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 1-2, 5-10, 12-13, and 16-22 are presented for examination.
Continued Examination under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on August 26, 2026 has been entered.
Response to Amendment
Applicant’s amendment has obviated most, but not all, of the claim objections, but has not obviated any of the specification objections. To the extent that an objection or rejection appears in the previous Office Action(s) but not this Office Action, that objection or rejection is withdrawn. To the extent that it appears both in a previous Office Action(s) and this Office Action, the objection or rejection is maintained.
Specification
The lengthy specification has not been checked to the extent necessary to determine the presence of all possible minor errors. Applicant’s cooperation is requested in correcting any errors of which applicant may become aware in the specification.
The abstract of the disclosure is objected to because (a) an “and” should precede “false negatives” in the penultimate sentence, and (b) “are generated” in the penultimate sentence should be “is generated”. A corrected abstract of the disclosure is required and must be presented on a separate sheet, apart from any other text. See MPEP § 608.01(b).
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 1-2, 5-10, 12-13, and 16-22 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”).
Claims 1 and 21
Step 1: The claims recite a method; therefore, they are directed to the statutory category of processes.
Step 2A Prong 1: The claims recite, inter alia:
[A]ccepting, … a single input that contains a plurality of objects, including inferring a single inference that contains an inferred frequency of each class of at least three classes:1 This limitation could encompass mentally classifying the objects and mentally inferring the frequency of each class based thereon.
[G]enerating a respective upscaled magnitude of each class of the at least three classes from the inferred frequency of the class by using a multiplicand that is based on the plurality of objects, wherein the multiplicand is not based on a count of the at least three classes: This limitation could encompass mentally generating the upscaled magnitude using a multiplicand not based on the count of the classes. Upscaling is also a mathematical concept.
[M]easuring a validation metric that is not based on a total of true negatives, the measuring comprising: a) generating a respective frequency integer of each class of the at least three classes from the upscaled magnitude of the class: This limitation could encompass mentally generating the validation metric by mentally generating the integer.
b) [E]stimating, based on said frequency integers of the at least three classes and a target integer respectively for each class of the at least three classes: a count of the plurality of objects that are true positives of the class, a count of the plurality of objects that are false positives of the class, and a count of the plurality of objects that are false negatives of the class: This limitation could encompass mentally generating the counts of true positives, false positives, and false negatives based on the integers and the target integer.
c) [G]enerating, based on the counts of true positives of the at least three classes, an estimated total of true positives that characterizes fitness of the trained classifier: This limitation could encompass mentally estimating a total of true positives and using that total as a measure of classifier fitness.
d) [G]enerating, based on the counts of false positives of the at least three classes, an estimated total of false positives that characterizes the fitness of the trained classifier: This limitation could encompass mentally estimating a total of false positives and using that total as a measure of classifier fitness.
e) [G]enerating, based on the counts of false negatives of the at least three classes, an estimated total of false negatives that characterizes fitness of the trained classifier; wherein the validation metric is based on at least two of: the estimated total of true positives, the estimated total of false positives, and the estimated total of false negatives: This limitation could encompass mentally estimating a total of false negatives and using that total as a measure of classifier fitness, then computing the validation metric mentally based on at least two of the three claimed items.
[D]etecting that the validation metric exceeds a predefined threshold: This limitation could encompass mentally detecting that the validation metric exceeds a threshold.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claims further recite that the inferred frequency is generated “by [a] trained classifier” and that “the method is performed by one or more computers.” However, these are mere instructions to apply the judicial exception using a generic computer programmed with a generic class of computer algorithm. MPEP § 2106.05(f).
Step 2B: The claim does not contain significantly more than the judicial exception. The analysis at this step mirrors that of step 2A, prong 2. As an ordered whole, the claims are directed to a mentally performable process of estimating true positives, false positives, and false negatives from the outputs of a classifier. Nothing in the claims provides significantly more than this. As such, the claims are not patent eligible.
Claim 2
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “said generating the estimated total of false positives comprises summing the counts of false positives of the at least three classes.” This is a mathematical concept; moreover, summing the counts could be performed in the mind given sufficiently small counts.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim 5
Step 1: A process, as above.
Step 2A Prong 1: The claim recites “predefining a distinct weight for each class of the at least three classes; and said generating the estimated total of false negatives comprises using the weights of the at least three classes as multiplicands.” Predefining a weight may be performed mentally, and generating a total using multiplicands is a mathematical concept that may be executed mentally given sufficiently simple data.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim 6
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the weight of each class of the at least three classes is less than one, and the weights of the at least three classes sum to one.” Predefining the weights remains mentally performable under these further assumptions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 5 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 5 analysis.
Claim 7
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “said generating the frequency integers of the at least three classes comprise one selected from a group consisting of: rounding up and rounding down.” Rounding is a mathematical concept and may be performed mentally.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim 8
Step 1: A process, as above.
Step 2A Prong 1: The claim recites “generating the plurality of objects from a parse tree; and generating the target integer for each class of the at least three classes based on a frequency of a respective n-gram in the parse tree.” The first limitation could encompass a human visually observing the parse tree and generating the objects based thereon. As for the second, generating the target integer remains mentally performable under these further assumptions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The claim further recites that “the trained classifier is a single neural network”. However, this amounts to a mere instruction to apply the judicial exception using a generic computer programmed with a generic class of computer algorithm. MPEP § 2106.05(f).
Step 2B: The claim does not contain significantly more than the judicial exception. The claim further recites that “the trained classifier is a single neural network”. However, this amounts to a mere instruction to apply the judicial exception using a generic computer programmed with a generic class of computer algorithm. MPEP § 2106.05(f).
Claim 9
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “generating the upscaled magnitude comprises using a multiplicand that is based solely on the parse tree.” Generating the upscaled magnitude remains a mathematical concept/mental process under these further assumptions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 8 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 8 analysis.
Claim 10
Step 1: A process, as above.
Step 2A Prong 1: The claim recites that “the estimated total of true positives, the estimated total of false positives, and the estimated total of false negatives are fractions that are less than one; and a sum of said fractions is less than one half.” Estimating these values remains mentally performable under these further conditions.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claims 12-13, 16-20
Step 1: The claims recite a non-transitory computer-readable medium; therefore, they are directed to the statutory category of articles of manufacture.
Step 2A Prong 1: The claims recite the same judicial exceptions as in claims 1-2 and 5-9, respectively, except insofar as claim 12 recites the original inferring limitation, which makes no substantive difference in the analysis.
Step 2A Prong 2: This judicial exception is not integrated into a practical application. The analysis at this step mirrors that of claims 1-2 and 5-9, respectively, except insofar as these claims additionally recite “[o]ne or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, [perform the method]”. However, this amounts to 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 mirrors that of claims 1-2 and 5-9, respectively, except insofar as these claims additionally recite “[o]ne or more non-transitory computer-readable media storing instructions that, when executed by one or more processors, [perform the method]”. However, this amounts to a mere instruction to apply the judicial exception using a generic computer. MPEP § 2106.05(f).
Claim 22
Step 1: A process, as above.
Step 2A, Prong 1: The claim recites that “the upscaling multiplicand is a count of: n-grams in a parse tree or distinct n-grams in the parse tree.” Upscaling the multiplicand remains a mathematical concept under these further assumptions.
Step 2A, Prong 2: This judicial exception is not integrated into a practical application. See claim 1 analysis.
Step 2B: The claim does not contain significantly more than the judicial exception. See claim 1 analysis.
Claim Rejections - 35 USC § 103
Claims 1-3, 11-13, and 21 are rejected under 35 U.S.C. 103 as being unpatentable over Amershi et al. (US 20180046935) (“Amershi”) in view of Csordás et al. (US 10380753) (“Csordás”) and further in view of Hsiung et al. (US 20190236333) (“Hsiung”).
Regarding claim 1, Amershi discloses “[a] validation method for a trained classifier, comprising:
accepting, by the trained classifier, a single input that contains a plurality of objects, including inferring a single inference that contains an inferred frequency of each class of at least three classes (many performance metrics in multi-class classification [i.e., classification of at least three classes] are derived from different categories of prediction counts; for example, accuracy is computed as the number of correct predictions over the total number of predictions (correct and incorrect) – Amershi, paragraph 49; analysis tool can count the amount of items designated as true positives, false positives, and false negatives for each class as well as for overall performance across classes – id. at paragraph 50 [note that the total of these three plus true negatives equals an inferred frequency of each class, and the set of all of these totals may be regarded as a single inference]; see also paragraphs 44 (disclosing that the multi-class classifiers are built by training binary classifiers and combining their outputs), 46 (disclosing that the calculations are performed based on a test data set accepted by the model, which can be regarded as a single input containing multiple objects)); …
generating a respective … magnitude of each class of the at least three classes from the inferred frequency of the class (analysis tool can count the amount of items designated as true positives, false positives, and false negatives for each class as well as for overall performance across classes; the amount [frequency] of items falling into each category [class] can be broken down into the amount of items falling into confidence score ranges [magnitudes] – Amershi, paragraph 50) …;
measuring a validation metric that is not based on a total of true negatives (accuracy is computed as the number of correct predictions over the total number of predictions (correct and incorrect); precision is computed as the number of true positives over the number of true and false positives and recall is the number of true positives over the number of true positives and false negatives [note that neither precision nor recall is based on true negatives] – Amershi, paragraph 49), the measuring comprising:
a) generating … a respective frequency integer of each class of the at least three classes from … the class (Amershi paragraph 62 describes the classes as being labeled 0-9; paragraph 49 discloses that the performance metrics are derived from categories of prediction counts including the total number of predictions [frequency integer]; paragraph 50 discloses that this count is performed for each class);
b) estimating, based on said frequency integers of the at least three classes and a target integer respectively for each class of the at least three classes:
a count of the plurality of objects that are true positives of the class,
a count of the plurality of objects that are false positives of the class, and
a count of the plurality of objects that are false negatives of the class (bidirectional bar graphs are concurrently displayed for each class available in a multi-class classifier; the true positives are portrayed in the color associated with the class; the false positives are displayed in a color associated with the class into which the item should have been classified; the false negatives are portrayed in the color of the class into which they were actually assigned – Amershi, paragraph 29; item of test data labeled as a 4 should be classified as a 4, but if it is classified as a 6 (a false positive), it will be depicted on the right side of the class 6 bidirectional graph in the color associated with class 4 [i.e., each classification is based on the actual integer to which the class being [i.e., the target class] and the integer predicted by the classifier] – id. at paragraph 67);
c) generating, based on the counts of true positives of the at least three classes, an estimated total of true positives that characterizes fitness of the trained classifier;
d) generating, based on the counts of false positives of the at least three classes, an estimated total of false positives that characterizes the fitness of the trained classifier; and
e) generating, based on the counts of false negatives of the at least three classes, an estimated total of false negatives that characterizes fitness of the trained classifier; [and]
wherein the validation metric is based on at least two of: the estimated total of true positives, the estimated total of false positives, and the estimated total of false negatives (accuracy is computed as the number of correct predictions over the total number of predictions (correct and incorrect); precision is computed as the number of true positives over the number of true and false positives and recall is the number of true positives over the number of true positives and false negatives [i.e., the true positives, false positives, and false negatives are aggregated across classes and used in precision and recall metrics that characterize the fitness of the classifier] – Amershi, paragraph 49); …
wherein the method is performed by one or more computers (Amershi Fig. 1 shows that the method is performed on a computer comprising a processor and a memory).”
Amershi appears not to disclose explicitly the further limitations of the claim. However, Csordás discloses “generating a respective upscaled magnitude (input disparity maps are upsampled in upsampling units to the next scale [upscaled magnitude]; in an exemplary case where the scale is 2, units are labeled by “x2” – Csordás, col. 30, l. 46-col. 31, l. 6) … by using a multiplicand that is based on the plurality of objects, wherein the multiplicand is not based on a count of the at least three classes (input disparity maps are upsampled in upsampling units to the next scale; in an exemplary case where the scale is 2, units are labeled by “x2” [2 = multiplicand]; a number being a power of 2 is a typical choice for the upscaling factor, but other integers can be used as the upscaling factor, taking into consideration the scale of the pair of feature maps [i.e., not taking into consideration a count of classes] – Csordás, col. 30, l. 46-col. 31, l. 6); [and]
generating [data] from the upscaled magnitude (output of left upscaling unit is given to a warping unit for warping the right feature map [warped map = data generated from upscaled magnitude] – Csordás, col. 31, ll. 17-29) ….”
Csordás 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 Amershi to upscale the magnitude and generate data therefrom, as disclosed by Csordás, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the data to be represented at a larger scale than previously, thereby magnifying its effect on downstream data when needed. See Csordás, col. 30, l. 46-col. 31, l. 6.
Neither Amershi nor Csordás appears to disclose explicitly the further limitations of the claim. However, Hsiung discloses “generating, after said inferring, a … class (Hsiung Figure 6 shows that information identifying a classification 670 [generating a class] is performed after performing one or more spectroscopic classifications 650 [inferring]) …; … [and]
detecting that the validation metric exceeds a predefined threshold (control device may determine that the classification model has an accuracy that exceeds the validation threshold – Hsiung, paragraph 66) ….”
Hsiung and the instant application both relate to validation of classifiers 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 Amershi and Csordás to detect that the validation metric exceeds a threshold, as disclosed by Hsiung, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would ensure that the classifier is sufficiently accurate to perform what it was trained to perform. See Hsiung, paragraph 66.
Claim 12 is identical to its counterpart in the last claim set presented on April 14, 2026. Therefore, the rejection of claim 12 in the final rejection of April 29, 2026 is maintained and is incorporated into this action by this reference. Similarly, claim 21 corresponds to claim 1 except insofar as claim 21 recites “a plurality of classes” instead of “at least three classes”. However, since three classes constitute a plurality, the rejection of claim 1 is equally applicable to claim 21.
Regarding claim 2, Amershi, as modified by Csordás/Hsiung, discloses that “said generating the estimated total of false positives comprises summing the counts of false positives of the at least three classes (analysis tool counts the amount of items designated as true positives, false positives, and false negatives for each class as well as for overall performance across classes [e.g., by summing the false positives across classes] – Amershi, paragraph 50).”
Claim 13 is a non-transitory computer-readable medium claim corresponding to method claim 2 and is rejected for the same reasons as given in the rejection of that claim.
Regarding claim 11, Amershi, as modified by Csordás/Hsiung, discloses that “the trained classifier is a single neural network (neural network-based feature extractor branch [part of a single neural network] is applied on an image – Csordás, col. 4, ll. 38-49).” 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 Amershi to employ a neural network, as disclosed by Csordás, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the user to utilize off-the-shelf models without having to build them manually. See Csordás, col. 4, ll. 38-49.
Claims 5-6 and 16-17 are rejected under 35 U.S.C. 103 as being unpatentable over Amershi in view of Csordás and vHsiung and further in view of Kalech et al. (US 20180173599) (“Kalech”).
Regarding claim 5, the rejection of claim 1 is incorporated. Amershi further discloses “each class of the at least three classes” and “generating the estimated total of false negatives”, as shown in the rejection of claim 1.
Neither Amershi, Hsiung, nor Csordás appears to disclose explicitly the further limitations of the claim. However, Kalech discloses that “the method further comprises predefining a distinct weight for each class (one approach to combine fault likelihood estimates [classes] is by using some weighted linear combination, such that the weights are positive and sum up to one – Kalech, paragraph 85) …; and
said generating the estimated total … comprises using the weights of the … classes as multiplicands (one approach to combine fault likelihood estimates [classes] is by using some weighted linear combination [total that uses weights as multiplicands], such that the weights are positive and sum up to one – Kalech, paragraph 85).”
Kalech and the instant application both relate to artificial intelligence 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 Amershi, Hsiung, and Csordás to use a weight for each class as a multiplicand, as disclosed by Kalech, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the weights to represent probability estimates, thereby increasing their utility. See Kalech, paragraph 85.
Claim 16 is a non-transitory computer-readable medium claim corresponding to method claim 5 and is rejected for the same reasons as given in the rejection of that claim.
Regarding claim 6, the rejection of claim 5 is incorporated. Amershi further discloses “at least three classes”, as shown in the rejection of claim 1.
Neither Amershi, Hsiung, nor Csordás appears to disclose explicitly the further limitations of the claim. However, Kalech discloses that “the weight of each class of the … classes is less than one, and
the weights of the … classes sum to one (one approach to combine fault likelihood estimates [classes] is by using some weighted linear combination, such that the weights are positive and sum up to one [implying that the weight of each class is less than one] – Kalech, paragraph 85).” 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 Amershi, Hsiung, and Csordás to use weights that sum to one, as disclosed by Kalech, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the weights to represent probability estimates, thereby increasing their utility. See Kalech, paragraph 85.
Claim 17 is a non-transitory computer-readable medium claim corresponding to method claim 6 and is rejected for the same reasons as given in the rejection of that claim.
Claims 7 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Amershi in view of Csordás and Hsiung and further in view of Soni et al. (US 20200120131) (“Soni”).
Regarding claim 7, the rejection of claim 1 is incorporated. Amershi further discloses “generating the integers of the at least three classes”, as shown above in the rejection of claim 1.
Neither Amershi, Hsiung, nor Csordás appears to disclose explicitly the further limitations of the claim. However, Soni discloses that “said generating the [data] … comprise[s] one selected from a group consisting of: rounding up and rounding down (in a model of k classes corresponding to protocols, the model predicts a real number, and this real number is rounded to the nearest classification in the range of integers [0, k – 1] [i.e., the real number is either rounded up or rounded down to the nearest integer] – Soni, paragraph 100).”
Soni 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 Amershi, Hsiung, and Csordás to round the data up or down, as disclosed by Soni, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow continuous quantities to be represented as discrete classes, thereby saving processing power. See Soni, paragraph 100.
Claim 18 is a non-transitory computer-readable medium claim corresponding to method claim 7 and is rejected for the same reasons as given in the rejection of that claim.
Claims 8-9, 19-20, and 22 are rejected under 35 U.S.C. 103 as being unpatentable over Amershi in view of Csordás and Hsiung and further in view of Ehlen et al. (US 20150186790) (“Ehlen”).
Regarding claim 8, the rejection of claim 1 is incorporated. Amershi further discloses “generating the target integer for each class of the at least three classes”, as shown above in the rejection of claim 1. Csordás further discloses that “the trained classifier is a single neural network (neural network-based feature extractor branch [part of a single neural network] is applied on an image – Csordás, col. 4, ll. 38-49).” 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 Amershi to employ a neural network, as disclosed by Csordás, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the user to utilize off-the-shelf models without having to build them manually. See Csordás, col. 4, ll. 38-49.
Neither Amershi, Hsiung, nor Csordás appears to disclose explicitly the further limitations of the claim. However, Ehlen discloses “generating the plurality of objects from a parse tree (using tokenized sentences, a parse tree for each sentence may be generated; any noun, noun-noun, or adjective-noun combinations [objects] are extracted from each noun phrase – Ehlen, paragraph 41); and
generating the [data] … based on a frequency of a respective n-gram in the parse tree (phrases or n-grams common to a trigger are extracted and counted, and phrases or n-grams are scored by frequency for their relevancy [data] to the buying decision trigger – Ehlen, paragraph 77; see also paragraphs 58-63 (disclosing that the n-grams are part of a parsing algorithm, e.g., a parse tree)).”
Ehlen and the instant application both relate to parse trees 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 Amershi, Hsiung, and Csordás to generate data based on frequency of elements of a parse tree, as disclosed by Ehlen, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the system to understand the syntactical structure of the objects being parsed. See Ehlen, paragraph 41.
Claim 19 is a non-transitory computer-readable medium claim corresponding to method claim 8 and is rejected for the same reasons as given in the rejection of that claim.
Regarding claim 9, the rejection of claim 8 is incorporated. Csordás further discloses that “said generating the upscaled magnitude comprises using a multiplicand”, as shown in the rejection of claim 1. 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 Amershi to perform upscaling using a multiplicand, as disclosed by Csordás, for substantially the reasons given in the rejection of claim 1.
Neither Amershi, Hsiung, nor Csordás appears to disclose explicitly the further limitations of the claim. However, Ehlen discloses “using a [datum] that is based solely on the parse tree (using tokenized sentences, a parse tree for each sentence may be generated; any noun, noun-noun, or adjective-noun combinations [objects] are extracted from each noun phrase and added to a set of attribute candidates [used data based solely on the parse tree] – Ehlen, paragraph 41).” 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 Amershi, Hsiung, and Csordás to derive data from a parse tree, as disclosed by Ehlen, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the system to understand the syntactical structure of the objects being parsed. See Ehlen, paragraph 41.
Claim 20 is a non-transitory computer-readable medium claim corresponding to method claim 9 and is rejected for the same reasons as given in the rejection of that claim.
Regarding claim 22, the rejection of claim 1 is incorporated. Csordás further discloses an “upscaling multiplicand,” as shown in the rejection of claim 1. 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 Amershi/Hsiung to employ an upscaling multiplicand, as disclosed by Csordás, for substantially the same reasons as given in the rejection of claim 1.
Neither Amershi, Csordás, nor Hsiung appears to disclose explicitly the further limitations of the claim. However, Ehlen discloses “a count of: n-grams in a parse tree or distinct n-grams in the parse tree (phrases or n-grams common to a trigger are extracted and counted, and phrases or n-grams are scored by frequency [count] for their relevancy to the buying decision trigger – Ehlen, paragraph 77; see also paragraphs 58-63 (disclosing that the n-grams are part of a parsing algorithm, e.g., a parse tree)).” 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 Amershi, Hsiung, and Csordás to count n-grams from a parse tree, as disclosed by Ehlen, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would allow the system to understand the syntactical structure of the objects being parsed. See Ehlen, paragraph 41.
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Amershi in view of Csordás and Hsiung and further in view of Burke et al. (WO 2017079398) (“Burke”).
Regarding claim 10, the rejection of claim 1 is incorporated. Amershi further discloses “the estimated total of true positives, the estimated total of false positives, and the estimated total of false negatives”, as shown in the rejection of claim 1.
Neither Amershi, Hsiung, nor Csordás appears to disclose explicitly the further limitations of the claim. However, Burke discloses that “the estimated total[s] … are fractions that are less than one; and
a sum of said fractions is less than one half (values of e(C, f) may be summed for small fractions for each conditions, e.g., the maximum is selected for the sum over a set of values f1, f2, f3, … for two or more fractions less than half and not equal to zero for one condition [e.g., for f1 and f2 < ¼, the sum is less than one half] – Burke, paragraph 54).”
Burke 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 Amershi, Hsiung, and Csordás to use fractional weights that sum to less than one half, as disclosed by Burke, and an ordinary artisan could reasonably expect to have done so successfully. Doing so would prevent certain objects from being excessively weighted. See Burke, paragraph 54.
Response to Arguments
Applicant's arguments filed July 22, 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 art rejection, Applicant argues that (a) Amershi does not disclose the newly claimed “frequency integer”, (b) that Amershi does not disclose a “single inference”, and (c) that Examiner is allegedly improperly providing a motivation to combine Kalech with itself and that Kalech allegedly disavows the use of weights to represent probability estimates. Remarks at 13-14. Regarding (a), Amershi discloses that the accuracy analysis includes taking a full count of the members of each class, which corresponds to the claimed “frequency integer”. Regarding (b), the claim does not require that there be only a single inference; insofar as a set of inferences is dividable into single inferences, a system that performs multiple discrete inferences discloses a “single inference”. Moreover, the entire set of inferences made by the system of Amershi may be regarded as a single inference. Regarding (c), Examiner is clearly not combining any reference with itself, but rather providing a reason why an ordinary artisan would be motivated to modify the other references such that the weights are each less than one and sum to one, as disclosed by Kalech. Moreover, the portions of paragraphs 85-86 of Kalech cited by Applicant contain nothing that can be reasonably characterized as a disavowal, and the use of the use of the weights as probabilities is cited as a motivation to modify the other references. Even assuming arguendo that Kalech does not disclose the use of weights as probability estimates, which Examiner does not concede, it can be easily gleaned by an ordinary artisan that setting the weights to the values claimed would allow for their representation as probabilities. Examiner reminds Applicant that the motivation to combine may be ascertained from general knowledge of the art as well as from the references themselves. MPEP § 2143(I)(G).
Regarding the rejections under 35 USC § 101, as best understood by Examiner, Applicant appears to be arguing (a) that the measuring step is performed in an unconventional way; and (b) that the claims as a whole are directed to a solution to a technical problem of accelerating neural network inference that is reflected at least in the recitation of inferring a single inference that contains an inferred frequency of each class, thereby integrating any judicial exception recited into a practical application. Remarks at 15-19. However, regarding (a), note that the measuring is part of the judicial exception itself and cannot provide the inventive concept. MPEP § 2106.05(I). Therefore, the question of whether the measuring is conventional or not is irrelevant to the analysis because the conventionality of a claim limitation is only a consideration for step 2B and the measuring limitation is analyzed at step 2A, prong 1. Argument (b) can be disposed of in much the following manner, namely by noting that performing the single inference is part of the abstract idea itself. Even assuming arguendo that the specification discloses a technical improvement to the field of neural network acceleration, which Examiner does not concede, this improvement is not reflected in the limitations cited by Applicant because those limitations are part of the judicial exception itself.
Conclusion
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 Claim 21 is identical to claim 1 except insofar as claim 21 recites “a plurality of classes” instead of “at least three classes”. However, this does not make a substantive difference in the eligibility analysis.
Prosecution Insights