DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Claims 1-5, 7, and 9-12 are presented for examination.
Priority
Receipt is acknowledged of certified copies of papers required by 37 CFR 1.55.
Claim Rejections - 35 USC § 101
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1-5, 7, and 9-12 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The analysis of the claims will follow the 2019 Revised Patent Subject Matter Eligibility Guidance, 84 Fed. Reg. 50 (“2019 PEG”).
Claim 1
Step 1: The claim recites a non-transitory computer-readable medium; therefore, it is directed to the statutory category of articles of manufacture.
Step 2A Prong 1: The claim recites, inter alia:
[D]etermining whether, in a training data set including a plurality of pieces of training data, a count of a first label that characterizes a greatest amount of the normal data among the training data and a count of a second label that characterizes a smallest amount of the abnormal data among the training data are imbalanced: This limitation could encompass the mental determination that a minority class and a majority class are imbalanced.
[G]enerating, when it is determined that the count of the first label and the count of the second label are imbalanced, a plurality of subsets each including first training data characterized by the first label and at least a portion of second training data characterized by the second label, the first training data having a count balanced with the count of the second label, the plurality of subsets being generated by dividing the training data set into the plurality of subsets while updating a number of divisions so that a different combination of the first training data is included in each subset: This limitation could encompass the mental generation of the different balanced subsets in the manner claimed.
[G]enerating a plurality of first learning models based on each subset in the generated plurality of subsets: Since the claim does not specify that the models in question are machine learning models, they could be simple mathematical models that can be generated in the mind.
[I]ntegrating, by majority vote, predicted values resulting when validation data are inputted to each first learning model: This limitation could encompass the mental integration of the predicted values.
[D]etermin[ing] that a value of a first evaluation index for the plurality of first learning models generated with the predetermined number of divisions is higher than a value of a second evaluation index for a second learning model generated based on the training data set without generation of the plurality of subsets: This limitation could encompass the mental determination that an evaluation index for one set of models is higher than that of another model.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim additionally recites a “non-transitory computer readable medium storing a program regarding a learning model for classifying data by characterizing the data with one label among a plurality of labels, the program being executable by one or more processors to cause an information processing apparatus to execute functions” and that “the determination data [are] newly classified by machine learning based on the stored plurality of first learning models”. However, these limitations are mere instructions to apply the judicial exception using a generic computer programmed with a generic class of computer algorithm. MPEP § 2106.05(f).
The claim further recites “inputting a plurality of pieces of training data required for generating the model, the training data including normal data and abnormal data of products obtained in manufacturing industries that use plants”; “saving the plurality of first learning models when [the criterion is satisfied]”; “inputting determination data”, and “outputting [a] classification result of the determination data to a user as information”. However, these limitations are directed to the insignificant extra-solution activity of mere data gathering and output. MPEP § 2106.05(g).
Step 2B: The claim does not contain significantly more than the judicial exception. The recitation that the operations are performed via instructions stored on a non-transitory medium and the recitation of the use of machine learning to perform the classification are mere instructions to apply the exception for the same reasons as given above. In addition to being insignificant extra-solution activity, the inputting, saving, inputting, and outputting limitations recite the well-understood, routine, and conventional activity of storing and retrieving information in memory. MPEP § 2106.05(d)(II); Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015). Otherwise, the analysis is identical to that of step 2A, prong 2. As an ordered whole, the claim is directed to a mentally performable method of generating models by balancing datasets. Nothing in the claim provides significantly more than this. As such, the claim is not patent eligible.
Claim 2
Step 1: An article of manufacture, as above.
Step 2A Prong 1: The claim recites “determining, before the generating of the plurality of subsets, the number of divisions when dividing the training data set into the plurality of subsets.” This limitation could encompass the mental determination of the subsets.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis supra.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis supra.
Claim 3
Step 1: An article of manufacture, as above.
Step 2A Prong 1: The claim recites “determining of the number of divisions comprises determining the number of divisions based on information inputted by a user.” This limitation could encompass mentally performing this determination.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 2 analysis supra.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 2 analysis supra.
Claim 4
Step 1: An article of manufacture, as above.
Step 2A Prong 1: The claim recites, inter alia, that “the determining of the number of divisions comprises determining the number of divisions … based on an initial setting.” This limitation could encompass the mental determination of the number of divisions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim additionally recites that this determination occurs “automatically”. However, this is a mere instruction to apply the judicial exception using a generic computer. MPEP § 2106.05(f).
Step 2B: The claim does not contain significantly more than the judicial exception. The claim additionally recites that this determination occurs “automatically”. However, this is a mere instruction to apply the judicial exception using a generic computer. MPEP § 2106.05(f).
Claim 5
Step 1: An article of manufacture, as above.
Step 2A Prong 1: The claim recites “repeatedly updating the determined number of divisions to a different value within a predetermined range, calculating the first evaluation index based on each updated number of divisions, and determining the number of divisions to be the number of divisions for which the value of the first evaluation index is highest.” These limitations could encompass the mental updating of the divisions, calculation of the evaluation index, and determining of the number of divisions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 2 analysis supra.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 2 analysis supra.
Claim 7
Step 1: An article of manufacture, as above.
Step 2A Prong 1: The claim recites that the “generating of the plurality of subsets comprises generating another subset by newly sampling the first training data from the training data set after excluding, from the training data set, the first training data sampled into one subset.” This limitation could encompass the mental generation of the subsets by sampling from a subset of the training data.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis supra.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis supra.
Claim 9
Step 1: The claim is directed to an information processing apparatus comprising a controller and a storage; therefore, the claim is directed to the statutory category of machines.
Step 2A Prong 1: The claim recites the same judicial exceptions as in claim 1.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The analysis at this step is identical to that of claim 1, with the exception that this claim is directed to an “information processing apparatus for generating a learning model for classifying data by characterizing the data with one label among a plurality of labels, the information processing apparatus comprising: a controller; a storage; an input interface; and an output interface, wherein the controller is configured to [perform the method]”. However, this is a mere instruction to apply the judicial exception using a generic computer. MPEP § 2106.05(f).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step is identical to that of claim 1, with the exception that this claim is directed to an “information processing apparatus for generating a learning model for classifying data by characterizing the data with one label among a plurality of labels, the information processing apparatus comprising: a controller; a storage; an input interface; and an output interface, wherein the controller is configured to [perform the method]”. However, this is a mere instruction to apply the judicial exception using a generic computer. MPEP § 2106.05(f).
Claim 10
Step 1: The claim is directed to a method; therefore, it is directed to the statutory category of processes.
Step 2A Prong 1: The claim recites the same judicial exceptions as in claim 1.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The analysis at this step is identical to that of claim 1, except insofar as claim 10 is directed to a method rather than to a non-transitory computer-readable medium.
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step is identical to that of claim 1, except insofar as claim 10 is directed to a method rather than to a non-transitory computer-readable medium.
Claim 11
Step 1: An article of manufacture, as above.
Step 2A Prong 1: The claim recites, inter alia, “determining the number of divisions so that a ratio of a count of the first label to a count of the second label in one subset is equal to or less than a threshold.” This limitation could encompass mentally determining the number of divisions in the claimed manner.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 2 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 2 analysis.
Claim 12
Step 1: An article of manufacture, as above.
Step 2A Prong 1: The claim recites “calculating the number of divisions n as number of divisions n = count(first label)/(count(second label) x a), where a is a coefficient.” This recites a mathematical calculation that could be performed in the mind.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 11 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 11 analysis.
Claim Rejections - 35 USC § 103
Claims 1-2, 4, and 9-10 are rejected under 35 U.S.C. 103 as being unpatentable over Schmidt (US 20210287222) (“Schmidt”) in view of Bailey et al. (US 20210350621) (“Bailey”) and further in view of Cohen et al. (US 20230046598) (“Cohen”) and Verma et al. (US 20180352045) (“Verma”).
Regarding claim 1, Schmidt discloses “[a] non-transitory computer readable medium storing a program for generating a learning model for classifying data by characterizing the data with one label among a plurality of labels, the program being executable by one or more processors to cause an information processing apparatus to execute functions (computer program may be stored in a non-transitory, tangible computer readable storage medium and may selectively activate a general-purpose computing device [processor] to perform the method – Schmidt, paragraph 109; system is for generating a classification model that can classify [characterize with a label among a plurality] imbalanced data – id. at paragraph 4) comprising:
inputting a plurality of pieces of training data required for generating the learning model, the training data including normal data and abnormal data (systems and methods may be used to classify imbalanced data, such as majority data sharing a first characteristic [normal data, “normal” by virtue of being in the minority] and minority data sharing a second characteristic [abnormal data, “abnormal” by virtue of being in the minority] – Schmidt, paragraph 2; see also paragraph 18 (disclosing that the data in question may be training data)) …;
determining whether, in a training data set including the plurality of pieces of training data, a count of a first label that characterizes a greatest amount of the normal data among the training data and a count of a second label that characterizes a smallest amount of the abnormal data among the training data are imbalanced (systems and methods may be used to classify imbalanced data, such as majority data [normal data] sharing a first characteristic [first label] and minority data [abnormal data] sharing a second characteristic [second label] – Schmidt, paragraph 2; see also paragraph 18 (disclosing that the data in question may be training data));
generating, when it is determined that the count of the first label and the count of the second label are imbalanced, a plurality of subsets each including first training data characterized by the first label and at least a portion of second training data characterized by the second label, … the plurality of subsets being generated by dividing the training data set into the plurality of subsets while updating a number of divisions so that a different combination of the first training data is included in each subset (to provide additional comparison data points for determining whether the classification model outperforms each of one or more existing classification models, the transaction classification system divides the data set into multiple subsets; each subset may be processed using the classification model such that a quantifiable loss may be determined for each of the multiple subsets; this allows the classification system to gauge the performance of the classification model using multiple comparisons from a single data set – Schmidt, paragraph 26; each subset may correspond to a time segment during which the data within the subset were obtained; for example, if the new data include data obtained over a period of several months, the new data may be divided into subsets each corresponding to a particular month of the data collection period – id. at paragraph 39 [i.e., each subset contains different data; updating the number of divisions = adding the new months]; see also paragraphs 5 (disclosing that the data in general are classified either as majority data or minority data, so in general a given months’ worth of data will contain both), 21-23 (indicating that the division of the dataset is triggered by the dataset being imbalanced, resulting in false positives and negatives), 20 (indicating that the dataset may be a training dataset)); …
saving [a] … first learning model[] when it is determined that a value of a first evaluation index for the generated … first learning model[] is higher than a value of a second evaluation index for a second learning model generated based on the training data set without generation of the plurality of subsets (if a set of criteria specify that a classification model [first learning model] is to outperform (e.g., produce a lower cost) [inverse cost of classification model = first evaluation index; inverse cost of existing models = second evaluation index] one or more existing classification models [second learning model] and those criteria are satisfied, the transaction processing system provides the classification model to the transaction processing system to fulfill the request [i.e., the classification model is saved] – Schmidt, paragraphs 26, 28; existing classification models are commercially available or control machine learning models trained to classify the data set subject to a standard loss function [i.e., they are not necessarily trained by dividing the data into subsets] – id. at paragraph 36); … [and]
outputting [a] classification result …, the … data newly classified by machine learning (transaction classification system [machine learning] may process the data set using an initial iteration of a classification model to generate an output [classification result] – Schmidt, paragraph 24) ….”
Schmidt appears not to disclose explicitly the further limitations of the claim. However, Bailey discloses that “the first training data hav[e] a count balanced with the count of the second label (to generate training data, existing animation may be augmented with multiplicative noise and data balancing may be applied to prevent common poses [second label] from being over-represented in the training data – Bailey, paragraph 105) …; [and the method further comprises]
generating a plurality of first learning models based on each subset in the generated plurality of subsets (to assist with model training, each network may be provided with only a subset of the parameters; all other parameters are excluded from the parameters that are input to the network – Bailey, paragraph 81; see also Fig. 3 (showing that there are multiple coarse models and multiple refinement models, each of which is provided with a subset of the parameters)); …
saving the plurality of first learning models (main memory may be connected to a bus and store [save] information and instructions to be executed by a processor – Bailey, paragraph 211; see also Fig. 3 (showing that the coarse models 304 and the refinement models 324 are among the instructions stored))) …;
inputting determination data (rig parameters [determination data] are used as inputs to neural networks which generate a deformation map for each of one or more mesh segments – Bailey, paragraph 64 [note that the rig parameters are determination data insofar as they are used to determine the overall deformation map result]); and
outputting [a] … result of the determination data to a user as information, the determination data newly [manipulated] by machine learning based on the stored plurality of first learning models (Bailey Fig. 3 shows a deformed mesh 316 [result of the determination data] being output based on deformation maps 306, 326 that are outputs of coarse models and refinement models [stored plurality of first learning models] that take as input rig parameters 302 [determination data]; see also paragraph 213 (disclosing that the display of the computer communicates information [including the deformed mesh] to a user of the computer)).”
Bailey and the instant application both relate to training multiple models with balanced data and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified Schmidt to generate multiple machine learning models with balanced data, as disclosed by Bailey, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would ensure the accuracy of the models by ensuring that the majority class is not overrepresented in the training data. See Bailey, paragraph 105.
Neither Schmidt nor Bailey appears to disclose explicitly the further limitations of the claim. However, Cohen discloses “integrating, by majority vote, predicted values resulting when validation data are inputted to each first learning model (plurality of training-validation cycles may be performed; in each train-validate cycle the dataset is randomly allocated to the training and validation datasets used to train a model; the best performing model or multiple good performing models may be identified, and the results combined using an ensemble learning approach; for example, each model could vote and a majority rule could be used to output the classification [integrate the predicted value by majority vote] – Cohen, paragraph 103) ….”
Cohen and the instant application both relate to ensemble learning by majority vote and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Schmidt and Bailey to predict values by majority vote after validating the models, as disclosed by Cohen, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would increase the accuracy of the overall classification by selecting the models that perform the best. See Cohen, paragraph 103.
Neither Schmidt, Bailey, nor Cohen appears to disclose explicitly the further limitations of the claim. However, Verma discloses that “the training data … [are] obtained in manufacturing industries that use plants (manufacturing plant provides a cloud service with training data that consist of several images that show what an acceptable product looks like and what typical defects in the products look like – Verma, paragraph 73) ….”
Verma and the instant application both relate to machine learning in manufacturing and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Schmidt, Cohen, and Bailey to employ the method in manufacturing plants, as disclosed by Verma, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would decrease the incidence of defective products being brought to market. See Verma, paragraph 73.
Claim 9 is an apparatus claim corresponding to non-transitory computer-readable medium claim 1 and is rejected for the same reasons as given in the rejection of that claim, except insofar as claim 9 recites an “information processing apparatus comprising: a controller; and storage”. These limitations are taught by Schmidt (computing system architecture includes system memory such as ROM and RAM [storage] and modules [controllers] that can control or be configured to control a processor to perform various actions – Schmidt, paragraph 72). Similarly, claim 10 is a method claim corresponding to non-transitory computer-readable medium claim 1 and is rejected for the same reasons as given in the rejection of that claim.
Regarding claim 2, Schmidt, as modified by Cohen, Verma, and Bailey, discloses that “the functions further comprise determining, before the generating of the plurality of subsets, a number of divisions when dividing the training data set into the plurality of subsets (each subset may correspond to a time segment during which the data within the subset were obtained; for example, if the new data include data obtained over a period of several months, the new data may be divided into subsets each corresponding to a particular month of the data collection period – Schmidt, paragraph 39 [i.e., the system predetermines that the number of divisions will be equal to the number of months of data analyzed]).”
Regarding claim 4, Schmidt, as modified by Cohen, Verma, and Bailey, discloses that “the determining of the number of divisions comprises determining the number of divisions automatically based on an initial setting (each subset may correspond to a time segment during which the data within the subset were obtained; for example, if the new data include data obtained over a period of several months, the new data may be divided into subsets each corresponding to a particular month of the data collection period – Schmidt, paragraph 39 [i.e., the system is initially set up automatically to divide the data up into monthly segments]).”
Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over Schmidt in view of Bailey, Verma, and Cohen and further in view of Liang et al. (US 20220383195) (“Liang”).
Regarding claim 3, neither Schmidt, Cohen, Verma, nor Bailey appears to disclose explicitly the further limitations of the claim. However, Liang discloses that “the determining of the number of divisions comprises determining the number of divisions based on information inputted by a user (system may receive an input from a user specifying which data that are already maintained by the system should be used for training the trainee neural network, and then divide the specified data into the training data and the validation set – Liang, paragraph 30 [i.e., the system determines that the data are to be divided into two subsets and determines which data are to be placed in which subset based on the user input]).”
Liang and the instant application both relate to machine learning algorithms that select data based on user input and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Schmidt, Cohen, Verma, and Bailey to divide the data based on user input, as disclosed by Liang, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would increase the flexibility of the system by allowing the user to determine how the data are to be divided. See Liang, paragraph 30.
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Schmidt in view of Bailey, Verma, and Cohen and further in view of Ishida (US 11170321) (“Ishida”).
Regarding claim 5, the rejection of claim 2 is incorporated. Schmidt further discloses that “the functions further comprise … updating the determined number of divisions to a different value within a predetermined range (each subset may correspond to a time segment during which the data within the subset were obtained; for example, if the new data include data obtained over a period of several months, the new data may be divided into subsets each corresponding to a particular month of the data collection period – Schmidt, paragraph 39 [i.e., when new data are received, the system updates the number of divisions of data by counting the number of months of data that have been received; predetermined range = [0,
∞
)]) ….”
Neither Schmidt, Cohen, Verma, nor Bailey appears to disclose explicitly the further limitations of the claim. However, Ishida discloses that “the functions further comprise repeatedly … calculating the first evaluation index based on each updated number of divisions, and determining the number of divisions to be the number of divisions for which the value of the first evaluation index is highest (remaining training data items obtained by excluding a test data group from training data items are divided into a first data group and a second data group; a determination process for determining whether a calculated prediction accuracy [evaluation index] is equal to or higher than a threshold is executed; when it is determined that the prediction accuracy is not equal to or higher than the threshold, the remaining data are re-divided into the first training group and the second data group; when it is determined that the prediction accuracy is equal to or higher than the threshold, an error prediction model is trained [i.e., the system determines the final number of divisions to be two, and divides the data such that the accuracy is above the threshold [i.e., highest]] – Ishida, claim 1).”
Ishida and the instant application both relate to machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Schmidt, Cohen, Verma, and Bailey to redivide the data until an evaluation index reaches a threshold, as disclosed by Ishida, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would ensure that the accuracy of the resulting model is acceptable for the use case. See Ishida, col. 1, l. 64-col. 2, l. 10.
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Schmidt in view of Bailey, Verma, and Cohen and further in view of Lyonnet et al. (US 20190318421) (“Lyonnet”).
Regarding claim 7, neither Schmidt, Cohen, Verma, nor Bailey appears to disclose explicitly the further limitations of the claim. However, Lyonnet discloses that “the generating of the plurality of subsets comprises generating another subset by newly sampling the first training data from the training data set after excluding, from the training data set, the first training data sampled into one subset (bootstrap aggregated (“bagged”) tree technique may be employed that builds multiple decision trees by repeatedly resampling training data without replacement – Lyonnet, paragraph 33 [note that resampling without replacement entails excluding the first data sampled]).”
Lyonnet and the instant application both relate to machine learning and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Schmidt, Cohen, Verma, and Bailey to resample the training data without replacement, as disclosed by Lyonnet, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would increase the diversity of the models trained by ensuring that each is trained on a different set of data. See Lyonnet, paragraph 33.
Claims 11-12 are rejected under 35 U.S.C. 103 as being unpatentable over Schmidt in view of Bailey, Verma, and Cohen and further in view of Chowdhury et al. (US 20180032900) (“Chowdhury”).
Regarding claim 11, the rejection of claim 2 is incorporated. The combination of Schmidt, Bailey, Verma, and Cohen teaches that “the functions further comprise determining the number of divisions,” as shown above in the rejections of claims 1-2.
Neither Schmidt, Bailey, Verma, nor Cohen appears to disclose explicitly the further limitations of the claim. However, Chowdhury discloses that “a ratio of a count of the first label to a count of the second label in one subset is equal to or less than a threshold (second threshold is [equal to] a ratio of a number of positive labeled images [count of first label] divided by a number of labeled negative instances [count of second label] – Chowdhury, paragraph 8).”
Chowdhury and the instant application both relate to machine learning with imbalanced data and are analogous. It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Schmidt, Bailey, Verma, and Cohen to use a ratio of the count of a first label to that of a second label as the threshold, as disclosed by Chowdhury, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would improve the performance of the model by ensuring that data imbalances are taken into account when training. See Chowdhury, paragraphs 4-5.
Regarding claim 12, the rejection of claim 11 is incorporated. The combination of Schmidt, Bailey, Verma, and Cohen teaches that “the functions further comprise calculating the number of divisions n as number of divisions n”, as shown in the rejections of claims 1-2 and 11.
Neither Schmidt, Bailey, Verma, nor Cohen appears to disclose explicitly the further limitations of the claim. However, Chowdhury discloses that “n = count(first label)/(count(second label) x a), where a is a coefficient (second threshold is a ratio of a number of positive labeled images [count of first label] divided by a number of labeled negative instances [count of second label] – Chowdhury, paragraph 8 [note that a = 1 in this example]).” It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to have modified the combination of Schmidt, Bailey, Verma, and Cohen to use a ratio of the count of a first label to that of a second label as the threshold, as disclosed by Chowdhury, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would improve the performance of the model by ensuring that data imbalances are taken into account when training. See Chowdhury, paragraphs 4-5.
Response to Arguments
Applicant's arguments filed January 23, 2026 (“Remarks”) have been fully considered but they are, except insofar as rendered moot by the introduction of a new ground of rejection, not persuasive.
Regarding the eligibility rejection, Applicant argues (a) that the amended claims as a whole are directed to an improvement in machine learning technology by suppressing overtraining; eliminating data bias; maintaining data fidelity; optimizing model architecture; and forming optimal decision boundaries and that these improvements are reflected in the specification. Applicant further argues (b) that the claims allegedly do not recite abstract ideas; and (c) that the recent changes to the MPEP in light of Ex parte Desjardins allegedly compel the conclusion that the claims are directed to an improvement in technology because it is not necessary that the improvement be explicitly recited in the claims and the claim limitations reflect the improvements indicated in argument (a). Remarks at 7-17.
Applicant’s arguments are unconvincing. Argument (b) is unconvincing at least because the claims provide no limits on how the imbalance in the training data is determined; how the generation of the balanced training data is accomplished (other than to say that the data are divided into subsets whole updating a number of divisions, which are themselves mentally performable processes); and how the predicted values are integrated, among other limitations. Thus, all of these limitations encompass mentally performable steps but for the recitation that they are performed by a generic machine learning algorithm, which is itself merely an instruction to apply the exception on a computer. MPEP § 2106.05(f).1 Arguments (a) and (c) may be considered together, as they are related. It suffices to make two observations. The first is that the large majority of the limitations that Applicant puts forth as allegedly providing a practical application – e.g., determining whether data are imbalanced, generating subsets of training data, generating multiple learning models, integrating predictions by majority vote, and comparing evaluation indices – are themselves part of the judicial exception itself, and as such cannot provide a practical application of the judicial exception. MPEP § 2106.05(I). The second is that, while the improvement need not be explicitly recited, it is at least necessary that the improvement be evident from the additional elements in ordered combination with the remainder of the claim. Here, the only additional elements of the claim recite either insignificant extra-solution activity that is well-understood, routine, and conventional or mere instructions to apply the exception on a computer programmed with a generic class of computer algorithm. These alone cannot provide the improvement.
Regarding the art rejection, Applicant argues that Schmidt allegedly fails to disclose that the number of divisions of data is actively updated as a design variable for the purpose of increasing machine learning model performance because Schmidt merely passively bins the data by months. Remarks at 17-20. This argument is unconvincing at least because the claims do not require that the data partitioning scheme be actively updated for the purpose of improving machine learning model performance. At most, the amended claims require that the training data be “divid[ed] … into the plurality of subsets while updating a number of divisions”. As Applicant itself notes, the system of Schmidt collects training data from various months and partitions the data by month. Thus, every time a new month of data is added, the “number of divisions” increases by one – i.e., is updated.
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to RYAN C VAUGHN whose telephone number is (571)272-4849. The examiner can normally be reached M-R 7:00a-5:00p ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Kamran Afshar, can be reached at 571-272-7796. 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.
/RYAN C VAUGHN/ Primary Examiner, Art Unit 2125
1 Note that the claims merely recite “generating a plurality of first learning models”, not “generating a plurality of first machine learning models”. Examiner would agree that this limitation would not be practically mentally performable if it were recited that machine learning models are generated. However, the broader term “learning models” actually recited arguably encompasses mental models.