DETAILED ACTION
The action is in response to the original filing on October 31, 2023 and the Remarks and Amendments filed on July 30, 2026. Claims 1-20 are pending and have been considered below. Claims 1, 2 and 15 are independent claims. Claims 1, 2, 5-12, 15, and 18-20 are amended.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to a judicial exception (an abstract idea) without reciting significantly more.
Regarding independent claims 1, 2, and 15
Step 1 -- whether the claim falls within any statutory category. See MPEP 2106.03.
Claim 1 is drawn to a system claim; claim 2 is drawn to a method claim; claim 15 is drawn to a non-transitory computer-readable medium claim. Claim 1 falls within the machine category because it recites one or more processors and one or more non-transitory, computer-readable mediums comprising instructions. Claim 2 falls within the process category because it recites a series of steps. Claim 15 falls within the manufacture category because it recites one or more non-transitory, computer-readable mediums comprising instructions. Therefore, each of these claims falls under one of the four categories of statutory subject matter (process/method, machine/product/apparatus, manufacture, or composition of matter).
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding independent claim 2, the claim is directed to a method for minimizing development time in artificial intelligence models by automating hyperparameter selection based on dataset fittings of time-series data, the method comprising:
The limitations of “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,” “selecting, based on the first aggregate statistical profile, a first untuned hyperparameter, for a first plurality of untrained models, for tuning to a specific value by: filtering a first plurality of untuned hyperparameters based on a characteristic 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,” and “tuning the first untuned hyperparameter to the specific value” recite an abstract idea.
These limitations recite mathematical concepts under MPEP 2106.04(a)(2), subsection I. Under the broadest reasonable interpretation, each recited “statistical analysis” is a mathematical calculation. The specification describes the statistical analysis as including “measures like mean, median, mode, standard deviation, variance, and percentiles” (specification ¶ [0036]). The “first aggregate statistical profile” is an “array of data values” (specification ¶ [0036]) that “may comprise a matrix” (specification ¶ [0083]). It therefore organizes and manipulates information through mathematical correlations.
These limitations also recite mental processes under MPEP 2106.04(a)(2), subsection III. Deriving a feature input from a dataset, aggregating statistical outputs into a profile, filtering a list of candidate hyperparameters according to a characteristic of the dataset, selecting a hyperparameter from the remaining candidates, and assigning that hyperparameter a specific value are observations, evaluations, judgments, and opinions. Each of them can practically be performed in the human mind, or by a human using pen and paper. For example, a data scientist can review a dataset, compute its mean and variance, note that the data is seasonal, strike from a list the hyperparameters of models unsuited to seasonal data, and choose a value for a remaining hyperparameter. Each limitation is recited at a high level of generality and does not require any operation that cannot practically be performed in the human mind. Performing these steps on a generic computer does not remove them from the mental-processes grouping. See MPEP 2106.04(a)(2), subsection III.C.
This treatment is consistent with the 2024 Subject Matter Eligibility Update, Example 47, claim 2. There, “discretizing,” “detecting,” and “analyzing” were found to recite mental processes, and training an artificial neural network with a backpropagation algorithm and a gradient descent algorithm was found to recite a mathematical concept.
Accordingly, independent claim 2 recites a judicial exception, namely an abstract idea, including mathematical concepts and mental processes under MPEP 2106.04(a)(2), subsections I and III. Therefore, the analysis proceeds to Step 2A Prong 2.
Independent claim 1 is a system claim reciting limitations similar to claim 2 and is directed towards the abstract idea for similar reasons. Claim 1 further recites that 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 recites “selecting, based on the respective effectiveness of the plurality of model types.” Computing a series of values that rates the effectiveness of model types, and selecting on that basis, recites a mathematical relationship and an evaluation. These limitations fall within the mathematical-concepts and mental-processes groupings. See MPEP 2106.04(a)(2), subsections I and III.
Independent claim 15 is a non-transitory computer-readable medium claim reciting limitations similar to claim 2 and is directed towards the abstract idea for similar reasons.
Step 2A Prong 2 -- whether the claim as a whole integrates the recited judicial exception into a practical application of the exception or whether the claim is “directed to” the judicial exception. This evaluation is performed by (1) identifying whether there are any additional elements recited in the claim beyond the judicial exception, and (2) evaluating those additional elements individually and in combination to determine whether the claim as a whole integrates the exception into a practical application. See MPEP 2106.04(d).
Regarding independent claim 2, this claim recites additional elements of “receiving a first dataset” and “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.”
The limitation of “receiving a first dataset” is mere data gathering recited at a high level of generality. It constitutes insignificant extra-solution activity. See MPEP 2106.05(g).
The limitation that the untrained models comprise algorithms for time-series forecasting with default hyperparameter tuning merely identifies the generic models whose hyperparameters are the subject of the abstract selection. This amounts to no more than generally linking the use of the judicial exception to a particular technological environment or field of use, namely time-series forecasting. See MPEP 2106.05(h). To the extent the recited “statistical routines” and “tuning” are considered additional elements, they are claimed at a high level of generality and merely use generic software as a tool to perform the abstract calculations and evaluations. They amount to no more than mere instructions to apply the exception. See MPEP 2106.05(f).
The claim does not recite an improvement to the functioning of a computer or to any other technology or technical field. See MPEP 2106.04(d)(1) and MPEP 2106.05(a). The benefit described in the specification is a less time-consuming process of “choosing the best model to fit a given dataset and optimizing its hyperparameters” (specification ¶ [0005]). That is an improvement to the abstract model-selection and hyperparameter-selection process itself, not an improvement to computer capabilities or to a technological process.
The claim also does not apply the judicial exception with a particular machine, effect a particular transformation, or include other meaningful limitations. See MPEP 2106.05(b), 2106.05(c), and 2106.05(e). In Example 47, claim 3 was found eligible because its additional steps used the results of the analysis to drop malicious network packets and block future traffic. Claim 2 contains no analogous limitation that applies the selected or tuned hyperparameter to effect a technological outcome.
Considered individually and in combination, the additional elements do not integrate the exception into a practical application. Therefore, independent claim 2 is directed to the abstract idea.
Regarding independent claim 1, this claim is drawn to a system claim reciting similar limitations of claim 2 and is rejected under the same rationale. Claim 1 also recites additional elements of:
“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, wherein the first dataset comprises one or more categories of data trends,” and “generating for display, on a user interface, a recommendation for using the tuned first model for time-series forecasting.”
The processors and computer-readable mediums are recited at a high level of generality; the specification describes them as “any suitable processing, storage, and/or I/O circuitry” (specification ¶ [0065]). They merely act as a tool on which the abstract idea operates and amount to no more than mere instructions to apply the exception using a generic computer. See MPEP 2106.05(f).
The receiving limitation is mere data gathering, and the wherein clause merely describes the type of data gathered. Together they constitute insignificant extra-solution activity. See MPEP 2106.05(g).
The limitation of “generating for display, on a user interface, a recommendation” merely outputs the result of the abstract idea. It is insignificant post-solution activity (MPEP 2106.05(g)) that generally links the exception to the field of time-series forecasting (MPEP 2106.05(h)). Therefore, these additional elements fail to integrate the exception into a practical application.
Regarding independent claim 15, this claim is drawn to a computer-readable medium claim reciting similar limitations of claim 2 and is rejected under the same rationale. Claim 15 also recites additional elements of:
“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.”
The computer-readable mediums and processors are generic computer components that merely act as a tool on which the method operates. They amount to no more than mere instructions to apply the exception. See MPEP 2106.05(f). The receiving limitation is insignificant extra-solution activity. See MPEP 2106.05(g). Therefore, these additional elements fail to integrate the exception into a practical application.
Step 2B -- whether the claim amounts to significantly more than the judicial exception. See MPEP 2106.05.
Regarding independent claim 2, the additional elements are “receiving a first dataset” and the untrained time-series forecasting models with default hyperparameter tuning. As explained at Step 2A Prong 2, receiving a dataset is insignificant extra-solution activity. Upon reevaluation at Step 2B, receiving data is a well-understood, routine, and conventional function. See MPEP 2106.05(d), subsection II (receiving or transmitting data over a network). Limiting the exception to time-series forecasting models merely links it to a field of use and does not provide an inventive concept. See MPEP 2106.05(h). Considered individually and as an ordered combination, the additional elements do not amount to significantly more than the judicial exception.
Regarding independent claim 1, to the extent claim 1 recites limitations overlapping with claim 2, those limitations do not amount to significantly more for the same reasons. The additional processors and computer-readable mediums remain mere instructions to apply the exception using generic computer components. See MPEP 2106.05(f). The receiving step and the step of generating a recommendation for display on a user interface, reevaluated at Step 2B, remain insignificant extra-solution activity. They recite generic computer functions of receiving, storing, and presenting data, which are well-understood, routine, and conventional. See MPEP 2106.05(d), subsection II. Considered individually and as an ordered combination, the additional elements of claim 1 do not provide an inventive concept.
Regarding independent claim 15, to the extent claim 15 recites limitations overlapping with claim 2, those limitations do not amount to significantly more for the same reasons. The additional computer-readable mediums and processors are generic computer components performing generic computer functions, which are well-understood, routine, and conventional. See MPEP 2106.05(d), subsection II. Considered individually and as an ordered combination, the additional elements of claim 15 do not provide an inventive concept.
Accordingly, claims 1, 2, and 15 do not recite additional elements, individually or in combination, that amount to significantly more than the recited judicial exception, and are therefore ineligible under 35 U.S.C. § 101.
Regarding dependent claims 3-14 and 16-20
Claims 3-14 and 16-20 merely narrow the previously identified abstract idea limitations recited in independent claims 2 and 15. For the reasons discussed above with respect to independent claims 1, 2, and 15, these judicial exceptions are not meaningfully integrated into a practical application and do not amount to significantly more than the abstract idea. The claims recite mathematical relationships and mental processes that are practically capable of being performed in the human mind with the assistance of pen and paper: determining statistical variations, comparing values to thresholds, filtering, ranking, predicting, and selecting. Therefore, claims 3-14 and 16-20 also recite abstract ideas that do not integrate into a practical application or amount to significantly more than the judicial exception, and are rejected under 35 U.S.C. § 101.
Step 1 -- whether the claim falls within any statutory category. See MPEP 2106.03.
Claims 3-14 are drawn to a method claim, and claims 16-20 are drawn to a non-transitory computer-readable medium claim. Therefore, each of these claims falls under one of the four categories of statutory subject matter (process/method, machine/product/apparatus, manufacture, or composition of matter).
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding claim 3, this claim recites limitations of “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.”
These limitations are directed towards the abstract idea of a mathematical relationship. A statistical variation, such as a variance or standard deviation, computed over a time period is a mathematical calculation. Selecting the time period and determining an output from the variation are evaluations that can be performed manually by a human being. See MPEP 2106.04(a)(2), subsections I and III.
Regarding claim 4, this claim recites limitations of “comparing a first respective output of the first plurality of respective outputs 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.”
These limitations are directed towards the abstract idea of a mental process, or a concept that can be performed in the human mind, including observation, evaluation, judgment, or opinion. They recite that a user can compare a value to a threshold, subtract one from the other, and select a hyperparameter accordingly. Computing a difference is also a mathematical calculation. See MPEP 2106.04(a)(2), subsections I and III.
Regarding claim 5, this claim recites limitation of “filtering the first plurality of untuned hyperparameters based on the first aggregate statistical profile.”
This limitation is directed towards the abstract idea of a mental process, including observation, evaluation, judgment, or opinion. It recites that a user can review a statistical profile and strike candidate hyperparameters from a list. See MPEP 2106.04(a)(2), subsection III.
Regarding claim 6, this claim recites limitations of “determining an age based on the characteristic” and “filtering the first plurality of untuned hyperparameters based on the age of the first dataset.”
These limitations are directed towards the abstract idea of a mental process, including observation, evaluation, judgment, or opinion. They recite that a user can observe how old the data is and strike candidate hyperparameters from a list on that basis. See MPEP 2106.04(a)(2), subsection III.
Regarding claim 7, this claim recites limitations of “determining a reliability based on the characteristic” and “filtering the first plurality of untuned hyperparameters based on the reliability of the first dataset.”
These limitations are directed towards the abstract idea of a mental process, including observation, evaluation, judgment, or opinion. They recite that a user can judge whether data is reliable and strike candidate hyperparameters from a list on that basis. See MPEP 2106.04(a)(2), subsection III.
Regarding claim 8, this claim recites limitations of “ranking the 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.”
These limitations are directed towards the abstract idea of a mental process, including evaluation and judgment. They recite that a user can order candidates and choose the highest-ranked one. Ordering values is also a mathematical relationship. See MPEP 2106.04(a)(2), subsections I and III.
Regarding claim 9, this claim recites limitations of “determining respective training time predictions for each of the 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.”
These limitations are directed towards the abstract idea of a mental process and a mathematical calculation. They recite that a user can estimate how long training would take and choose accordingly. See MPEP 2106.04(a)(2), subsections I and III.
Regarding claim 10, this claim recites limitations of “determining respective performance predictions for each of the first plurality of untuned hyperparameters based on the first aggregate statistical profile” and “selecting the first untuned hyperparameter based on the respective performance predictions.”
These limitations are directed towards the abstract idea of a mental process and a mathematical calculation. They recite that a user can estimate the expected performance of each candidate and choose accordingly. See MPEP 2106.04(a)(2), subsections I and III.
Regarding claim 11, this claim recites limitations of “determining respective predictions for a number of hyperparameters requiring training for each of the 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.”
These limitations are directed towards the abstract idea of a mental process, including evaluation and judgment. They recite that a user can count the hyperparameters each candidate would require to be trained and choose accordingly. See MPEP 2106.04(a)(2), subsection III.
Regarding claim 12, this claim recites limitations of “determining respective sample size requirements for training for each of the 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.”
These limitations are directed towards the abstract idea of a mental process and a mathematical calculation. They recite that a user can estimate the amount of data each candidate needs and choose accordingly. See MPEP 2106.04(a)(2), subsections I and III.
Regarding claim 13, this claim recites limitations of “determining respective processing power 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 processing power requirements for training.”
These limitations are directed towards the abstract idea of a mental process and a mathematical calculation. They recite that a user can estimate the computing cost of each candidate and choose accordingly. See MPEP 2106.04(a)(2), subsections I and III.
Regarding claim 14, this claim recites limitations of “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.”
These limitations are directed towards the abstract idea of a mathematical relationship, specifically organizing information and manipulating information through mathematical correlations. The recited comparison is an evaluation that can also be performed manually by a human being. See MPEP 2106.04(a)(2), subsections I and III.
Regarding claims 16, 17, 19, and 20, these claims recite limitations similar to claims 3, 4, 6, and 13, respectively, and are directed towards the abstract idea for the same reasons.
Regarding claim 18, this claim recites limitations of “filtering the 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 “selecting the first untuned hyperparameter from the filtered subset.”
These limitations are directed towards the abstract idea of a mental process, including observation, evaluation, judgment, or opinion. They recite that a user can strike candidates from a list and choose one of the remaining candidates. See MPEP 2106.04(a)(2), subsection III.
Step 2A Prong 2 -- whether the claim as a whole integrates the recited judicial exception into a practical application of the exception or whether the claim is “directed to” the judicial exception. This evaluation is performed by (1) identifying whether there are any additional elements recited in the claim beyond the judicial exception, and (2) evaluating those additional elements individually and in combination to determine whether the claim as a whole integrates the exception into a practical application. See MPEP 2106.04(d).
Regarding claims 3-14, these claims recite no additional elements beyond those identified for independent claim 2. Each added limitation is itself part of the abstract idea: it specifies how the statistical analysis is performed (claims 3 and 14) or how the hyperparameter is filtered or selected (claims 4-13). The additional elements inherited from claim 2 remain insignificant extra-solution activity and a field-of-use limitation, and thus fail to integrate the exception into a practical application. See MPEP 2106.05(g) and 2106.05(h).
Regarding claims 16-20, these claims recite the additional elements of “one or more non-transitory, computer-readable mediums” and “one or more processors” inherited from claim 15. These limitations amount to no more than generally linking the use of the judicial exception to a particular technological environment (see MPEP 2106.05(h)). They recite generic computer components that merely act as a tool on which the method operates (see MPEP 2106.05(f)), and thus fail to integrate the exception into a practical application.
Step 2B -- whether the claim amounts to significantly more than the judicial exception. See MPEP 2106.05.
Regarding claims 3-14 and 16-20, for the same reasons stated for independent claims 2 and 15, the additional elements, considered individually and as an ordered combination, do not amount to significantly more than the judicial exception. The only additional elements are data gathering, a field-of-use limitation, and generic computer components performing generic computer functions, which are well-understood, routine, and conventional. See MPEP 2106.05(d), subsection II. Therefore, claims 3-14 and 16-20 are ineligible under 35 U.S.C. § 101.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
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.
Claims 1, 2, 3, 4, 5, 14, 15, 16, 17, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Ahuja et al. (Ahuja), US 2023/0153394 A1, in view of Rossi et al. (Rossi), US 2024/0152769 A1, and further in view of Hetherington et al. (Hetherington), 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) teaches this preamble by disclosing a computer that "applies timeseries preprocessing, model selection, and hyperparameter tuning techniques based on temporal statistics 120 of original timeseries 111" (Ahuja, ¶ [0033]), in which the least promising algorithms "receive very few computer resources" so that the pipeline will "consume fewer computer resources to achieve a given accuracy" (Ahuja, ¶ [0032]).
(Ahuja) further teaches:
"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 "one or more general purpose hardware processors" (Ahuja, ¶ [0108]) and instructions "stored in non-transitory storage media accessible to processor 504" (Ahuja, ¶ [0110]).
"receiving a first dataset" — (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]). The original timeseries 111 corresponds to the recited 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" — (Ahuja) teaches a preprocessor that "estimates data seasonality, heteroskedasticity, a required degree of differencing to achieve data stationarity, and the frequency of samples in the timeseries" (Ahuja, ¶ [0020]), deriving "temporal statistics 120 by quantitatively analyzing the tuples in original timeseries 111" (Ahuja, ¶ [0036]), including "a mean, mode, maximum, entropy, or variance" (Ahuja, ¶ [0039]) and "determining the differencing via statistical tests" (Ahuja, ¶ [0098]). These estimations correspond to the first plurality of statistical routines, each performing a respective statistical analysis, and the resulting temporal statistics 120 correspond to the first plurality of respective outputs. The first feature input on which these routines operate is generated by (Rossi), as set forth below. In the combination of (Ahuja) and (Rossi), the feature input generated by (Rossi) is input into the statistical routines of (Ahuja).
"selecting, based on the respective effectiveness of the plurality of model types, a first untuned hyperparameter, for a first plurality of untrained models, for tuning to a specific value by:" — (Ahuja) teaches proxy models validated with a fitness score, where "if proxy model 151 has the highest fitness score, then ML algorithm 131 is selected for further processing" (Ahuja, ¶ [0054]), after which (Ahuja) "tunes hyperparameters of best model 160" (Ahuja, ¶ [0075]). The fitness scores of the proxy models correspond to the respective effectiveness of the plurality of model types.
"filtering a first plurality of untuned hyperparameters based on a characteristic of the first dataset to generate a filtered subset of the first plurality of untuned hyperparameters; and" — (Ahuja, ¶[0074]: “step 209 identifies a respective values subrange for some hyperparameter(s) of best model 160 as a preface to hyperparameters tuning” and “For example if seasonality 122 is weekly, then step 209 may limit a seasonality hyperparameter to a range of 5-10 days or exactly seven days”; ¶ [0036]: “Temporal statistics 120 may include frequency 121 and seasonality 122 that characterize original timeseries 111 as a whole”). Before tuning, the untuned candidate values of the hyperparameters of best model 160 correspond to the first plurality of untuned hyperparameters. Step 209 limits those candidates to a subrange, which corresponds to the filtered subset. Seasonality 122 characterizes original timeseries 111 as a whole and corresponds to the characteristic of the first dataset on which the filtering is based.
"selecting the first untuned hyperparameter from the filtered subset," — (Ahuja) teaches that "based on temporal statistics 120, step 209 identifies a respective values subrange for some hyperparameter(s) of best model 160 as a preface to hyperparameters tuning" (Ahuja, ¶ [0074]), best model 160 being instantiated from the algorithm selected from among the non-excluded algorithms (Ahuja, ¶¶ [0054]–[0055]).
"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) teaches "[a] set of many forecasting algorithms" evaluated "using at least some of a predefined set of default model hyperparameters for each algorithm" (Ahuja, ¶ [0017]), each proxy model being "configured with respective default hyperparameters values" (Ahuja, ¶ [0052]).
"tuning the first untuned hyperparameter to the specific value to generate a tuned first model; and" — (Ahuja) teaches that "step 210A tunes hyperparameters of best model 160 such as by gradient descent" (Ahuja, ¶ [0075]), the result being "a rigorously trained ML model" (Ahuja, ¶ [0031]).
(Ahuja) teaches measuring temporal statistics of a timeseries and selecting and tuning a forecasting algorithm based on those statistics. However, (Ahuja) does not teach the following limitations, which (Rossi) teaches in the same field of endeavor:
"generating a first feature input based on the first dataset;" — (Rossi) teaches to "compute a time-series meta-feature vector based on the time-series dataset" (Rossi, Abstract), where "the system extracts meta-features from the time-series dataset" (Rossi, ¶ [0075]). The meta-feature vector corresponds to the first feature input.
"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;" — (Rossi) teaches "generating first predicted performance data for each of the plurality of forecasting models" (Rossi, ¶ [0070]) using a meta-learner that learns "which characteristics of time-series data are most suitable to certain types of forecasting models" (Rossi, ¶ [0055]), to "determine a forecasting model with the highest expected performance for an input time-series dataset" (Rossi, ¶ [0038]). The predicted performance data for each forecasting model corresponds to the series of values based on the respective effectiveness of the plurality of model types.
"generating for display, on a user interface, a recommendation for using the tuned first model for time-series forecasting." — (Rossi) teaches that the apparatus "may display the selected model or the forecasted data from the selected model to the user via user interface 215" (Rossi, ¶ [0037]).
(Ahuja) and (Rossi) are analogous to the claimed invention as both are from the same field of endeavor of automated selection of time-series forecasting models based on statistical characteristics of an input 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 combine the meta-feature vector, per-model performance prediction, and user-interface display of (Rossi) with the forecasting pipeline of (Ahuja). The motivation to combine is to identify "a forecasting model with the highest expected performance for an input time-series dataset" (Rossi, ¶ [0038]), consistent with the goal of (Ahuja) to "consume fewer computer resources to achieve a given accuracy" (Ahuja, ¶ [0032]).
The combination of (Ahuja) and (Rossi), however, does not teach the following limitations, which (Hetherington) teaches in the same field of endeavor:
"wherein the first dataset comprises one or more categories of data trends;" — (Hetherington) teaches "a framework for automatically generating statistical features describing trends in time-series data that may become inputs to deep learning models" (Hetherington, ¶ [0001]; see also ¶ [0005]). The trends described by the statistical features correspond to the one or more categories of data trends of the first dataset.
"wherein each of the first plurality of statistical routines is based on a first respective algorithm;" — (Hetherington) teaches statistical tests including "Grubb's test, student's t-test, generalized extreme studentized deviate (G-ESD) test" (Hetherington, ¶ [0035]) and window based statistical functions including "moving averages, weighted moving averages, moving variance, moving gradient, exponential smoothing" (Hetherington, ¶ [0048]), each of which is based on its own respective algorithm.
(Ahuja), (Rossi), and (Hetherington) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning that derives statistical descriptors from time-series data for forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the trend-describing statistical features and algorithm-specific statistical functions of (Hetherington) with the combination of (Ahuja) and (Rossi). The motivation to combine is that (Hetherington) recognizes "the benefits of using moving window based statistical functions for time-series forecasting" (Hetherington, ¶ [0048]) and automates "generating, evaluating, and selecting an optimal set of moving average features" (Hetherington, ¶ [0052]), which improves the accuracy of the forecasting model selected and tuned by the combination.
As to independent Claim 2, the preamble recites "A method for minimizing development time in artificial intelligence models by automating hyperparameter selection based on dataset fittings of time-series data, the method comprising:". (Ahuja) teaches this preamble by disclosing "[a] method comprising" automatically measuring a temporal statistic of a timeseries, selecting an ML algorithm, and tuning its hyperparameters (Ahuja, claim 1), so that the pipeline will "consume fewer computer resources to achieve a given accuracy" (Ahuja, ¶ [0032]).
Claim 2 is the method counterpart of Claim 1 and recites fewer limitations than Claim 1. The following limitations of Claim 2 are identical to, or contained within, the corresponding limitations of Claim 1, and are rejected under the same rationale as Claim 1:
"receiving a first dataset;"
"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,"
"determining a first aggregate statistical profile for the first dataset based on the first plurality of respective outputs;"
"filtering a first plurality of untuned hyperparameters based on a characteristic 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,"
"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; and"
"tuning the first untuned hyperparameter to the specific value."
Regarding the selecting limitation of Claim 2, (Ahuja) teaches:
"a first untuned hyperparameter, for a first plurality of untrained models, for tuning to a specific value by:" — (Ahuja) teaches that "based on temporal statistics 120, step 209 identifies a respective values subrange for some hyperparameter(s) of best model 160 as a preface to hyperparameters tuning" (Ahuja, ¶ [0074]). Best model 160 is instantiated from forecasting algorithms whose proxy models are "configured with respective default hyperparameters values" (Ahuja, ¶ [0052]).
(Ahuja) teaches selecting a hyperparameter for tuning based on temporal statistics 120. However, (Ahuja) does not teach the following limitation:
"selecting, based on the first aggregate statistical profile,"
In the same field of endeavor, (Rossi) teaches the following limitation. (Rossi) teaches to "select the forecasting model from a plurality of forecasting models based on the performance score" (Rossi, Abstract), "wherein the forecasting model is selected based on the first predicted performance data" (Rossi, ¶ [0070]). The predicted performance data corresponds to the first aggregate statistical profile. In the combination of (Ahuja) and (Rossi), the first untuned hyperparameter of (Ahuja) is selected based on the predicted performance data of (Rossi).
(Ahuja) and (Rossi) are analogous to the claimed invention as both are from the same field of endeavor of automated selection of time-series forecasting models based on statistical characteristics of an input 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 combine the profile-based selection of (Rossi) with the hyperparameter selection and tuning of (Ahuja). The motivation to combine (Ahuja) and (Rossi) is to select "a forecasting model with the highest expected performance for an input time-series dataset" (Rossi, ¶ [0038]) before committing computer resources to tuning, consistent with the goal of (Ahuja) to "consume fewer computer resources to achieve a given accuracy" (Ahuja, ¶ [0032]).
However, the combination of (Ahuja) and (Rossi) does not teach the following limitation:
"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 the following limitation. (Hetherington) teaches statistical tests including "Grubb's test, student's t-test, generalized extreme studentized deviate (G-ESD) test" (Hetherington, ¶ [0035]) and window based statistical functions including "moving averages, weighted moving averages, moving variance, moving gradient, exponential smoothing" (Hetherington, ¶ [0048]), each of which is based on its own respective algorithm.
(Ahuja), (Rossi), and (Hetherington) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning that derives statistical descriptors from time-series data for forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the algorithm-specific statistical functions of (Hetherington) with the combination of (Ahuja) and (Rossi). The motivation to combine (Ahuja), (Rossi), and (Hetherington) is that (Hetherington) recognizes "the benefits of using moving window based statistical functions for time-series forecasting" (Hetherington, ¶ [0048]), which improves the accuracy of the forecasting model selected and tuned by the combination.
As to dependent Claim 3, which recites "The method of claim 2, wherein determining the first plurality of respective outputs further comprises:", (Ahuja) teaches deriving temporal statistics 120, including seasonality 122 with measurements "such as a period and an amplitude," by "quantitatively analyzing the tuples in original timeseries 111" (Ahuja, ¶¶ [0036]–[0037]). However, the combination of (Ahuja) and (Rossi) does not teach the following limitations.
In the same field of endeavor, (Hetherington) teaches the following limitations:
"determining a first time period for a first model of the first plurality of statistical routines;" — (Hetherington) teaches that "the framework automatically generates multiple time-series features by constructing window based statistical features (e.g. moving average) of varying window sizes" (Hetherington, ¶ [0027]), using "[a] set of algorithms for selecting the number and window size of moving averages used as input features" (Hetherington, ¶ [0060]). The window size corresponds to the first time period, and the window based statistical function to which it applies corresponds to the first model of the first plurality of statistical routines.
"determining a first statistical variation for the first model over the first time period; and" — (Hetherington) teaches window based statistical functions including "moving variance," which are "used to smooth time-series to evaluate statistical deviations of data points from the smoothed time-series" (Hetherington, ¶ [0048]). The moving variance computed over the window corresponds to the first statistical variation for the first model over the first time period.
"determining a respective output, of the first plurality of respective outputs, for the first model based on the first statistical variation." — (Hetherington) teaches that "moving average information is presented as additional features to machine learning models" (Hetherington, ¶ [0050]), with "the original time-series data … augmented with automatically generated statistical features" (Hetherington, ¶ [0051]). The generated window based statistical feature corresponds to the respective output for the first model, determined based on the statistical variation computed over the window.
(Ahuja), (Rossi), and (Hetherington) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning that derives statistical descriptors from time-series data for forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the window based statistical features of (Hetherington) with the combination of (Ahuja) and (Rossi). The motivation to combine (Ahuja), (Rossi), and (Hetherington) is that "[t]ime-series forecasting using machine learning models can improve with the use of additional statistical features" and that "generating multiple moving average features with different window sizes preserves different levels of detail from the original time-series" (Hetherington, ¶ [0050]).
As to dependent Claim 4, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches the following limitations:
"comparing a first respective output of the first plurality of respective outputs to a threshold value; and" — (Ahuja) teaches that "[s]tep 303 applies an AutoCorrelation Function (ACF) over the differenced timeseries to calculate a respective correlation coefficient at specific lagged points in the timeseries for multiple seasonalities" (Ahuja, ¶ [0080]), and that "[s]tep 306 discards any of the k largest seasonalities whose correlation coefficient does not exceed a critical value threshold of autocorrelation (e.g. 5%)" (Ahuja, ¶ [0083]). The correlation coefficient output by the ACF corresponds to the first respective output, and the critical value threshold corresponds to the threshold value.
"determining a difference between the first respective output and the threshold value, wherein selecting the first untuned hyperparameter is based on the difference." — (Ahuja) determines whether each correlation coefficient "does not exceed" the critical value threshold (Ahuja, ¶ [0083]), which is a determination of the difference between the correlation coefficient and the threshold. The seasonalities not discarded include "the dominant seasonality" (Ahuja, ¶ [0083]). Based on seasonality 122, "step 209 identifies a respective values subrange for some hyperparameter(s) of best model 160 as a preface to hyperparameters tuning," for example by limiting "a seasonality hyperparameter to a range of 5-10 days" (Ahuja, ¶ [0074]). The seasonality hyperparameter is therefore selected for tuning based on the difference.
As to dependent Claim 5, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches:
"filtering the first plurality of untuned hyperparameters" — (Ahuja) teaches that "if proxy model 151 has the highest fitness score, then ML algorithm 131 is selected for further processing by the ML pipeline," and "ML algorithm 132 is excluded from further processing by the ML pipeline" (Ahuja, ¶ [0054]). Each proxy model is "configured with respective default hyperparameters values" (Ahuja, ¶ [0052]). Excluding an ML algorithm removes its untuned hyperparameters from further processing, which corresponds to filtering the first plurality of untuned hyperparameters.
(Ahuja) teaches excluding ML algorithms based on proxy-model fitness scores. However, (Ahuja) does not teach the following limitation:
"based on the first aggregate statistical profile."
In the same field of endeavor, (Rossi) teaches the following limitation. (Rossi) teaches "generating first predicted performance data for each of the plurality of forecasting models," "wherein the forecasting model is selected based on the first predicted performance data" (Rossi, ¶ [0070]), so that the forecasting model is selected "from a plurality of forecasting models based on the performance score" (Rossi, Abstract). As set forth for Claim 2, the predicted performance data corresponds to the first aggregate statistical profile. In the combination of (Ahuja) and (Rossi), the exclusion of ML algorithms, and of their untuned hyperparameters, taught by (Ahuja) is performed based on the predicted performance data of (Rossi).
(Ahuja) and (Rossi) are analogous to the claimed invention as both are from the same field of endeavor of automated selection of time-series forecasting models based on statistical characteristics of an input 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 perform the exclusion of (Ahuja) based on the predicted performance data of (Rossi). The motivation to combine (Ahuja) and (Rossi) is to retain only "a forecasting model with the highest expected performance for an input time-series dataset" (Rossi, ¶ [0038]), so that the least promising ML algorithms "receive very few computer resources" (Ahuja, ¶ [0032]).
As to dependent Claim 14, which recites "The method of claim 2, wherein determining the first aggregate statistical profile for the first dataset based on the first plurality of respective outputs further comprises:", (Ahuja) teaches:
"based on a comparison of the first plurality of respective outputs and respective model requirements for the first plurality of untrained models." — (Ahuja) teaches that "[e]ach ML algorithm is best suited to a varying degree of strength of basic timeseries characteristics such as stationarity, predictability, trend and seasonality" (Ahuja, ¶ [0017]). It further teaches that "based on temporal statistics 120, computer 100 may decide that ML algorithm 133 would have low accuracy for original timeseries 111," for example because "frequency 121 may be too high or too low for ML algorithm 133 to accurately perform" (Ahuja, ¶ [0051]). The timeseries characteristics to which each ML algorithm is best suited correspond to the respective model requirements, and the decision of whether frequency 121 is too high or too low for an ML algorithm corresponds to a comparison of the respective outputs and the respective model requirements.
(Ahuja) teaches comparing temporal statistics against the suitability of each ML algorithm. However, (Ahuja) does not teach the following limitations:
"generating a profile matrix for the first dataset; and"
"populating values of the profile matrix"
In the same field of endeavor, (Rossi) teaches the following limitations. (Rossi) teaches that "[r]unning the models produces a set of N=T×n distinct training samples in a meta-features matrix and a performance vector" (Rossi, ¶ [0108]). It further teaches that "[p]erformance data for the models may be represented in a performance tensor P," whose elements denote "the jth model Mj's performance on the time window wt of the ith training dataset Di" (Rossi, ¶ [0110]). The performance tensor for a dataset corresponds to the profile matrix for the first dataset, and assigning each element the performance of a respective model corresponds to populating values of the profile matrix. In the combination of (Ahuja) and (Rossi), the values of the performance tensor of (Rossi) are populated based on the comparison of temporal statistics and algorithm suitability taught by (Ahuja).
(Ahuja) and (Rossi) are analogous to the claimed invention as both are from the same field of endeavor of automated selection of time-series forecasting models based on statistical characteristics of an input 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 combine the performance tensor of (Rossi) with the algorithm-suitability comparison of (Ahuja). The motivation to combine (Ahuja) and (Rossi) is to organize per-model performance so that the meta-learner can determine "a forecasting model with the highest expected performance for an input time-series dataset" (Rossi, ¶ [0038]).
Claim 15 is rejected under the same rationale as Claim 2. Claim 15 recites one or more non-transitory, computer-readable mediums comprising instructions that, when executed by one or more processors, cause operations corresponding to the method steps of Claim 2.
Claim 16 is rejected under the same rationale as Claim 3. Claim 16 recites the non-transitory, computer-readable mediums of Claim 15 with the limitations of Claim 3.
Claim 17 is rejected under the same rationale as Claim 4. Claim 17 recites the non-transitory, computer-readable mediums of Claim 15 with the limitations of Claim 4.
As to dependent Claim 18, which recites "The one or more non-transitory, computer-readable mediums of claim 15, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", the limitation "filtering the first plurality of untuned hyperparameters based on the first aggregate statistical profile" is rejected under the same rationale as Claim 5.
Claim 18 further recites limitations not recited in Claim 5. (Ahuja) teaches:
"to generate a filtered subset of the first plurality of untuned hyperparameters; and" — (Ahuja) teaches that "if proxy model 151 has the highest fitness score, then ML algorithm 131 is selected for further processing," and "ML algorithm 132 is excluded from further processing by the ML pipeline" (Ahuja, ¶ [0054]). The untuned hyperparameters of the ML algorithms that remain after the exclusion correspond to the filtered subset of the first plurality of untuned hyperparameters.
"selecting the first untuned hyperparameter from the filtered subset." — (Ahuja) teaches that "based on temporal statistics 120, step 209 identifies a respective values subrange for some hyperparameter(s) of best model 160 as a preface to hyperparameters tuning" (Ahuja, ¶ [0074]), best model 160 being instantiated from the ML algorithm that was not excluded (Ahuja, ¶ [0055]). The hyperparameter of best model 160 identified for tuning corresponds to the first untuned hyperparameter selected from the filtered subset.
Claims 6 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Ahuja et al. (Ahuja), US 2023/0153394 A1, in view of Rossi et al. (Rossi), US 2024/0152769 A1, further in view of Hetherington et al. (Hetherington), US 2020/0097810 A1, further in view of Vu et al. (Vu), US 2022/0327058 A1, and further in view of Higginson et al. (Higginson), US 2023/0195591 A1.
As to dependent Claim 6, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches that in original timeseries 111, "[e]ach tuple may correspond to a distinct time in the past" (Ahuja, ¶ [0034]). However, the combination of (Ahuja), (Rossi), and (Hetherington) does not teach the following limitation:
"determining an age based on the characteristic; and"
In the same field of endeavor, (Vu) teaches the following limitation. (Vu) teaches that time series "data evolves over time, so the historical data become less and less relevant as time passes" (Vu, ¶ [0023]), and that "a sequential order of the time series data set may be used while allocating the time series data," with "[t]he time series data … allocated backward in time" (Vu, ¶ [0029]), "starting from most recent data" (Vu, ¶ [0111]). The sequential order of the time series data set corresponds to the characteristic, and the recency of the data derived from that sequential order corresponds to the age.
(Ahuja), (Rossi), (Hetherington), and (Vu) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for time-series forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the recency-based data handling of (Vu) with the combination of (Ahuja), (Rossi), and (Hetherington). The motivation to combine (Ahuja), (Rossi), (Hetherington), and (Vu) is that "the historical data become less and less relevant as time passes" (Vu, ¶ [0023]), so accounting for the age of the data improves the accuracy of the selected forecasting model.
However, the combination of (Ahuja), (Rossi), (Hetherington), and (Vu) does not teach the following limitation:
"filtering the first plurality of untuned hyperparameters based on the age of the first dataset."
In the same field of endeavor, (Higginson) teaches the following limitation. (Higginson) teaches that a time-series model is stale "if a predetermined period has elapsed" (Higginson, ¶ [0127]). After that period, the system "retrains the time-series model using more recent time-series data," obtains "a new training data set and test data set from metrics collected over a recent number of days, weeks, months," and uses it "to generate a set of time-series models with different combinations of parameter values" (Higginson, ¶ [0128]). For those candidate models, the "model parameter selection engine 181 … filters a parameter space" (Higginson, ¶ [0040]). The parameter space of the candidate time-series models is thus filtered for data selected according to its age, which corresponds to filtering the first plurality of untuned hyperparameters based on the age of the first dataset.
(Ahuja), (Rossi), (Hetherington), (Vu), and (Higginson) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for time-series forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the age-based parameter filtering of (Higginson) with the combination of (Ahuja), (Rossi), (Hetherington), and (Vu). The motivation to combine (Ahuja), (Rossi), (Hetherington), (Vu), and (Higginson) is that filtering the parameter space of candidate models "reduces the overall time it takes to compute and perform the predictions" (Higginson, ¶ [0044]).
Claim 19 is rejected under the same rationale as Claim 6. Claim 19 recites the non-transitory, computer-readable mediums of Claim 15 with the limitations of Claim 6.
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Ahuja et al. (Ahuja), US 2023/0153394 A1, in view of Rossi et al. (Rossi), US 2024/0152769 A1, further in view of Hetherington et al. (Hetherington), US 2020/0097810 A1, and further in view of Rahimi et al. (Rahimi), US 2024/0311650 A1.
As to dependent Claim 7, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches temporal statistics including "heteroskedasticity that is a kind of volatility based on measuring fluctuating variance" (Ahuja, ¶ [0039]). However, the combination of (Ahuja), (Rossi), and (Hetherington) does not teach the following limitations.
In the same field of endeavor, (Rahimi) teaches the following limitations:
"determining a reliability based on the characteristic; and" — (Rahimi) teaches an outlier removal technique "configured to ensure that any impossible data or human error is not being fed into the model creation pipeline," in which "[o]utliers may be initially detected by setting a threshold value for a maximum amount of standard deviation presented" (Rahimi, ¶ [0042]). It further teaches assessing whether "the quality of the data is bad," including whether "there are too many missing values or a high number of outliers" (Rahimi, ¶ [0043]). The standard deviation, outlier count, and missing-value count correspond to the characteristic, and the resulting determination of the quality of the data corresponds to the reliability.
"filtering the first plurality of untuned hyperparameters based on the reliability of the first dataset." — (Rahimi) teaches that model search space 216 enables "the inclusion or removal of one or more models in the model search space 216, for example manually by a user or automatically based on one or more characteristics" (Rahimi, ¶ [0055]). For example, "the Prophet model is effective with data that has many change points and outliners" (Rahimi, ¶ [0070]). Each model carries "hyperparameters that are specific to each of the set of univariate models 510" (Rahimi, ¶ [0071]). Removing models from the search space based on the outlier and missing-value characteristics of the data removes their untuned hyperparameters, which corresponds to filtering the first plurality of untuned hyperparameters based on the reliability of the first dataset.
(Ahuja), (Rossi), (Hetherington), and (Rahimi) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for time-series forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the data-quality assessment and characteristic-based model removal of (Rahimi) with the combination of (Ahuja), (Rossi), and (Hetherington). The motivation to combine (Ahuja), (Rossi), (Hetherington), and (Rahimi) is to "ensure that any impossible data or human error is not being fed into the model creation pipeline" (Rahimi, ¶ [0042]) and to retain only the models effective for the data presented (Rahimi, ¶ [0070]).
Claims 8 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Ahuja et al. (Ahuja), US 2023/0153394 A1, in view of Rossi et al. (Rossi), US 2024/0152769 A1, further in view of Hetherington et al. (Hetherington), US 2020/0097810 A1, and further in view of Vu et al. (Vu), US 2022/0327058 A1.
As to dependent Claim 8, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches that "if proxy model 151 has the highest fitness score, then ML algorithm 131 is selected for further processing" (Ahuja, ¶ [0054]). However, the combination of (Ahuja), (Rossi), and (Hetherington) does not teach the following limitations.
In the same field of endeavor, (Vu) teaches the following limitations:
"ranking the 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" — (Vu) teaches that "the predicted score vectors … may be ranked from a minimum ('min') to a maximum ('max')," with "the corresponding pipelines … maintained in a priority queue" (Vu, ¶ [0117]), so that "[a] ranked list of machine learning pipelines may be automatically generated from the candidate machine learning pipelines for time series forecasting" (Vu, Abstract). Each pipeline embodies a decision as to "which machine algorithm should be used with which settings (hyperparameters)" (Vu, ¶ [0021]). Ranking the pipelines therefore ranks their untuned hyperparameters. The predicted score of each pipeline corresponds to the respective value of the first aggregate statistical profile, and the priority queue corresponds to the ranked order.
"selecting the first untuned hyperparameter based on the ranked order." — (Vu) teaches that "additional next_allocation data points may be allocated to a top/optimized machine learning pipeline … in the priority queue" (Vu, ¶ [0119]). Selecting the top pipeline from the priority queue, and with it its hyperparameter settings, corresponds to selecting the first untuned hyperparameter based on the ranked order.
(Ahuja), (Rossi), (Hetherington), and (Vu) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for time-series forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the ranking of candidate pipelines of (Vu) with the combination of (Ahuja), (Rossi), and (Hetherington). The motivation to combine (Ahuja), (Rossi), (Hetherington), and (Vu) is that "[i]f each candidate machine learning pipeline is provided the entire input dataset, the automated AI machine learning system run time may be too time consuming" (Vu, ¶ [0027]); ranking the candidates directs computation to the most promising pipeline.
As to dependent Claim 12, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches "automatically selecting, based on said temporal statistic of the timeseries, a length of a subsequence of the timeseries for training the ML model" (Ahuja, claim 6). However, the combination of (Ahuja), (Rossi), and (Hetherington) does not teach the following limitations.
In the same field of endeavor, (Vu) teaches the following limitations:
"determining respective sample size requirements for training for each of the first plurality of untuned hyperparameters based on the first aggregate statistical profile; and" — (Vu) teaches that "[a] data allocation size of time series data may be determined based on one or more characteristics of a time series data set" (Vu, ¶ [0027]), where "the allocation_increment may be set as equal to the seasonality length" estimated "using Fast Fourier Transformation" (Vu, ¶ [0113]). For each pipeline, "a regression fit may be performed on target variables scores V's predictor sample sizes" (Vu, ¶ [0116]). Each pipeline carries its own hyperparameter settings (Vu, ¶ [0021]). The data allocation size determined for each pipeline corresponds to the respective sample size requirements for training, and the dataset characteristics, such as seasonality, on which it is based correspond to the first aggregate statistical profile.
"selecting the first untuned hyperparameter based on the respective sample size requirements for training." — (Vu) teaches that "[a] score may be predicted when a sample size is equal to total length of input time series data 'L'" for each pipeline (Vu, ¶ [0116]), that the pipelines are ranked by the predicted scores (Vu, ¶ [0117]), and that additional data points are "allocated to a top/optimized machine learning pipeline … in the priority queue" (Vu, ¶ [0119]). Selecting the top pipeline, together with its hyperparameter settings, based on scores projected from the allocated sample sizes corresponds to selecting the first untuned hyperparameter based on the respective sample size requirements for training.
(Ahuja), (Rossi), (Hetherington), and (Vu) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for time-series forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the sample-size-based data allocation of (Vu) with the combination of (Ahuja), (Rossi), and (Hetherington). The motivation to combine (Ahuja), (Rossi), (Hetherington), and (Vu) is that "[i]f each candidate machine learning pipeline is provided the entire input dataset, the automated AI machine learning system run time may be too time consuming" (Vu, ¶ [0027]).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Ahuja et al. (Ahuja), US 2023/0153394 A1, in view of Rossi et al. (Rossi), US 2024/0152769 A1, further in view of Hetherington et al. (Hetherington), US 2020/0097810 A1, and further in view of Yakovlev et al. (Yakovlev), US 11,429,895 B2.
As to dependent Claim 9, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches "asymmetry of investment of computer resources in the ML algorithms," in which "the least promising ML algorithms receive very few computer resources" (Ahuja, ¶ [0032]). However, the combination of (Ahuja), (Rossi), and (Hetherington) does not teach the following limitations.
In the same field of endeavor, (Yakovlev) teaches the following limitations:
"determining respective training time predictions for each of the first plurality of untuned hyperparameters based on the first aggregate statistical profile; and" — (Yakovlev) teaches training "a regressor to predict how much time would be needed to train the ML model based on a hyperparameter configuration and a dataset" (Yakovlev, col. 2, ll. 25–29). It describes the dataset by meta-features that "may be statistical such as: mean, min, max, quantile values of features, skew, and kurtosis" (Yakovlev, col. 2, ll. 46–48), which "can be put together into an encoded feature vector" (Yakovlev, col. 2, ll. 51–52). "Each unique hyperparameter configuration vector, together with dataset meta-features, makes up a unique training example for a regressor that can be trained to predict training times for the ML model" (Yakovlev, col. 2, ll. 56–59). The regressor "predicts an inferred duration needed to train the ML model based on: a) a proposed configuration of the ML model, b) dataset meta-feature values" (Yakovlev, col. 3, ll. 21–25). The predicted duration for each hyperparameter configuration corresponds to the respective training time predictions, and the encoded vector of statistical dataset meta-features corresponds to the first aggregate statistical profile.
"selecting the first untuned hyperparameter based on the respective training time predictions." — (Yakovlev) teaches that "[p]redicting training time may facilitate cost-aware hyperparameter tuning" (Yakovlev, col. 3, ll. 32–33) and "may also facilitate cost-aware model selection" (Yakovlev, col. 3, ll. 41–42). Cost-aware selection among hyperparameter configurations based on their predicted training times corresponds to selecting the first untuned hyperparameter based on the respective training time predictions.
(Ahuja), (Rossi), (Hetherington), and (Yakovlev) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for model and hyperparameter selection. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the training-time prediction of (Yakovlev) with the combination of (Ahuja), (Rossi), and (Hetherington). The motivation to combine (Ahuja), (Rossi), (Hetherington), and (Yakovlev) is that predicting training time enables "cost-based task scheduling and task management, cost-based scaling and resource allocation" (Yakovlev, col. 3, ll. 33–35), furthering the goal of (Ahuja) to "consume fewer computer resources to achieve a given accuracy" (Ahuja, ¶ [0032]).
Claims 10 and 11 are rejected under 35 U.S.C. 103 as being unpatentable over Ahuja et al. (Ahuja), US 2023/0153394 A1, in view of Rossi et al. (Rossi), US 2024/0152769 A1, further in view of Hetherington et al. (Hetherington), US 2020/0097810 A1, and further in view of Moghadam et al. (Moghadam), US 11,620,568 B2.
As to dependent Claim 10, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches proxy models "configured with respective default hyperparameters values" (Ahuja, ¶ [0052]) and selecting the ML algorithm whose proxy model "has the highest fitness score" (Ahuja, ¶ [0054]). However, the combination of (Ahuja), (Rossi), and (Hetherington) does not teach the following limitations.
In the same field of endeavor, (Moghadam) teaches the following limitations:
"determining respective performance predictions for each of the first plurality of untuned hyperparameters based on the first aggregate statistical profile; and" — (Moghadam) teaches "selection of machine learning algorithms based on performance predictions by trained algorithm-specific regressors using hyperparameter predictors" (Moghadam, col. 2, ll. 19–22), in which "[t]he trained mini-models use hyper-parameters predicted by hyperparameter predictors based on meta-features of the data set" (Moghadam, col. 2, ll. 29–31). For each reference model, "a RML predictor is trained … using a mini-model having optimal hyperparameter settings predicted by the hyperparameter predictor," where "the training inputs include meta-features generated for data set samples" and "[t]he target is the scores generated by the AHT algorithm for full data set" (Moghadam, col. 2, ll. 52–60). The predicted score for each model under its predicted hyperparameter settings corresponds to the respective performance predictions, and the data set meta-features correspond to the first aggregate statistical profile.
"selecting the first untuned hyperparameter based on the respective performance predictions." — (Moghadam) teaches that "[t]he use of mini-models and improved scores are used to more efficiently and accurately rank and select the best algorithm for the given dataset" (Moghadam, col. 2, ll. 65–67). Selecting the best algorithm, together with its predicted hyperparameter settings, based on the predicted scores corresponds to selecting the first untuned hyperparameter based on the respective performance predictions.
(Ahuja), (Rossi), (Hetherington), and (Moghadam) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for model and hyperparameter selection. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the hyperparameter-based performance prediction of (Moghadam) with the combination of (Ahuja), (Rossi), and (Hetherington). The motivation to combine (Ahuja), (Rossi), (Hetherington), and (Moghadam) is that "[u]sing predicted hyperparameters for a mini-model improves the accuracy of scores of each mini-model" (Moghadam, col. 2, ll. 61–63), so that the best algorithm is selected "more efficiently and accurately" (Moghadam, col. 2, ll. 65–67).
As to dependent Claim 11, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches that "[e]ach of proxy models 150 may have respective different counts and kinds of hyperparameters" (Ahuja, ¶ [0052]). However, the combination of (Ahuja), (Rossi), and (Hetherington) does not teach the following limitations.
In the same field of endeavor, (Moghadam) teaches the following limitations:
"determining respective predictions for a number of hyperparameters requiring training for each of the first plurality of untuned hyperparameters based on the first aggregate statistical profile; and" — (Moghadam) teaches that "a respective hyperparameter predictor set is trained that predicts a respective set of hyperparameter settings for the first data set," where "[t]he hyperparameter predictor set may comprise one or more hyperparameter models" (Moghadam, col. 7, ll. 13–16), and that "there is one RF predictor for each mini-model hyperparameter" (Moghadam, col. 6, ll. 1–2). The predictions are made "based on meta-features of the data set" (Moghadam, col. 2, ll. 29–31). Each model's predicted set of hyperparameter settings, which has one predictor per hyperparameter, corresponds to the respective prediction for the number of hyperparameters requiring training, and the data set meta-features correspond to the first aggregate statistical profile.
"selecting the first untuned hyperparameter based on the respective predictions for the number of hyperparameters requiring training." — (Moghadam) teaches that "[t]he meta-features are then used to predict hyperparameters of the mini-models," the mini-model scores are used "to predict reference-model scores," and "[t]he machine learning algorithms are then ranked based on their scores" (Moghadam, col. 6, ll. 22–29), so as to "rank and select the best algorithm for the given dataset" (Moghadam, col. 2, ll. 65–67). Selecting the best algorithm, together with its predicted set of hyperparameter settings, corresponds to selecting the first untuned hyperparameter based on the respective predictions.
(Ahuja), (Rossi), (Hetherington), and (Moghadam) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for model and hyperparameter selection. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the per-model hyperparameter predictor sets of (Moghadam) with the combination of (Ahuja), (Rossi), and (Hetherington). The motivation to combine (Ahuja), (Rossi), (Hetherington), and (Moghadam) is that "[u]sing predicted hyperparameters for a mini-model improves the accuracy of scores of each mini-model" (Moghadam, col. 2, ll. 61–63), so that the best algorithm is selected "more efficiently and accurately" (Moghadam, col. 2, ll. 65–67).
Claims 13 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Ahuja et al. (Ahuja), US 2023/0153394 A1, in view of Rossi et al. (Rossi), US 2024/0152769 A1, further in view of Hetherington et al. (Hetherington), US 2020/0097810 A1, and further in view of Higginson et al. (Higginson), US 2023/0195591 A1.
As to dependent Claim 13, which recites "The method of claim 2, wherein selecting, based on the first aggregate statistical profile, the first untuned hyperparameter further comprises:", (Ahuja) teaches "asymmetry of investment of computer resources in the ML algorithms as decided based on temporal statistics," in which "the least promising ML algorithms receive very few computer resources" (Ahuja, ¶ [0032]). However, the combination of (Ahuja), (Rossi), and (Hetherington) does not teach the following limitations.
In the same field of endeavor, (Higginson) teaches the following limitations:
"determining respective processing power requirements for training for each of a first plurality of untuned hyperparameters based on the first aggregate statistical profile; and" — (Higginson) teaches that "[t]he computational cost and the time to perform the computations to test each version of every model with different parameter values may exceed a performance threshold," and that "the system may require the forecast while consuming only a predefined amount of resources, such as processing resources" (Higginson, ¶ [0040]). The parameter values are evaluated "[b]ased on the identified characteristics of the historical time-series data" (Higginson, ¶ [0043]). The computational cost of testing each version of a model with different parameter values corresponds to the respective processing power requirements for training, and the identified characteristics of the time-series data correspond to the first aggregate statistical profile.
"selecting the first untuned hyperparameter based on the respective processing power requirements for training." — (Higginson) teaches that "[t]he model parameter selection engine 181 filters the search space of parameters into a range of parameters corresponding to an execution time and a resource cost that meets system specifications" (Higginson, ¶ [0040]). Selecting parameter values from the range that meets the resource-cost specification corresponds to selecting the first untuned hyperparameter based on the respective processing power requirements for training.
(Ahuja), (Rossi), (Hetherington), and (Higginson) are analogous to the claimed invention as all are from the same field of endeavor of automated machine learning for time-series forecasting. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention, to combine the resource-cost-based parameter selection of (Higginson) with the combination of (Ahuja), (Rossi), and (Hetherington). The motivation to combine (Ahuja), (Rossi), (Hetherington), and (Higginson) is that training models over every parameter combination "takes too much time to be useful and consumes too many computing resources to be practical" (Higginson, ¶ [0028]), which is the problem (Ahuja) addresses by seeking to "consume fewer computer resources to achieve a given accuracy" (Ahuja, ¶ [0032]).
Claim 20 is rejected under the same rationale as Claim 13. Claim 20 recites the non-transitory, computer-readable mediums of Claim 15 with the limitations of Claim 13.
Response to Arguments
Applicant's arguments filed 07/30/2026 have been fully considered. The arguments regarding the rejection under 35 U.S.C. 101 are not persuasive. The arguments regarding the rejection under 35 U.S.C. 103 are not persuasive with respect to claims 1, 2, and 15, and are moot with respect to claims 6, 7, and 19 because those rejections rely on new grounds of rejection necessitated by Applicant's amendment. The arguments are addressed below.
Rejection under 35 U.S.C. 101
Argument A: Applicant argues that, under the USPTO memorandum issued to Technology Centers 2100, 2600, and 3600 on August 4, 2025 ("Guidance"), "[c]laim limitations that encompass AI in a way that cannot be practically performed in the human mind do not fall within this grouping." Applicant contends that the claims recite "datasets," "tuning the first untuned hyperparameter," "a plurality of statistical routines," and "user interfaces," and therefore cannot be practically performed in the human mind. (See Remarks, page 10.)
In reply to argument A, Examiner respectfully disagrees. Under MPEP 2106.04(a)(2), subsection III, a limitation falls within the mental-processes grouping when it can practically be performed in the human mind, including with the aid of pen and paper. A limitation that cannot practically be performed in the human mind does not fall within that grouping.
The limitations identified at Step 2A Prong 1 do not encompass AI in a way that cannot be practically performed in the human mind:
Generating a feature input from a dataset.
Computing statistical outputs such as a mean or variance (specification ¶ [0036]).
Aggregating those outputs into a profile.
Filtering a list of candidate hyperparameters based on a characteristic of the dataset.
Selecting a hyperparameter from the remaining candidates.
Assigning that hyperparameter a specific value.
The claims do not recite training a neural network, executing a model on a processor in a manner that the human mind cannot perform, or any operation of comparable complexity. They recite the evaluation and selection of hyperparameters at a high level of generality.
Further, the "statistical routines" were identified as reciting mathematical concepts under MPEP 2106.04(a)(2), subsection I, independently of the mental-processes grouping. The Guidance's statement concerning the mental-processes grouping does not affect that determination.
The "datasets" and "user interfaces" relied on by Applicant were not characterized as mental processes. "Receiving a first dataset" and "generating for display, on a user interface, a recommendation" were evaluated as additional elements at Step 2A Prong 2 and Step 2B.
Applicant also characterizes the claims as "automating model selection based on dataset fittings of time-series data prior to hyperparameter optimization." The claims do not recite model selection prior to hyperparameter optimization; they recite selecting and tuning an untuned hyperparameter. The preamble's statement of "minimizing development time" expresses an intended result and does not change the character of the recited steps. Accordingly, the claims recite a judicial exception at Step 2A Prong 1.
Argument B: Applicant argues that the claims reflect an improvement to the functioning of a computer or another technology under Step 2A Prong 2. Applicant contends that specification ¶¶ [0005], [0007], and [0008] describe a technological problem, namely redundant training, fitting, and tuning time and the absence of a linear relationship between dataset attributes and model effectiveness. Applicant contends that the use of an aggregate statistical profile is "a technological solution to a technological problem." (See Remarks, pages 10-12.)
In reply to argument B, Examiner respectfully disagrees. Under MPEP 2106.04(d)(1) and MPEP 2106.05(a), the improvement must be to the functioning of a computer or to another technology or technical field, and the claim itself must reflect the disclosed improvement. An improvement in the abstract idea itself is not an improvement in technology. See MPEP 2106.05(a), subsection II; SAP America, Inc. v. InvestPic, LLC, 898 F.3d 1161, 1163 (Fed. Cir. 2018).
The solution Applicant identifies in ¶ [0008] is "the use of an aggregate statistical profile based on the results of a plurality of known statistical analyses." That aggregate statistical profile, and the statistical analyses from which it is computed, are the mathematical concepts and mental processes identified at Step 2A Prong 1. Any reduction in redundant training, fitting, or tuning time results from performing that abstract selection process more efficiently. It is not an improvement to how a computer, a processor, or a training process operates. The claims recite no change to computer functionality: the processors and computer-readable mediums are generic and merely execute the abstract steps (MPEP 2106.05(f)).
Moreover, the Guidance quoted by Applicant directs examiners to consider "the extent to which the claim covers a particular solution to a problem or a particular way to achieve a desired outcome, as opposed to merely claiming the idea of a solution or outcome." The claims recite the outcome of selecting and tuning a hyperparameter based on statistical outputs. They do not recite a particular technological way of achieving that outcome, and they do not apply the result to effect a technological change. This is in contrast to Example 47, claim 3, of the 2024 Subject Matter Eligibility Update, where the detected anomalies were used to drop malicious network packets and block future traffic. Accordingly, the claims do not integrate the judicial exception into a practical application.
Argument C: Applicant argues that "as the system use[s] unconventional technical features, the claim is patentable subject matter under Step 2B." (See Remarks, page 12.)
In reply to argument C, Examiner respectfully disagrees. Applicant does not identify any additional element that is unconventional. A general allegation of patentability that does not specifically point out how the claims distinguish over the rejection does not comply with 37 C.F.R. 1.111(b).
As set forth in the rejection, the additional elements are the following:
Generic processors and computer-readable mediums.
Receiving a dataset.
Displaying a recommendation on a user interface.
Time-series forecasting models with default hyperparameters.
Considered individually and as an ordered combination, these elements amount to mere instructions to apply the exception, insignificant extra-solution activity, and a field-of-use limitation. Generic receiving, storing, and presenting of data are well-understood, routine, and conventional functions recognized by the court decisions cited in MPEP 2106.05(d), subsection II. Accordingly, the claims do not amount to significantly more than the judicial exception, and the rejection under 35 U.S.C. 101 is maintained.
Rejection under 35 U.S.C. 103
Argument D: Applicant argues that the amended claims recite "filtering a first plurality of untuned hyperparameters based on a characteristic 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." Applicant contends that (Vu) does not teach any filtering based on a characteristic of a dataset, because the sequential order relied upon is "specific to the individual datum, not the dataset as a whole," and is "never used for filtering." (See Remarks, pages 12-13.)
In reply to argument D, as it applies to independent claims 1, 2, and 15, Examiner respectfully disagrees. The rejection of claims 1, 2, and 15 does not rely on (Vu). The amended limitation is taught by (Ahuja): (Ahuja) teaches that "based on temporal statistics 120, computer 100 may decide that ML algorithm 133 would have low accuracy for original timeseries 111 and thus is unsuitable and should be excluded from further processing," for example because "frequency 121 may be too high or too low" (Ahuja, ¶ [0051]). It further teaches that "step 209 identifies a respective values subrange for some hyperparameter(s) of best model 160 as a preface to hyperparameters tuning" (Ahuja, ¶ [0074]).
The characteristic on which (Ahuja) filters is a characteristic of the dataset as a whole, because temporal statistics 120, including frequency 121 and seasonality 122, "characterize original timeseries 111 as a whole" (Ahuja, ¶ [0036]). Applicant's arguments against (Vu) individually do not address the reference actually relied upon for this limitation. One cannot show nonobviousness by attacking references individually where the rejections are based on combinations of references. See MPEP 2145, subsection IV; In re Keller, 642 F.2d 413, 426 (CCPA 1981); In re Merck & Co., 800 F.2d 1091, 1097 (Fed. Cir. 1986). Accordingly, the rejection of claims 1, 2, and 15, and of the claims depending therefrom, is maintained.
In reply to argument D, as it applies to claims 6 and 19, the arguments are moot because the rejection relies on (Higginson), a new reference applied in the new ground of rejection necessitated by Applicant's amendment. In the current rejection, (Vu) is relied upon only for "determining an age based on the characteristic." It teaches that time series "data evolves over time, so the historical data become less and less relevant as time passes" (Vu, ¶ [0023]), and that "a sequential order of the time series data set may be used while allocating the time series data" (Vu, ¶ [0029]). The sequential order is that of the time series data set, not of an individual datum. The limitation "filtering the first plurality of untuned hyperparameters based on the age of the first dataset" is taught by (Higginson) (Higginson, ¶¶ [0040], [0127]-[0128]), as set forth in the rejection above.
In reply to argument D, as it applies to claim 7, the arguments are moot because the rejection no longer relies on (Vu). Claim 7 is rejected under a new ground of rejection, necessitated by Applicant's amendment, further in view of (Rahimi).
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to 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
Prosecution Insights