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 . This action is in response to an amendment filed on April 6th, 2026. Claims 1-20 are pending in the current application. Claims 1-3, and 11-13 are amended in the current application.
Claim Rejections - 35 USC § 112
Claims 2 and 12 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 2 and 12 recites the limitation “(1-a2)n3”. The variables are not defined, and therefore, there is insufficient antecedent basis for this limitation in the claim which renders the claim indefinite.
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 towards a process an abstract idea without significantly more.
Regarding claim 1 Under Step 1 of the Subject Matter Eligibility Test of Products and
Processes, the claim is directed towards a process, which is one of the four statutory categories.
Next, under a Step 2A Prong 1 Analysis, the following limitations in the claim recites an abstract idea.
generating an initial ground set based on a first candidate set of outer-If conditions (SD), and a second candidate set used for selecting Inner-If or Then conditions (RL); (mental process)
evaluating a fixed number of triples and forming a new ground set that provides a recourse accuracy level above a reference threshold, wherein the fixed number of triples included in the new ground set is less than a total number of triples included in the initial ground set, and wherein each triple included in the new ground set increases below-threshold recourse accuracy levels of the new ground set; (mental process)
sorting the new ground set by recourse accuracy; (mental process)
producing a two-level recourse set by selecting a predetermined number of triples based on corresponding recourse accuracies indicated in the sorting; (mental process)
performing calculation based on the selected number of triples. (mental process)
and constructing the global counterfactual explanation based on at least the performing of the calculation. (mental process)
Therefore, we have to examine the claim under Step 2A prong 2, which considers the additional elements within the claim. The claim does not contain any additional elements. Therefore there are no additional elements that integrate the abstract idea into a practical application. The claim is directed to an abstract idea.
Under a Step 2B analysis, since the claim does not contain any additional elements, there is nothing that amounts to significantly more than the judicial exception as explained above in Step 2A prong 2. Therefore, the claim is ineligible.
Regarding claim 11 Under Step 1 of the Subject Matter Eligibility Test of Products and
Processes, the claim is directed towards a machine, which is one of the four statutory categories.
Next, under a Step 2A Prong 1 Analysis, the following limitations in the claim recites an abstract idea.
generating an initial ground set based on a first candidate set of outer-If conditions (SD), and a second candidate set used for selecting Inner-If or Then conditions (RL); (mental process)
evaluating a fixed number of triples and forming a new ground set that provides a recourse accuracy level above a reference threshold, wherein the fixed number of triples included in the new ground set is less than a total number of triples included in the initial ground set, and wherein each triple included in the new ground set increases below-threshold recourse accuracy levels of the new ground set; (mental process)
sorting the new ground set by recourse accuracy; (mental process)
producing a two-level recourse set by selecting a predetermined number of triples based on corresponding recourse accuracies indicated in the sorting; (mental process)
performing calculation based on the selected number of triples. (mental process)
and constructing the global counterfactual explanation based on at least the performing of the calculation. (mental process)
Therefore, we have to examine the claim under Step 2A prong 2, which considers the additional elements within the claim. The claim’s additional elements are:
at least one processor
at least one memory
at least one communication circuit
The limitations, as drafted, is interpreted to be “generally linked” to the abstract idea, as the processor, memory, and circuit are merely used to perform the abstract idea. (See MPEP 2106.05(h)) Therefore, the additional elements do not integrate the abstract idea into a practical application. The claim is directed to an abstract idea.
Under a Step 2B analysis, the claim' s additional elements do not amount to significantly more than the judicial exception as explained above in Step 2A prong 2. Therefore, the claim is ineligible.
Regarding claims 2 and 12, the claims recite “storing up to (1-α2)n3 iterations from among the initial ground set, wherein the generating of the initial ground set is performed by iterating over the second candidate set in O(n) time and computing feature combinations, before removing any items that contain a feature combination that only occurs once, for yielding a new RL with size an, and wherein α is greater than or equal to 0 and less than or equal 1.” The limitation, “storing up to (1-α2)n3 iterations from among the initial ground set” is considered to be insignificant extra-solution activity, (See MPEP 2106.05(g)) as well as well-understood routine and conventional, as it is considered to be storing or retrieving data in memory. (See MPEP2106.05(d)(II)) The limitation, “the generating of the initial ground set is performed by iterating over the second candidate set in O(n) time and computing feature combinations, before removing any items that contain a feature combination that only occurs once, for yielding a new RL with size an, and wherein α is greater than or equal to 0 and less than or equal 1.”, is interpreted to be, under the broadest reasonable interpretation, a “mental process”. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
Regarding claims 3 and 13, the claims recite “generating of the initial ground set is performed by filtering a dataset based on the outer-If or the inner-If conditions, and separately deploying a method for generating Then conditions.” The limitation, as drafted, is interpreted to be, under the broadest reasonable interpretation, a “mental process”. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
Regarding claims 4 and 14, the claims recite “each triple includes an outer-If condition, an inner-If condition, and a Then condition.” The limitation, as drafted, is interpreted to be, under the broadest reasonable interpretation, part of a “mental process”, as it contains further details of the judicial exception and the claims do not recite any further additional elements. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
Regarding claims 5 and 15, the claims recite “the selecting of the predetermined number of triples includes selecting highest-performing triples within the new ground set.” The limitation, as drafted, is interpreted to be, under the broadest reasonable interpretation, a “mental process”. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
Regarding claims 6 and 16, the claims recite “each triple forming the new ground set increases the recourse accuracy level.” The limitation, as drafted, is interpreted to be, under the broadest reasonable interpretation, part of a “mental process”, as it contains further details of the judicial exception and the claims do not recite any further additional elements. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
Regarding claims 7 and 17, the claims recite “one or more constraints are applied during the generating of the initial ground set.” The limitation, as drafted, is interpreted to be, under the broadest reasonable interpretation, part of a “mental process”, as it contains further details the judicial exception and the claims do not recite any further additional elements. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
Regarding claims 8 and 18, the claims recite “the initial ground set removes a feature combination that only occurs once.” The limitation, as drafted, is interpreted to be, under the broadest reasonable interpretation, a “mental process”. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
Regarding claims 9 and 19, the claims recite “an upper bound defined as acc(R) < acc(V) is reached before an algorithm for providing the global counterfactual explanation has completed execution, wherein acc(R) is a percentage of instances in Xaff that are provided with a successful recourse, wherein Xaff is a set of individuals with an unfavorable prediction from a model, and wherein acc(v) is a recourse accuracy.” The limitation, as drafted, is interpreted to be, under the broadest reasonable interpretation, a “mental process”. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
Regarding claims 10 and 20, the claims recite “the algorithm is terminated prior to its completion when the upper bound for saturation is reached.” The limitation, as drafted, is interpreted to be, under the broadest reasonable interpretation, a “mental process”. Therefore, the claims are rejected on the same basis as claims 1 and 11 respectively.
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.
Claim(s) 1, 3-7, 9, 11, 13-17, and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Rawal and Lakkaraju. (Herein referred to as Rawal) (Beyond Individualized Recourse: Interpretable and Interactive Summaries of Actionable Recourses) in view of Pedro Tabacof et al. (Herein referred to as Tabacof) (PROBABILITY CALIBRATION FOR KNOWLEDGE GRAPH EMBEDDING MODELS)
Regarding claim 1, Rawal teaches a method for providing a global counterfactual explanation that summarizes a machine learning model behavior, the method comprising: generating an initial ground set based on a first candidate set of outer-If conditions (SD), and a second candidate set used for selecting Inner-If or Then conditions (RL); (See Figure 1 on pg. 2. In the example given in Figure 1, the Subgroup descriptor corresponds to a first candidate set, with the recourse rules corresponding to second candidate sets.) evaluating a fixed number of triples and forming a new ground set that provides a recourse accuracy level above a reference threshold, wherein the fixed number of triples included in the new ground set is less than a total number of triples included in the initial ground set, (“we use the recourse rules outlined by our explanation to prescribe recourses to affected individuals, they should be able to obtain the desired predictions from the black box model upon acting on the recourse. To quantify this notion, we define incorrectrecourse(R) which is defined as the number of instances in Xaff for which acting upon the recourse prescribed by R does not result in the desired prediction (Table 1). Our goal would therefore be to construct explanations that minimize this metric… we evaluate how the recourse accuracy metric (defined above) changes as we vary the size of the two-level recourse set i.e., number of triples of the form (q, c, c0 ) in the two-level recourse set (Section 2.2).”, pg. 4, second paragraph; pg. 8, third paragraph; See also Figure 2 on pg. 8.) (The number of instances in Xaff acting upon the recourse prescribed by R corresponds to recourse accuracy levels. This metric is shown in Figure 2. Figure 2 also shows evaluating the recourse accuracy at different sizes. (Size corresponding to a number of triples.) The previous accuracy at every iteration of size corresponds to an accuracy threshold, as the goal of the algorithm is to find the lowest value for size without compromising accuracy. The set with most optimal values between size and recourse accuracy would then be the new ground set. The varying of the size i.e., the number of triples correspond to a number of triples less than the number of triples in the initial ground set. (For instance, the graph at 50 triples which correspond to a total number of triples, vs 30 triples corresponding to a fixed number of triples) sorting the new ground set by recourse accuracy; producing a two-level recourse set by selecting a predetermined number of triples based on corresponding recourse accuracies indicated in the sorting; (“Formally, a two-level recourse set is a set of triples… we evaluate how the recourse accuracy metric (defined above) changes as we vary the size of the two-level recourse set i.e., number of triples of the form (q, c, c0 ) in the two-level recourse set (Section 2.2). Results for the same are shown in Figure 2. It can be seen that recourse accuracies converge to their maximum values at explanation sizes of about 10 to 15 rules across all the datasets.”, pg. 3, bottom paragraph; pg. 8, third paragraph) (They choose 10 to 15 rules to have a reasonable mix of interpretability and accuracy, with the sorting being indicated by the higher accuracies at higher sizes.) performing calculation based on the selected number of triples; and constructing the global counterfactual explanation based on at least the performing of the calculation. (“If a conjunction q ∧c consists of the entire feature-vector of a particular data-point x ∈ Xaff, then the triple (q,c,c) represents a single instance level counterfactual. This is how AReS can be used to output individual recourses.”, pg. 15, second paragraph; See also Algorithm 1 on pg. 14) (The calculation is made. Based on the calculation meeting a condition according to the algorithm, the triple can represent a counterfactual.)
However Rawal does not teach wherein each triple included in the new ground set increases below-threshold recourse accuracy levels of the new ground set;
Tabacof teaches each triple included in the new ground set increases below-threshold recourse accuracy levels of the new ground set. (“Since the learned models are not calibrated, multiple decision thresholds τi must be picked, where 0 < i < |R|, i.e. one for each relation type… Classification metrics apply (e.g. accuracy)… Table 4: Effect of calibration on triple classification accuracy. Best results in bold. For all calibration methods there is one single threshold, τ=0.5. For the per-relation τ, we learned multiple thresholds from validation sets.”, pg. 3, under “Triple Classification.”; pg. 8, under “Table 4”) (The triples of Rawal combine with the thresholding of Tabacof, which increases the accuracy of the triples’ classification. This corresponds to increasing below-threshold recourse accuracy levels of the new ground set.)
Therefore it would have been considered obvious to one of ordinary skill in the art, prior to the filing date of the current application, to combine the method of Rawal with the accuracy thresholds of Tabacof. One would have been motivated to combine the two teaching, prior to the filing date of the current application, as the calibration leads to higher triple classification accuracy, as disclosed in Tabacof. (See Table 4 on pg. 8)
Regarding claim 11, Rawal teaches A system for providing a global counterfactual explanation that summarizes a machine learning model behavior, the system comprising: at least one processor; at least one memory; and at least one communication circuit (While the method of Rawal never explicitly discloses a processor, memory, or a communication circuit, one would implicitly need those components to perform the method of Rawal.) wherein the at least one processor performs operations that comprise: generating an initial ground set based on a first candidate set of outer-If conditions (SD), and a second candidate set used for selecting Inner-If or Then conditions (RL); (See Figure 1 on pg. 2. In the example given in Figure 1, the Subgroup descriptor corresponds to a first candidate set, with the recourse rules corresponding to second candidate sets.) evaluating a fixed number of triples and forming a new ground set that provides a recourse accuracy level above a reference threshold, wherein the fixed number of triples included in the new ground set is less than a total number of triples included in the initial ground set, (“we use the recourse rules outlined by our explanation to prescribe recourses to affected individuals, they should be able to obtain the desired predictions from the black box model upon acting on the recourse. To quantify this notion, we define incorrectrecourse(R) which is defined as the number of instances in Xaff for which acting upon the recourse prescribed by R does not result in the desired prediction (Table 1). Our goal would therefore be to construct explanations that minimize this metric… we evaluate how the recourse accuracy metric (defined above) changes as we vary the size of the two-level recourse set i.e., number of triples of the form (q, c, c0 ) in the two-level recourse set (Section 2.2).”, pg. 4, second paragraph; pg. 8, third paragraph; See also Figure 2 on pg. 8.) (The number of instances in Xaff acting upon the recourse prescribed by R corresponds to recourse accuracy levels. This metric is shown in Figure 2. Figure 2 also shows evaluating the recourse accuracy at different sizes. (Size corresponding to a number of triples.) The previous accuracy at every iteration of size corresponds to an accuracy threshold, as the goal of the algorithm is to find the lowest value for size without compromising accuracy. The set with most optimal values between size and recourse accuracy would then be the new ground set. The varying of the size i.e., the number of triples correspond to a number of triples less than the number of triples in the initial ground set. (For instance, the graph at 50 triples which correspond to a total number of triples, vs 30 triples corresponding to a fixed number of triples) sorting the new ground set by recourse accuracy; producing a two-level recourse set by selecting a predetermined number of triples based on corresponding recourse accuracies indicated in the sorting; (“Formally, a two-level recourse set is a set of triples… we evaluate how the recourse accuracy metric (defined above) changes as we vary the size of the two-level recourse set i.e., number of triples of the form (q, c, c0 ) in the two-level recourse set (Section 2.2). Results for the same are shown in Figure 2. It can be seen that recourse accuracies converge to their maximum values at explanation sizes of about 10 to 15 rules across all the datasets.”, pg. 3, bottom paragraph; pg. 8, third paragraph) (They choose 10 to 15 rules to have a reasonable mix of interpretability and accuracy, with the sorting being indicated by the higher accuracies at higher sizes.) performing calculation based on the selected number of triples; and constructing the global counterfactual explanation based on at least the performing of the calculation. (“If a conjunction q ∧c consists of the entire feature-vector of a particular data-point x ∈ Xaff, then the triple (q,c,c) represents a single instance level counterfactual. This is how AReS can be used to output individual recourses.”, pg. 15, second paragraph; See also Algorithm 1 on pg. 14) (The calculation is made. Based on the calculation meeting a condition according to the algorithm, the triple can represent a counterfactual.)
However Rawal does not teach wherein each triple included in the new ground set increases below-threshold recourse accuracy levels of the new ground set;
Tabacof teaches each triple included in the new ground set increases below-threshold recourse accuracy levels of the new ground set. (“Since the learned models are not calibrated, multiple decision thresholds τi must be picked, where 0 < i < |R|, i.e. one for each relation type… Classification metrics apply (e.g. accuracy)… Table 4: Effect of calibration on triple classification accuracy. Best results in bold. For all calibration methods there is one single threshold, τ=0.5. For the per-relation τ, we learned multiple thresholds from validation sets.”, pg. 3, under “Triple Classification.”; pg. 8, under “Table 4”) (The triples of Rawal combine with the thresholding of Tabacof, which increases the accuracy of the triples’ classification. This corresponds to increasing below-threshold recourse accuracy levels of the new ground set.)
Therefore it would have been considered obvious to one of ordinary skill in the art, prior to the filing date of the current application, to combine the method of Rawal with the accuracy thresholds of Tabacof. One would have been motivated to combine the two teaching, prior to the filing date of the current application, as the calibration leads to higher triple classification accuracy, as disclosed in Tabacof. (See Table 4 on pg. 8)
Regarding claims 3 and 13, Rawal, as modified by Tabacof, teaches the method and system according to claims 1 and 11 respectively, as well as the generating of the initial ground set is performed by filtering a dataset based on the outer-If or the inner-If conditions, and separately deploying a method for generating Then conditions. (Figure 1 on pg. 2 shows Subgroup descriptors and recourse rule containing if and then statements which acts as a filter for dataset.)
Regarding claims 4 and 14, Rawal, as modified by Tabacof, teaches the method and system according to claims 1 and 11 respectively, as well as each triple includes an outer-If condition, an inner-If condition, and a Then condition. (“a two-level recourse set is a set of triples and has the following form: R = {(q1, c11, c0 11),(q1, c12, c0 12)· · ·(q2, c21, c0 21)· · · } where qi corresponds to the subgroup descriptor, and (cij , c0 ij ) together represent the inner if-then recourse rules with cij denoting the antecedent (i.e., the if condition) and c 0 ij denoting the consequent (i.e., the recourse).”, pg. 3, bottom paragraph; See also Figure 1 on pg. 2)
Regarding claims 5 and 15, Rawal, as modified by Tabacof, teaches the method and system according to claims 1 and 11 respectively, as well as the selecting of the predetermined number of triples includes selecting highest-performing triples within the new ground set. (“we evaluate how the recourse accuracy metric (defined above) changes as we vary the size of the two-level recourse set i.e., number of triples of the form (q, c, c0 ) in the two-level recourse set (Section 2.2). Results for the same are shown in Figure 2. It can be seen that recourse accuracies converge to their maximum values at explanation sizes of about 10 to 15 rules across all the datasets.” Pg. 8, third paragraph) (The triples with the maximum values are selected.)
Regarding claims 6 and 16, Rawal, as modified by Tabacof, teaches the method and system according to claims 1 and 11 respectively, as well as each triple forming the new ground set increases the recourse accuracy level. (“AReS always had the lowest FCost, and provided recourses that either had the highest Recourse Accuracy or were within 0.5% of the best performing individual recourse generation method.”, pg. 8, second paragraph)
Regarding claims 7 and 17, Rawal, as modified by Tabacof, teaches the method and system according to claims 1 and 11 respectively, as well as one or more constraints are applied during the generating of the initial ground set. (See Figure 1 on pg.2. The variables within the if statements correspond to constraints that are applied during the generation of ground sets.)
Regarding claims 9 and 19, Rawal, as modified by Tabacof, teaches the method and system according to claims 1 and 11 respectively, as well as an upper bound defined as acc(R) < acc(V) is reached before an algorithm for providing the global counterfactual explanation has completed execution. (“To construct non-negative reward functions, penalty terms (metrics in Table 1) are subtracted from their corresponding upper bound values (U1, U3, U4) which are computed with respect to the sets SD and RL.”, pg. 5, second paragraph of “2.3 Learning Two Level Recourse Sets”) (The upper bounds can be easily configured to be reached before the algorithm has completed execution.) wherein acc(R) is a percentage of instances in Xaff that are provided with a successful recourse, wherein Xaff is a set of individuals with an unfavorable prediction from a model, and wherein acc(v) is a recourse accuracy. (“we define incorrectrecourse(R) which is defined as the number of instances in Xaff for which acting upon the recourse prescribed by R does not result in the desired prediction (Table 1)… we outline the following metrics: 1) Recourse Accuracy: percentage of instances in Xaff for which acting upon the prescribed recourse (e.g., changing the feature values as prescribed by the recourse) obtains the desired prediction”, pg. 4, second paragraph; pg. 7, bottom paragraph)
Claims 2 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Rawal in view of Tabacof, and in further view of Manivannan et al. (Herein referred to as Manivannan) (U.S. Patent Application Publication No. US 20230008115 A1)
Regarding claims 2 and 12, Rawal, as modified by Tabacof teaches the method and system according to claims 1 and 11 respectively, but does not explicitly teach storing up to (1-a2)n3 iterations from among the initial ground set, wherein the generating of the initial ground set is performed by iterating over the second candidate set in O(n) time and computing feature combinations, before removing any items that contain a feature combination that only occurs once, for yielding a new RL with size an, and wherein a is greater than or equal to 0 and less than or equal 1.
Rawashdeh teaches storing up to (1-a2)n3 iterations from among the initial ground set (“Once validated, the model can be moved to a model repository, which can track deployed global models and store past iterations/version of models.”, Paragraph 33) (Under the broadest reasonable interpretation, the claim requires the storing of any kind of an iteration. Rawashdeh teaches storing iteration of a machine learning model, corresponding to the storing of an iteration, which in combination with Rawal and Tabacof, fully teaches the limitation.)
Therefore it would have been considered obvious to one of ordinary skill in the art, prior to the filing date of the current application, to combine the method of Rawal, as modified by Tabacof, with the storing of iterations, as detailed in Rawashdeh. On would have been motivated to combine the teachings, prior to the application’s filing date, as this allows for proper storage for tracking and later distribution, as outlined in Rawashdeh. (“Once validated, the model can be moved to a model repository, which can track deployed global models and store past iterations/version of models. In federated learning, globally trained models may be validated in the cloud prior to distribution or may be distributed to a limited subset of vehicles for testing/validation prior to wide distribution.”, Paragraph 33)
However, the combination still does not teach the generating of the initial ground set is performed by iterating over the second candidate set in O(n) time and computing feature combinations, before removing any items that contain a feature combination that only occurs once, for yielding a new RL with size αn, and wherein α is greater than or equal to 0 and less than or equal 1.
Manivannan teaches the generating of the initial ground set is performed by iterating over the second candidate set in O(n) time and computing feature combinations, before removing any items that contain a feature combination that only occurs once, for yielding a new RL with size αn, and wherein α is greater than or equal to 0 and less than or equal 1. (“The process of cleaning the data may include fixing or removing incorrect, corrupted, incorrectly formatted, duplicate, or incomplete data within a dataset.”, Paragraph 58) (The removal of duplicate data from a dataset would make it to where the size of the dataset goes from n to αn where α is between 0 and 1.)
Therefore it would have been considered obvious to one of ordinary skill in the art, prior to the filing date of the current application, to combine the method of Rawal, as modified by Tabacof and Rawashdeh, with the data removal of Manivannan. One would have been motivated to combine the two teaching, prior to the filing date of the current application, as data removal allows for the cleaned data to be maintainable, reproducible, and traceable, as disclosed in Manivannan. (“This “clean” form of data ensures that the feature engineering logic is maintainable, the target datasets are reproducible and, sometimes, that the whole pipeline is traceable to the source representations.”, Paragraph 58)
Claim(s) 8 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Rawal in view of Tabacof, and in further view of BANERJEE et al. (Herein referred to as Banerjee) (U.S. Patent Application Publication No. US 20210125207 A1)
Regarding claims 8 and 18, Rawal, as modified by Tabacof teaches the method and system according to claims 1 and 11 respectively, but does not explicitly teach the initial ground set removes a feature combination that only occurs once.
Banerjee teaches removing a feature combination that only occurs once. (“data cleaning can be implemented to improve the quality of the data. This includes imputation of missing data and the removal of erroneous values and outliers.”, Paragraph 34) (Removal of erroneous values and outliers corresponds to data that occurs once. Combined with and configured to the removal of the feature combinations of Rawal, as modified by Tabacof, teaches the limitation.)
Therefore it would have been considered obvious to one of ordinary skill in the art, prior to the filing date of the current application, to combine the method of Rawal, as modified by Tabacof, with the data removal of Banerjee. One would have been motivated to combine the two teaching, prior to the filing date of the current application, as this improved the quality of the data, as disclosed in Banerjee. (“data cleaning can be implemented to improve the quality of the data. This includes imputation of missing data and the removal of erroneous values and outliers.”, Paragraph 34)
Claim(s) 10 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Rawal in view of Tabacof, and in further view of Tesauro et al. (Herein referred to as Tesauro) (U.S. Patent Application Publication No. US 20070203871 A1)
Regarding claims 10 and 20, Rawal, as modified by Tabacof, teaches the method and system according to claims 9 and 19 respectively, but does not explicitly teach the algorithm is terminated prior to its completion when the upper bound for saturation is reached.
Tesauro teaches an algorithm is terminated prior to its completion when the upper bound for saturation is reached. (“training is stopped if an upper bound on the number of training iterations has been reached.” Paragraph 27) (The stopping criteria of Das is related to a training algorithm, but can be applied in a similar way to the algorithm with upper bounds of Rawal, fully teaching the limitation.)
Therefore it would have been considered obvious to one of ordinary skill in the art, prior to the filing date of the current application, to combine the method of Rawal, as modified by Tabacof, with the stopping criteria of Tesauro. One would have been motivated to combine the two teaching, prior to the filing date of the current application, as this saves computational resources via dynamic allocation, as disclosed in Tesauro. (“This entails the development of effective policies pertaining to, for example, dynamic allocation of computational resources, performance tuning of system control parameters, dynamic configuration management, automatic repair or remediation of system faults and actions to mitigate or avoid observed or predicted malicious attacks or cascading system failures.” Paragraph 2)
Response to Arguments
Applicant's arguments filed on April 6th, 2026, have been fully considered but they are not persuasive. The applicant argues in substance:
Argument 1: The outstanding subject matter eligibility is improper under 2106.07 because the rejection fails to explain why the Examiner concluded that the cited claim language is directed to an abstract idea and did not contain additional elements.
The examiner respectfully disagrees. Under MPEP 2106 Subject Matter Eligibility Test for Products and Processes, the steps according to the flowchart were followed properly. Claim 11 is directed towards a machine, recites an abstract idea, and comprises additional elements, such as a processor, memory, and communication circuit that do not amount to significantly more, as they are interpreted to be tools to perform the abstract idea. Claim 1 also recites the use of a processor and memory. While it was improper in the last rejection to state that those elements do not amount to additional elements, the claim would be examined similarly to claim 11, in which the additional elements (the processor and memory) do not amount to significantly more than the judicial exception, according to the broadest reasonable interpretation of the claim as a whole. As such, the claims were rejected under 35 U.S.C. 101.
Argument 2: The newly added limitations incorporates improvements that integrate the supposed abstract ideas into a practical application. Specifically, the added limitations recite an improvement in ascertaining global properties of machine learning models.
The examiner respectfully disagrees. The newly added limitations do not successfully incorporate the abstract idea into a practical application. The new limitations, such as “a total number of triples”, “each triple included in the new ground set increase below-threshold recourse accuracy levels of the new ground set”, “producing a two-level recourse set”, and “constructing the global counterfactual explanation based on at least the performing of the calculation”, are merely parts of the abstract ideas themselves, with the limitation, “summarizes a machine learning model behavior” being interpreted to be intended use which is given negligible patentable weight. Furthermore, the additional elements in the claims do not reflect an improvement in ascertaining global properties of machine learning models, but at most recites improvements to an abstract idea. Therefore, the 101 rejections are maintained.
Argument 3: Rawal fails to teach “evaluating a fixed number of triples… wherein the fixed number of triples… is less than a number of triples included in the initial ground set” and “producing a two-level recourse set.”
The applicant respectfully disagrees. As explained in the previous action and in this one, the number of triples is evaluated at every size, which includes the entirety of the ground set and a number less than a total number of triples included in the initial ground set, teaching the limitation. As for “producing a two-level recourse set”, Rawal define a two-level recourse set as a set of triples, which Rawal describes throughout the disclosure on how they produce a set of triples. Evidence is also shows how the produced triples can be used to generate counterfactuals as shown in the Appendix and Algorithm 1 on pgs. 14 and 15. Therefore, the rejection is proper.
Argument 4: Manivannan does not teach “removing items that contain a feature combination that only occurs once.”
The examiner respectfully disagrees. Manivannan teaches the removal of specifically “incorrect, corrupted, incorrectly formatted, duplicate, or incomplete data”. Under the broadest reasonable interpretation, “incorrect data” corresponds to data that only occurs once. The explanation provided was to explain how removing this data would make the size of the dataset go from n to an, where a is between 0 and 1, explaining away the additional limitation. As such, the rejection is proper.
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 Tyler E Iles whose telephone number is (571)272-5442. The examiner can normally be reached 9:00am - 5:00pm.
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, Kakali Chaki can be reached at (571) 272-3719. 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.
/T.E.I./ Patent Examiner, Art Unit 2122
/KAKALI CHAKI/ Supervisory Patent Examiner, Art Unit 2122
Prosecution Insights