combined Ahuja – Rossi – Hetherington pipeline to discard unreliable candidate hyperparameter configurations before the resource-intensive tuning stage, directly furthering Ahuja 's express objective of consuming "fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract). As to dependent Claim 8 , which recites "wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises," the combination of Ahuja , Rossi , and Hetherington does not teach: "ranking a first plurality of untuned hyperparameters based on the first aggregate statistical profile to generate a ranked order of the first plurality of untuned hyperparameters" ; and "selecting the first untuned hyperparameter based on the ranked order." In the same field of endeavor, Vu teaches "ranking a first plurality of untuned hyperparameters based on the first aggregate statistical profile to generate a ranked order of the first plurality of untuned hyperparameters" by automatically generating a ranked list of candidate machine-learning pipelines on the basis of cached features computed from the time-series data ( Vu , claim 1: "executing the machine learning logic to determine and cache one or more features for the time series data by the one or more candidate machine learning pipelines; evaluating predictions of each of the one or more candidate machine learning pipelines using at least the one or more features; and automatically generating a ranked list of machine learning pipelines from the one or more candidate machine learning pipelines for time series forecasting based upon evaluating predictions of each of the one or more candidate machine learning pipelines"; Vu , Abstract: "A ranked list of machine learning pipelines may be automatically generated from the candidate machine learning pipelines for time series forecasting based upon evaluating predictions of each of the one or more candidate machine learning pipelines"). The cached features computed from the time-series data are the recited first aggregate statistical profile, and because each candidate machine-learning pipeline carries its own untuned hyperparameters, the automatically generated ranked list of pipelines is a ranked order of the corresponding first plurality of untuned hyperparameters generated on the basis of the first aggregate statistical profile. Vu further teaches "selecting the first untuned hyperparameter based on the ranked order" by selecting the highest-ranked candidate pipeline from the automatically generated ranked list ( Vu , claim 1; Vu , Abstract). The untuned hyperparameter of the highest-ranked pipeline is the recited first untuned hyperparameter selected based on the ranked order. Ahuja , Rossi , Hetherington , and Vu are analogous to the claimed invention as all four are from the same field of endeavor of automated machine learning for time-series forecasting in which candidate forecasting pipelines and their hyperparameters are selected on the basis of characteristics computed from an input time-series dataset. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the combined system of Ahuja , Rossi , and Hetherington to rank the first plurality of untuned hyperparameters based on the first aggregate statistical profile and to select on the basis of the ranked order, as taught by Vu . The motivation to combine Ahuja , Rossi , Hetherington , and Vu is as recited by Vu , which teaches that automatically generating a ranked list of candidate pipelines from the cached time-series features produces an explicit, accuracy-driven ordering of candidates from which the best can be deterministically selected ( Vu , claim 1; Vu , Abstract); a person of ordinary skill would therefore have applied Vu 's ranking within the combined Ahuja – Rossi – Hetherington pipeline to make the automated selection step transparent and reproducible, furthering Ahuja 's objective of an efficient, automated forecasting pipeline that consumes "fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract). As to dependent Claim 12 , which recites "wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises," the combination of Ahuja , Rossi , and Hetherington does not teach: "determining respective sample size requirements for training for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile" ; and "selecting the first untuned hyperparameter based on the respective sample size requirements for training." In the same field of endeavor, Vu teaches "determining respective sample size requirements for training for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile" by determining a data-allocation size for the time-series data from the dataset's characteristics and allocating that quantity of data to the candidate machine-learning pipelines for training ( Vu , claim 1: "determining a data allocation size of time series data based on one or more characteristics of a time series data set"; Vu , Abstract: "A data allocation size of time series data may be determined based on one or more characteristics of a time series data set. The time series data may be allocated for use by candidate machine learning pipelines based on the data allocation size"; Vu , Abstract: "A holdout data set, a test data set, and a training data set may be identified" based on the data-allocation size). The data-allocation size is the quantity, i.e., sample size, of training data committed to a candidate machine-learning pipeline, and it is determined from the characteristics of the time-series dataset, which are the recited first aggregate statistical profile. Because each candidate machine-learning pipeline carries its own untuned hyperparameters, the data-allocation size determined for each pipeline is the recited respective sample size requirement for training for each of the first plurality of untuned hyperparameters, determined based on the first aggregate statistical profile. Vu further teaches "selecting the first untuned hyperparameter based on the respective sample size requirements for training" by allocating the determined quantity of data to each candidate pipeline, evaluating the pipelines on the data allocated in accordance with the data-allocation size, and selecting from among them ( Vu , Abstract; Vu , claim 1: "evaluating predictions of each of the one or more candidate machine learning pipelines … and automatically generating a ranked list of machine learning pipelines"). The untuned hyperparameter of the candidate pipeline selected on the basis of its data-allocation-size-governed evaluation is the recited first untuned hyperparameter selected based on the respective sample size requirements for training. Ahuja , Rossi , Hetherington , and Vu are analogous to the claimed invention as all four are from the same field of endeavor of automated machine learning for time-series forecasting in which candidate forecasting pipelines and their hyperparameters are selected on the basis of characteristics computed from an input time-series dataset. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the combined system of Ahuja , Rossi , and Hetherington to determine respective sample size requirements for training for each of the first plurality of untuned hyperparameters and to select on the basis of those requirements, as taught by Vu . The motivation to combine Ahuja , Rossi , Hetherington , and Vu is as recited by Vu , which teaches that determining the data-allocation size from the time-series characteristics and allocating training data accordingly enables efficient pipeline evaluation by committing only the necessary quantity of data to each candidate ( Vu , Abstract); a person of ordinary skill would therefore have applied Vu 's data-allocation-size determination within the combined Ahuja – Rossi – Hetherington pipeline to avoid over-committing training data to candidate hyperparameter configurations, directly furthering Ahuja 's express objective of consuming "fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract). As to Claim 19, Claim 19 is the non-transitory computer-readable medium counterpart of Claim 6 and is rejected on the same grounds as Claim 6 . 07-21-aia AIA Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Ahuja , and Rossi , in view of Hetherington , and further in view of Yakovlev et al. (Yakovlev) , U.S. Patent No. US 11,429,895 B2 . As to dependent Claim 9 , which recites "wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises," the combination of Ahuja , Rossi , and Hetherington does not teach: "determining respective training time predictions for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile" ; and "selecting the first untuned hyperparameter based on the respective training time predictions." In the same field of endeavor, Yakovlev teaches "determining respective training time predictions for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile" by training a regressor that predicts the time needed to train a machine-learning model for each of a plurality of hyperparameter configurations as a function of the dataset ( Yakovlev , Title: "Predicting Machine Learning or Deep Learning Model Training Time"; Yakovlev , Abstract: "techniques for exploring hyperparameters of a machine learning model (MLM) and to train a regressor to predict a time needed to train the MLM based on a hyperparameter configuration"; Yakovlev , Abstract: "the regressor predicts time needed to train the MLM based on … training durations and hyperparameter values of landmark configurations of the MLM," and "to predict a training time for the MLM for new datasets and/or new hyperparameter configurations"). Because the regressor produces a training-time prediction "for … new hyperparameter configurations" and does so "for new datasets," and because the recited first aggregate statistical profile is the set of statistical descriptors that characterizes the first dataset, Yakovlev determines a respective training-time prediction for each of the first plurality of untuned hyperparameters based on the first aggregate statistical profile. Yakovlev further teaches "selecting the first untuned hyperparameter based on the respective training time predictions" because the predicted training times are produced precisely so that the automated machine-learning process can choose among hyperparameter configurations on the basis of their predicted training cost ( Yakovlev , Abstract: predicting "a training time for the MLM for new datasets and/or new hyperparameter configurations"); in the combination, the hyperparameter-selection step of Ahuja ( Ahuja , ¶¶ [0017], [0031]) is performed on the basis of Yakovlev 's respective training-time predictions, so that the first untuned hyperparameter is selected based on those predictions. Ahuja , Rossi , Hetherington , and Yakovlev are analogous to the claimed invention as all four are from the same field of endeavor of automated machine-learning configuration in which the hyperparameters of candidate models are selected based on properties computed from the input dataset. Ahuja , Hetherington , and Yakovlev are further co-assigned to Oracle International Corporation and share inventors (e.g., Yakovlev, Varadarajan, Agrawal, Moghadam, Idicula, Agarwal) of the same line of Oracle AutoML research, providing an express expectation that their teachings would be combined. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the combined system of Ahuja , Rossi , and Hetherington to determine respective training-time predictions for each of the first plurality of untuned hyperparameters and to select on the basis of those predictions, as taught by Yakovlev . The motivation to combine Ahuja , Rossi , Hetherington , and Yakovlev is as recited by Yakovlev , which teaches predicting model training time so that expensive hyperparameter configurations can be identified in advance ( Yakovlev , Title; Yakovlev , Abstract); a person of ordinary skill would therefore have applied Yakovlev 's training-time prediction within the combined Ahuja – Rossi – Hetherington pipeline to avoid committing computational resources to hyperparameter configurations predicted to be excessively expensive to train, directly furthering Ahuja 's express objective of consuming "fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract) . 07-21-aia AIA Claim s 10, 11, and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Ahuja , and Rossi , in view of Hetherington , and further in view of Moghadam et al. Moghadam , U.S. Patent No. US 11,620,568 B2 . As to dependent Claim 10 , which recites "wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises," the combination of Ahuja , Rossi , and Hetherington does not teach: "determining respective performance predictions for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile" ; and "selecting the first untuned hyperparameter based on the respective performance predictions." In the same field of endeavor, Moghadam teaches "determining respective performance predictions for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile" by predicting, through hyperparameter predictors, a performance prediction for each candidate machine-learning model and its hyperparameter settings from the dataset's meta-features ( Moghadam , Title: "Using hyperparameter predictors to improve accuracy of automatic machine learning model selection"; Moghadam , col. 2, ll. 18-22 (Detailed Description, General Overview): "selection of machine learning algorithms based on performance predictions by using hyperparameter predictors"; Moghadam , Abstract (front page): "for each mini-machine learning model (MML model), a respective hyperparameter predictor set that predicts a respective set of predicted hyperparameter settings"; Moghadam , col. 2, ll. 28-32 (General Overview): "The trained mini-models use hyper-parameters predicted by hyperparameter predictors based on meta-features of the data set"). The dataset meta-features are the recited first aggregate statistical profile, and the per-model performance predictions produced by the hyperparameter predictors for each model's predicted hyperparameter settings are the recited respective performance predictions for each of the first plurality of untuned hyperparameters, determined based on the first aggregate statistical profile. Moghadam further teaches "selecting the first untuned hyperparameter based on the respective performance predictions" because it discloses a "computer that optimally selects trainable algorithms based on performance predictions by using hyperparameter predictors" ( Moghadam , col. 1 ll. 52-56 (Brief Description of Drawing, FIG.1) and uses the predicted scores "to more efficiently and accurately rank and select the best algorithm for the given dataset" ( Moghadam , col. 2, ll. 66-68 (General Overview)). Selecting the algorithm -- and accordingly its untuned hyperparameter -- on the basis of the respective performance predictions is the recited selection based on the respective performance predictions. Ahuja , Rossi , Hetherington , and Moghadam are analogous to the claimed invention as all four are from the same field of endeavor of automated machine-learning configuration in which models or hyperparameters are selected based on predictions derived from dataset statistics. Ahuja , Hetherington , and Moghadam are further co-assigned to Oracle International Corporation and share inventors (e.g., Moghadam, Agrawal, Varadarajan, Yakovlev, Idicula, Agarwal) of the same line of Oracle AutoML research, providing an express expectation that their teachings would be combined. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the combined system of Ahuja , Rossi , and Hetherington to determine respective performance predictions for each of the first plurality of untuned hyperparameters and to select on the basis of those predictions, as taught by Moghadam . The motivation to combine Ahuja , Rossi , Hetherington , and Moghadam is as recited by Moghadam , which teaches that selecting machine-learning algorithms "based on performance predictions by using hyperparameter predictors" improves the accuracy of automatic model selection ( Moghadam , Title; Moghadam , col. 1 ll. 52-56 (Brief Description of Drawing, FIG.1)); a person of ordinary skill would therefore have applied Moghadam 's performance-prediction-based selection within the combined Ahuja – Rossi – Hetherington pipeline to commit tuning resources to the hyperparameter configurations predicted to perform best, directly furthering Ahuja 's express objective of consuming "fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract). As to dependent Claim 11 , which recites "wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises," the combination of Ahuja , Rossi , and Hetherington does not teach: "determining respective predictions for a number of hyperparameters requiring training for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile" ; and "selecting the first untuned hyperparameter based on the respective predictions for the number of hyperparameters requiring training." In the same field of endeavor, Moghadam teaches "determining respective predictions for a number of hyperparameters requiring training for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile" by training, for each of a plurality of machine-learning models, a hyperparameter predictor set that predicts a respective set of hyperparameter settings from the dataset's meta-features ( Moghadam , Title: "Using hyperparameter predictors to improve accuracy of automatic machine learning model selection"; Moghadam , claim 1 / Abstract (front page): "for each mini-machine learning model (MML model), a respective hyperparameter predictor set that predicts a respective set of predicted hyperparameter settings"; Moghadam , col. 2, ll. 35-40 (General Overview): "for a given mini-model and for each hypermeter of the given model, a hyperparameter predictor is trained to predicate an optimal hyperparameter setting, using as training input meta-features of a data set sample"; Moghadam , col. 2, ll. 28-32 (General Overview): "The trained mini-models use hyper-parameters predicted by hyperparameter predictors based on meta-features of the data set"). The dataset meta-features are the recited first aggregate statistical profile, and the respective set of predicted hyperparameter settings for each model identifies which, and therefore how many, hyperparameters of that model are predicted to require training. Moghadam thus determines, for each of the first plurality of untuned hyperparameters, a respective prediction for the number of hyperparameters requiring training, based on the first aggregate statistical profile. Moghadam further teaches "selecting the first untuned hyperparameter based on the respective predictions for the number of hyperparameters requiring training" because the per-model predicted hyperparameter sets are used "to more efficiently and accurately rank and select the best algorithm for the given dataset" ( Moghadam , col. 2, ll. 66-68 (General Overview)). Selecting the algorithm -- and accordingly its hyperparameter -- on the basis of those per-model predicted hyperparameter sets is the recited selection based on the respective predictions for the number of hyperparameters requiring training. Ahuja , Rossi , Hetherington , and Moghadam are analogous to the claimed invention as all four are from the same field of endeavor of automated machine-learning configuration in which models or hyperparameters are selected based on predictions derived from dataset statistics. Ahuja , Hetherington , and Moghadam are further co-assigned to Oracle International Corporation and share inventors (e.g., Moghadam, Agrawal, Varadarajan, Yakovlev, Idicula, Agarwal) of the same line of Oracle AutoML research, providing an express expectation that their teachings would be combined. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to modify the combined system of Ahuja , Rossi , and Hetherington to determine, for each of the first plurality of untuned hyperparameters, a respective prediction for the number of hyperparameters requiring training and to select on the basis of those predictions, as taught by Moghadam . The motivation to combine Ahuja , Rossi , Hetherington , and Moghadam is as recited by Moghadam , which teaches that hyperparameter-predictor-based prediction of per-model hyperparameter settings is used "to more efficiently and accurately rank and select the best algorithm for the given dataset" ( Moghadam , col. 2, ll. 66-68 (General Overview)); a person of ordinary skill would therefore have applied Moghadam 's per-model hyperparameter-set prediction within the combined Ahuja – Rossi – Hetherington pipeline to avoid committing tuning resources to models requiring an impractical number of hyperparameters to be trained, directly furthering Ahuja 's express objective of consuming "fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract). As to dependent Claim 14 , which recites "wherein determining the first aggregate statistical profile for the first dataset based on the first plurality of respective outputs further comprises," Rossi teaches "generating a profile matrix for the first dataset" by forming the per-model performance data as a tensor ( Rossi , ¶ [0058]: "Performance data 315 may be formed as a tensor"; Rossi , ¶ [0055]: "performance data which measures the performance of a plurality of forecasting models as applied to" the time-series data). The per-model performance tensor derived from the first dataset is the recited profile matrix for the first dataset. The combination of Ahuja , Rossi , and Hetherington does not teach: "populating values of the profile matrix based on a comparison of the first plurality of respective outputs and respective model requirements for the first plurality of untrained models." In the same field of endeavor, Moghadam teaches "populating values of the profile matrix based on a comparison of the first plurality of respective outputs and respective model requirements for the first plurality of untrained models" by computing, for each candidate model, a score derived from a comparison of the dataset's meta-features against that model's predicted hyperparameter settings ( Moghadam , col. 2, ll. 35-40 (General Overview): "for a given mini-model and for each hypermeter of the given model, a hyperparameter predictor is trained to predicate an optimal hyperparameter setting, using as training input meta-features of a data set sample"; Moghadam , Abstract (front page): "for each mini-machine learning model (MML model), a respective hyperparameter predictor set that predicts a respective set of predicted hyperparameter settings"; Moghadam , detailed description: the hyperparameter predictors "improve the accuracy of scores of each mini-model"). The dataset meta-features are the recited first plurality of respective outputs, the per-model predicted hyperparameter settings are the recited respective model requirements for the first plurality of untrained models, and the resulting per-model scores computed from the comparison of the two are the recited values populated into the profile matrix. Ahuja , Rossi , Hetherington , and Moghadam are analogous to the claimed invention as all four are from the same field of endeavor of automated machine-learning configuration in which models or hyperparameters are selected based on descriptors and predictions derived from dataset statistics. Ahuja , Hetherington , and Moghadam are further co-assigned to Oracle International Corporation and share inventors (e.g., Moghadam, Agrawal, Varadarajan, Yakovlev, Idicula, Agarwal) of the same line of Oracle AutoML research, providing an express expectation that their teachings would be combined. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to populate the values of Rossi 's profile matrix based on a comparison of the first plurality of respective outputs and the respective model requirements for the first plurality of untrained models, as taught by Moghadam . The motivation to combine Ahuja , Rossi , Hetherington , and Moghadam is as recited by Moghadam , which teaches that comparing dataset meta-features against per-model predicted hyperparameter settings "improves the accuracy of scores of each mini-model" and thereby supports more efficient and accurate selection of the best algorithm for the given dataset ( Moghadam , Summary); a person of ordinary skill would therefore have populated the profile matrix of the combined Ahuja – Rossi – Hetherington pipeline by Moghadam 's comparison to improve the accuracy of the per-model values driving selection, furthering Ahuja 's objective of an efficient automated forecasting pipeline that consumes "fewer computer resources to achieve a given accuracy" ( Ahuja , Abstract). Conclusion Any inquiry concerning this communication or earlier communications from the examiner should be directed to HUNG VAN LE whose telephone number is (571)270-0164. The examiner can normally be reached 8 a.m. - 5 p.m.. 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, Cesar Paula can be reached at (571) 272-4128. 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. /HUNG VAN LE/Examiner, Art Unit 2145 /CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145 Application/Control Number: 18/498,406 Page 2 Art Unit: 2145 Application/Control Number: 18/498,406 Page 3 Art Unit: 2145 Application/Control Number: 18/498,406 Page 4 Art Unit: 2145 Application/Control Number: 18/498,406 Page 5 Art Unit: 2145 Application/Control Number: 18/498,406 Page 6 Art Unit: 2145 Application/Control Number: 18/498,406 Page 7 Art Unit: 2145 Application/Control Number: 18/498,406 Page 8 Art Unit: 2145 Application/Control Number: 18/498,406 Page 9 Art Unit: 2145 Application/Control Number: 18/498,406 Page 10 Art Unit: 2145 Application/Control Number: 18/498,406 Page 11 Art Unit: 2145 Application/Control Number: 18/498,406 Page 12 Art Unit: 2145 Application/Control Number: 18/498,406 Page 13 Art Unit: 2145 Application/Control Number: 18/498,406 Page 14 Art Unit: 2145 Application/Control Number: 18/498,406 Page 15 Art Unit: 2145 Application/Control Number: 18/498,406 Page 16 Art Unit: 2145 Application/Control Number: 18/498,406 Page 17 Art Unit: 2145 Application/Control Number: 18/498,406 Page 18 Art Unit: 2145 Application/Control Number: 18/498,406 Page 19 Art Unit: 2145 Application/Control Number: 18/498,406 Page 20 Art Unit: 2145 Application/Control Number: 18/498,406 Page 21 Art Unit: 2145 Application/Control Number: 18/498,406 Page 22 Art Unit: 2145 Application/Control Number: 18/498,406 Page 23 Art Unit: 2145 Application/Control Number: 18/498,406 Page 24 Art Unit: 2145 Application/Control Number: 18/498,406 Page 25 Art Unit: 2145 Application/Control Number: 18/498,406 Page 26 Art Unit: 2145 Application/Control Number: 18/498,406 Page 27 Art Unit: 2145 Application/Control Number: 18/498,406 Page 28 Art Unit: 2145 Application/Control Number: 18/498,406 Page 29 Art Unit: 2145 Application/Control Number: 18/498,406 Page 30 Art Unit: 2145 Application/Control Number: 18/498,406 Page 31 Art Unit: 2145 Application/Control Number: 18/498,406 Page 32 Art Unit: 2145 Application/Control Number: 18/498,406 Page 33 Art Unit: 2145 Application/Control Number: 18/498,406 Page 34 Art Unit: 2145 Application/Control Number: 18/498,406 Page 35 Art Unit: 2145 Application/Control Number: 18/498,406 Page 36 Art Unit: 2145
Prosecution Insights