DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This Office Action is responsive to Applicant's amendment filed on 25 March 2026. Applicant’s amendment on 25 March 2026 amended Claims 1, 8, and 15. Currently Claims 1, 3, 4, 7, 8, 10, 11, 14, 15, 17, 18, and 20-23 are pending and have been examined. Claim 21-23 is newly presented. Claims 2, 5, 6, 9, 12, 13, 16, and 19 were previously canceled. The Examiner notes that the 103 rejections have been withdrawn, however the 101 rejection has been maintained.
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 25 March 2026 has been entered.
Response to Arguments
The Applicant argues on pages 13-14 that “Srivastava in view of Opdycke in further view of Dirac does no teach or suggest training a machine learning model to predict duplication factors for a media campaign based on total exposure metrics associated with respective media platforms for the media campaign as in claim 1”.
The Examiner respectfully disagrees.
In response to the arguments the Examiner notes that as provided in the previous Office action in par. [0027] of Opdycke it is taught that the training a machine learning model to detect duplication of media content, this de-duplicate implies knowing what is a duplicate and what can then be de-duplicated. Furthermore, for further clarification in par. [0063] it is disclosed that understanding the long-term impact on campaigns and utilizing production modeling by gathering and enough data to understand behavior based on exposure data (metrics)). Opdycke teaching the scheduling of marketing campaigns based on measurements which is then used to optimize audience responses. This includes as previously provided the par. [0027] that discloses a machine learning model where the audience will be exposed to stimuli and this exposure is analyzed based on the weight and decay of the influence.
It is viewed that it is inherent that if a de-duplication can occur the analysis and understanding and with the addition of understanding the exposure it is viewed that the machine learning model takes this information and then predict duplication factors for a media campaign based on total exposure metrics associated with respective media platforms for the media campaign. Therefore, the rejection is maintained.
The Applicant argues on page 11 that “The invention described in amended claim I provides a technical solution to the problem of determining deduplicated audience metrics when cross-platform duplication data is unavailable. The application explains that for some media campaigns, advertisement exposure for individual media platforms (e.g., TV, online, mobile) is known, but duplication across platform combinations is unknown. See As-Fi/ed Specification, paragraph [0011]. The application further explains that duplication factors from a reference media campaign cannot be used directly for a query campaign because doing so would cause resulting audience metrics to be produced with illogical trends, inconsistent volumetrics, and/or an audience which is not reflective of the actual query media campaign's performance. See As-Filed Specification, paragraph [0053]”.
The Examiner respectfully disagrees:
With respect to the argument the Examiner notes that the Applicant's argument that the claims are directed to a "technical solution to a technical problem" has been fully considered but is unpersuasive. While the Examiner acknowledges that the specification identifies a real-world challenge specifically, that cross-platform audience duplication data is unknown for query media campaigns and that directly applying reference campaign duplication factors yields illogical or inconsistent audience metrics the characterization of this challenge as a "technical problem" does not transform the claimed solution into patent-eligible subject matter. The problem identified is fundamentally a data availability and data quality problem, not a problem rooted in the technical operation of a computer or computing system. The absence of cross-platform duplication data for a query campaign is a measurement or information gap within the field of audience analytics an issue that would exist regardless of whether a computer were used to address it. Similarly, the challenge of inconsistent or illogical volumetrics arising from directly transplanting duplication factors from one campaign to another is a mathematical consistency problem: the numbers do not add up correctly. These are not problems caused by deficiencies in how a computer stores data, processes instructions, or manages memory, and therefore they cannot serve as the basis for a technical solution argument under the Enfish line of cases. Enfish, LLC v. Microsoft Corp. (an improvement to computer functionality must be directed to how the computer operates e.g., how it organizes memory not merely to the quality of data the computer produces). The claimed solution training a machine learning model, performing a nearest-neighbor campaign selection via KD tree and Euclidean distance, and applying a cross-entropy minimization with linear consistency and logical constraints is a mathematical solution to a mathematical and statistical problem. Each component of the claimed process is a well-known mathematical technique applied to media campaign data. The fact that these mathematical techniques are applied within the specific domain of audience measurement does not elevate the claims above the abstract idea; applying an abstract idea in a particular technological environment, without more, does not constitute a practical application. Alice Corp. Pty. Ltd. v. CLS Bank Int'l, (2014); BSG Tech LLC v. BuySeasons, Inc. Furthermore, Applicant's citation to the specification paragraphs describing the nature of the problem does not overcome the rejection, because the characterization of an invention in the specification does not control the eligibility analysis it is the claims themselves that must be examined. Interval Licensing LLC v. AOL, Inc. For these reasons, the rejection is maintained.
The Applicant argues pages 11-12 that “the application also explains a technical solution to the problem. The max entropy engine utilizes numeric optimization techniques to ensure that any estimates that are produced satisfy a set of requirements. See As-Filed Specification, paragraph [0054]. The metric used is the cross entropy between the media platforms of the reference media campaign and the query media campaign, and the selected solution is one which minimizes the cross-entropy. See As-Filed Specification, paragraph [0060]. The specification confirms that these processes "increase the efficiency of the data engine" and "increase the computational efficiency of the data engine by removing illogical data that would require additional processing cycles to analyze." As-Filed
Specification, paragraph [0065].
Hence, the disclosed systems and methods address a technical problem in media audience measurement technology with a specific, improved solution. Based on the above-referenced aspects of the specification, one of ordinary skill in the art would recognize the claimed invention as providing an improvement”.
The Examiner respectfully disagrees.
Applicant's argument that the max entropy engine's cross-entropy minimization constitutes a technical solution that improves computational efficiency has been fully considered but remains unpersuasive, and the rejection is hereby maintained. Applicant relies primarily on the specification's characterization at paragraph [0065] that the claimed processes "increase the computational efficiency of the data engine by removing illogical data that would require additional processing cycles to analyze." However, the Examiner notes as an initial matter that Applicant has cited paragraph [0065] for this proposition, whereas the specification language describing computational efficiency improvements to the data engine appears at paragraph [0070], not paragraph [0065]. Paragraph [0065] describes the consistent and logical requirements that permit infinitely many solutions and introduces cross-entropy as the metric used to select among those solutions. This distinction is significant because Applicant's argument conflates the mathematical mechanism of the invention cross-entropy minimization subject to linear consistency and logical constraints with the downstream operational benefit of having consistent output data, and attempts to reframe the latter as a technical improvement to computer functionality. These are not the same thing, and the Federal Circuit has drawn this distinction clearly.
The specification's assertion that the claimed processes improve the operating efficiency of the data engine does not, without more, establish patent eligibility. As the Federal Circuit stated in Interval Licensing LLC v. AOL, Inc., "we look to whether the claims focus on a specific means or method that improves the relevant technology, or whether they are directed to a result or effect that itself is the abstract idea and merely invoke generic processes and machinery." Here, the claims focus squarely on the mathematical result final duplication factors that are internally consistent and close to the reference campaign's duplication factors and the recited improvement to computational efficiency is simply a consequence of having mathematically consistent output rather than a specific technological mechanism by which the computer itself operates differently. The "data engine" referenced in the specification is a downstream processing component that benefits from receiving cleaner input data; the claim does not specify any structural change to how the computing system processes instructions, manages memory, or executes operations. Receiving logically consistent data rather than inconsistent data as input is not a technical improvement to the computer it is simply better data. Any processing system, computing or otherwise, benefits from receiving accurate inputs, yet this truism cannot serve as the foundation for patent eligibility under 101.
Furthermore, the specific mathematical technique at issue cross-entropy minimization subject to linear consistency constraints and logical bounding constraints, solved via a numeric dual optimization solver is a well-established mathematical optimization methodology. The Examiner further notes that the specification itself at paragraph [0067] confirms that the max entropy engine "utilizes numeric solvers that optimize the dual and minimize the duality gap," which is a standard mathematical technique in convex optimization, not a novel computational architecture or unconventional computer operation. Under Alice Corp. Pty. Ltd. v. CLS Bank Int'l, implementing an abstract idea using well-understood, routine, and conventional computer functions does not amount to significantly more than the abstract idea itself. The use of a numeric solver to perform mathematical optimization, however computationally sophisticated, remains a mathematical operation performed on a generic computing platform.
Applicant's reliance on the proposition that "one of ordinary skill in the art would recognize the claimed invention as providing an improvement" also does not advance the eligibility analysis. Patent eligibility under 101 is a question of law, and the perception of practitioners in the field does not substitute for the legal inquiry into whether the claims are directed to an abstract idea and whether they recite significantly more. Berkheimer v. HP Inc., (while the question of whether claim elements are well-understood, routine, and conventional may involve underlying factual questions, the 101 inquiry itself remains one of law). The improvement the Applicant identifies is an improvement in the quality and logical consistency of audience measurement data a valuable commercial and analytical benefit, but one that is achieved entirely through the application of mathematical optimization techniques on a generic computing system, which does not satisfy the requirements for patent eligibility. For all of the foregoing reasons, the rejection is maintained.
The Applicant argues on page 12 that the “Amended claim 1 itself reflects this disclosed improvement. It outlines a detailed, specific process. That process involves training a machine learning model to predict duplication factors, predicting estimated duplication factors for a query media campaign, selecting a reference media campaign from the plurality of media campaigns based on a comparison of the estimated duplication factors, and determining final duplication factors by a max entropy engine utilizing a numeric solver by minimizing a cross-entropy between the reference duplication factors and the final duplication factors. The claim explicitly recites that "the determining improves a computational efficiency of a data engine by removing illogical data prior to determination of a total audience size. " The claim further recites that the determining is "constrained by consistency'
PNG
media_image1.png
9
2
media_image1.png
Greyscale
between the total exposure metrics and unique audiences derived from the final duplication factors. This is a specific, sequential, and technologically-bound process that reflects the improvement described in the specification”.
The Examiner respectfully disagrees.
Applicant's argument that amended claim 1 recites a "specific, sequential, and technologically-bound process" that reflects the improvement described in the specification has been fully considered but is unpersuasive, and the rejection is hereby maintained. Applicant essentially contends that the level of specificity and the sequential ordering of the claimed steps, combined with the explicit recitation that "the determining improves a computational efficiency of a data engine by removing illogical data prior to determination of a total audience size," is sufficient to establish patent eligibility. The Examiner disagrees for the reasons set forth below, and notes that this argument substantially overlaps with Arguments 1 and 2 previously addressed, which are incorporated herein by reference.
As an initial matter, the specificity of a claim's recitation of an abstract idea does not transform that abstract idea into patent-eligible subject matter. The Supreme Court in Alice Corp. Pty. Ltd. v. CLS Bank Int'l, expressly cautioned against allowing the "draftsman's art" to circumvent the prohibition against patenting abstract ideas. A detailed or particularized recitation of steps that, at their core, implement mathematical concepts does not move those steps outside the realm of the abstract. Synopsys, Inc. v. Mentor Graphics Corp., ("the level of abstraction at which we describe a claimed concept is irrelevant to whether the claim is directed to an abstract idea"). Here, each of the four sequential steps recited in claim 1 training a machine learning model, predicting estimated duplication factors, selecting a reference media campaign via comparison, and determining final duplication factors via cross-entropy minimization is, as established in the rejection and in the Examiner's responses to Arguments 1 and 2, a mathematical concept or a combination of mathematical concepts. The fact that these mathematical steps are arranged in a particular sequential order and described with specificity does not convert them into a technical improvement to computer functionality. Elec. Power Grp., LLC v. Alstom S.A., ("the claims are not directed to a specific improvement to the way computers or networks perform their basic functions but instead are directed to certain independently abstract ideas that use computers and networks as tools").
Regarding Applicant's emphasis on the claim language that "the determining improves a computational efficiency of a data engine by removing illogical data prior to determination of a total audience size," the Examiner has already addressed this language in the response to above, but reiterates the following critical point: this limitation recites a functional result or intended outcome of performing the abstract mathematical process, not a structural or operational mechanism by which the claimed computing system itself functions differently from a conventional computer. The claim does not specify how the data engine's computational efficiency is improved at a technical level it does not recite a specific memory management technique, a particular processing architecture, a modified data structure, or any unconventional hardware or software configuration that achieves the stated efficiency improvement. Rather, the claim simply asserts that performing the cross-entropy minimization produces logically consistent outputs, and that those outputs are easier for a downstream data engine to process. This is precisely the type of result-oriented claiming that the Federal Circuit has found insufficient to establish eligibility. Two-Way Media Ltd. v. Comcast Cable Commc'ns, LLC, (claims that recite functional results without specifying how those results are achieved at a technical level do not satisfy 101). Critically, the claim language itself does not meaningfully limit the manner in which the computational efficiency improvement is achieved beyond the mathematical steps already recited the improvement is simply described as occurring as a consequence of those steps, not as a separate and distinct technical mechanism.
With respect to the "constrained by consistency" limitation, the Examiner likewise reiterates that the consistency constraints requiring that the final duplication factors add up to the total exposure metrics and that unique audiences for different platform combinations are self-consistent are mathematical constraints expressed as a linear system and inequality bounds, as confirmed by the specification at paragraphs [0059] through [0064]. Characterizing these mathematical relationships as a technological constraint does not alter their fundamental nature as mathematical requirements. Digitech Image Techs., LLC v. Elecs. for Imaging, Inc., (a process that employs mathematical algorithms to manipulate existing information to generate additional information is not patent-eligible). The use of mathematical consistency constraints in an optimization problem is a standard feature of constrained optimization a well-established mathematical discipline and does not constitute a technological improvement to a computer system.
Finally, the Examiner addresses Applicant's characterization of the claim as "technologically-bound." A claim is not rendered technologically bound merely because it operates within a technological domain such as media audience measurement, or because it is implemented on a computing system. The computing system recited in claim 1 is a generic processor and memory platform, and as noted in the rejection, the specification confirms at paragraph [0085] that this platform encompasses conventional computing devices such as servers and personal computers. Tethering abstract mathematical operations to a generic computing environment does not transform those operations into a patent-eligible technical improvement. Alice, 573. For all of the foregoing reasons, the claims remain directed to patent-ineligible subject matter, and the rejection is hereby maintained.
The Applicant argues on page 13 that “Additionally, amended claim I is analogous to the claims found eligible in Ex parie Kamaih, Appeal 2018-000030 (P TAB 2019). In Kamaih, the Board found that claims directed to generating, optimizing, and using a predictive model to determine marketing options were patent eligible because the claims "implement the recited abstract idea with the particular computing device that is integral to the claim and apply or use the recited abstract idea in a meaningful way beyond generally linking the use of the abstract idea to a particular technological environment, such that the claim as a whole is more than a drafting effort designed to monopolize the exception. Kamalh at 6. Similarly, amended claim I implements the recited abstract idea with a specific computing system comprising a max entropy engine that utilizes a numeric solver to minimize cross-entropy subject to consistency constraints a particular computing device that is integral to the claim.
The Examiner respectfully disagrees.
Applicant's reliance on Ex parte Kamaih, Appeal No. 2018-000030 (PTAB 2019), as analogous authority supporting the eligibility of amended claim 1 has been fully considered but is unpersuasive, and the rejection is hereby maintained for the reasons set forth below.
As a threshold matter, the Examiner notes that decisions by the Patent Trial and Appeal Board in ex parte appeals are not precedential authority binding on examiners unless designated as such by the USPTO Director. Ex parte PTAB decisions are, at most, persuasive and must be evaluated in light of the specific claim language, specification, and record before the Board in that particular case. Each application must be evaluated on its own merits based on the specific claims presented and the applicable legal framework. In re Comiskey. Accordingly, the Examiner is not bound by Kamaih and must independently evaluate the claims of the present application under the governing legal standards established by the Supreme Court and the Federal Circuit, as well as the 2019 Revised Patent Subject Matter Eligibility Guidance.
Turning to the substance of Applicant's analogy, even assuming arguendo that Kamaih was correctly decided and is persuasive authority, the present claims are factually and legally distinguishable in a critical respect. Applicant quotes Kamaih for the proposition that the claims in that case were eligible because they "implement the recited abstract idea with the particular computing device that is integral to the claim" in a manner that applies or uses the abstract idea in a meaningful way beyond generally linking it to a particular technological environment. Applicant then argues that amended claim 1 similarly implements the abstract idea with a "specific computing system comprising a max entropy engine that utilizes a numeric solver to minimize cross-entropy subject to consistency constraints a particular computing device that is integral to the claim." However, this characterization fundamentally misidentifies what constitutes a "particular computing device" within the meaning of the eligibility analysis.
The "max entropy engine utilizing a numeric solver" is not a particular computing device it is a software component implementing a mathematical algorithm. The specification confirms this at paragraphs [0067] and [0070], where the max entropy engine is described as a functional module that "utilizes numeric solvers that optimize the dual and minimize the duality gap" a mathematical optimization routine executed on a general-purpose processor. The specification further confirms at paragraph [0080] that the max entropy engine, along with all other components of the machine learning engine, "may be implemented by hardware, software, firmware and/or any combination of hardware, software and/or firmware," and may be executed on conventional programmable processors, GPUs, DSPs, and ASICs. A software module executing a mathematical algorithm on a generic processor does not constitute a "particular computing device" in any meaningful technical sense. To accept Applicant's characterization would be to hold that any software-defined functional component implementing an algorithm qualifies as a "particular computing device" integral to a claim a result that would effectively nullify the 101 inquiry for virtually all software-implemented inventions, which the Supreme Court expressly rejected in Alice Corp. Pty. Ltd. v. CLS Bank Int'l.
The distinction between the present claims and the type of claims found eligible under the Enfish/McRO line of cases which appears to underlie the Kamaih Board's reasoning is equally significant. In cases where courts have found that a particular computing implementation renders an otherwise abstract idea patent-eligible, the computing system at issue was configured in a specific, unconventional manner that itself constituted the improvement. See Enfish, LLC v. Microsoft Corp.) (self-referential table representing a specific improvement to the way a computer organizes and retrieves data); McRO, Inc. v. Bandai Namco Games Am. Inc., (specific rules-based morph weight and phoneme sequence process that automated a previously human-performed task in a technically specific manner). In the present case, there is no analogous specific, unconventional computing configuration. The claimed computing system a generic processor and memory running a machine learning model, a KD tree nearest-neighbor algorithm, and a cross-entropy minimization solver employs entirely conventional computing components in their conventional capacities. The "max entropy engine" does not represent an improvement to computer architecture; it is simply a label assigned to the software component that performs the claimed mathematical optimization. Labeling a conventional software routine with a specific functional name does not transform it into a particular computing device. In re TLI Commc'ns LLC Patent Litig., (a server and a telephone network are not "particular machines" that render abstract claims patent-eligible when they are invoked merely as tools to execute an abstract process).
Furthermore, to the extent Kamaih involved claims where the predictive model and its optimization were implemented in a manner that was specifically tied to and improved the computing environment such that the computing device was truly integral rather than merely instrumental those facts are not present here. In the present application, the computing system is the environment in which the abstract mathematical process runs, not a component that is technically improved or reconfigured by the claimed invention. As the Federal Circuit held in Two-Way Media Ltd. v. Comcast Cable Commc'ns, LLC, the mere fact that a claim requires a computing system to carry out the recited steps does not mean the computing system is "integral" in the patent-eligibility sense the system must do something beyond what a generic computer conventionally does, and here it does not. For all of the foregoing reasons, the analogy to Kamaih is unpersuasive, and the 101 rejection is hereby maintained as to claims 1, 3, 4, 7, 8, 10, 11, 14, 15, 17, 18, 20, 21, 22, and 23.
The Applicant argues on pgs. 13-14 that “Furthermore, the recent precedential decision in Ex parte Desjardins, Appeal No. 2024000567 (P TAB September 26, 2025) supports the eligibility of amended claim l. Deputy Commissioner for Patents Charles Kim's "Advance notice of change to the MPEP in light of Ex Parte Desjardins" (hereinafter "Desjardins memo") instructs that "when evaluating a claim as a whole, examiners should not dismiss additional elements as mere 'generic computer components' without considering whether such elements confer a technological improvement to a technical problem, especially as to improvements to computer components or the computer system." The Desjardins memo specifically recognizes as eligible " [i Improvements to computer component or system performance based upon adjustments to parameters of a machine learning model associated with tasks or workstreams." Amended claim I is analogous because it explicitly recites that "the determining improves a computational efficiency of a data engine by removing illogical data prior to determination of a total audience size." This is a direct improvement to computer component performance the max entropy engine's constraint-based optimization prevents the data engine from having to process illogical data that would otherwise require additional processing cycles to analyze. Like the claims in Desjardins, the specification identifies this specific improvement to computer system performance, and the claim explicitly reflects that improvement.
Thus, the claimed invention provides an improvement to a technical field, namely that of media audience measurement technology, and provides a particular solution to a problem arising in that technical field. Because the claimed invention provides such an improvement, the additional elements in combination integrate the alleged exception into a practical application. See MPEP 2106.04(d)(II). Therefore, amended claim I satisfies the criteria for subject matter eligibility because the claim as a whole integrates the recited judicial exception into a practical application of the exception”.
The Examiner respectfully disagrees.
Applicant's reliance on Ex parte Desjardins, Appeal No. 2024-000567 (PTAB September 26, 2025, Appeals Review Panel Decision) (precedential), and the accompanying memorandum from Deputy Commissioner for Patents Charles Kim dated December 5, 2025 ("Desjardins Memo"), has been fully and carefully considered. The Examiner acknowledges the precedential weight of Desjardins and the corresponding MPEP updates, and has re-evaluated the claims of the present application in their entirety in light of that decision and the revised guidance. Notwithstanding this re-evaluation, the 101 rejection is maintained for the reasons set forth below, as the present claims are factually and legally distinguishable from the claims found eligible in Desjardins in material respects that are dispositive of the eligibility analysis.
The Desjardins Memo instructs that the eligibility determination under Step 2A, Prong 2 involves a two-part inquiry: first, whether the specification provides sufficient details such that one of ordinary skill in the art would recognize the claimed invention as providing an improvement in the functioning of a computer or an improvement to other technology or a technical field and critically, the specification must not set forth the improvement only in a conclusory manner, i.e., as a bare assertion without the detail necessary to be apparent to a person of ordinary skill in the art; and second, whether the claim itself reflects the disclosed improvement, meaning the claim must include the components or steps of the invention that provide the improvement described in the specification, though the claim need not explicitly recite the improvement verbatim. The Examiner has applied both prongs of this framework to the present claims and concludes that neither is satisfied for the reasons below.
Applicant identifies the specification's statement at paragraph [0070] that the processes "increase the computational efficiency of the data engine 128 by removing illogical data that would require additional processing cycles to analyze" as the disclosed improvement. The Examiner acknowledges that this language appears in the specification and has been incorporated into the claim language. However, as the Desjardins Memo expressly warns, an improvement stated "only in a conclusory manner (i.e., a bare assertion of an improvement without the detail necessary to be apparent to a person of ordinary skill in the art)" does not satisfy the first prong. The statement at paragraph [0070] is precisely such a bare assertion. The specification does not describe the technical architecture of the data engine, does not explain what specific computational mechanisms are affected by receiving logically consistent versus inconsistent data, does not quantify or technically characterize the processing cycles that are reduced, and does not describe how the data engine itself operates differently as a result of the claimed process. The disclosure amounts to the general proposition that processing consistent data requires fewer computational resources than processing inconsistent data a truism that applies to virtually any data processing system and that does not constitute a technically specific disclosure of an improvement to a computer or computing system. This stands in stark contrast to the Desjardins specification, which described with technical specificity how the machine learning model was trained to overcome the known and technically characterized problem of "catastrophic forgetting" in continual learning systems, including specific disclosures of reduced storage capacity, reduced system complexity, and the preservation of performance attributes across sequential tasks all of which were identified as improvements to how the machine learning model itself operates in operation. Desjardins at pp. 8-9; Desjardins Memo at pp. 1-2. The technical specificity of the Desjardins specification in describing how the ML model itself functioned differently is the precise type of disclosure the first prong requires, and no analogous specificity is present in the present application's specification with respect to the data engine.
Even assuming arguendo that the specification sufficiently disclosed an improvement, the second prong requires that the claim include the components or steps that provide the improvement. The Desjardins Memo identifies the critical claim limitation in Desjardins as "adjust the first values of the plurality of parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task." This limitation directly recited a specific structural and operational mechanism the specific manner in which the ML model's parameters were adjusted that was the mechanism by which the improvement (overcoming catastrophic forgetting) was achieved. The improvement was inseparable from the technical operation of the ML model itself, and the claim recited the specific components and steps through which the ML model was configured to achieve that improvement.
In the present claims, the step that purportedly provides the improvement to the data engine is the "determining" step specifically, determining final duplication factors by a max entropy engine utilizing a numeric solver by minimizing cross-entropy subject to consistency and logical constraints. The Examiner finds that this step does not include components or steps that provide an improvement to a computer or computing system in the manner required by the Desjardins framework. The determining step is, in its entirety, a mathematical optimization process: it takes numerical inputs (reference duplication factors, total exposure metrics), applies mathematical constraints (linear consistency requirements expressed as Ax = b, logical bounds as confirmed in specification paragraphs [0059]-[0064]), and solves a mathematical optimization problem (cross-entropy minimization via a numeric dual solver, as confirmed in specification paragraph [0067]) to produce numerical outputs (final duplication factors). The "removal of illogical data" is not a separate technical step it is the natural and expected mathematical result of enforcing consistency and logical constraints in the optimization, which prevents the solution from producing values that violate those mathematical constraints. There is no additional technical mechanism by which the claimed computing system removes illogical data; the mathematical constraints themselves enforce logical consistency. Characterizing the mathematical result of enforcing mathematical constraints as a technical improvement to a data engine does not convert the mathematical operation into a technical improvement to a computer. Digitech Image Techs., LLC v. Elecs. for Imaging, Inc.
A fundamental distinction between Desjardins and the present application is the locus and nature of the alleged improvement. In Desjardins, the improvement was to how the machine learning model itself operates the ML model was technically configured in a specific and unconventional manner to address a known technical problem inherent to continual learning AI systems, namely catastrophic forgetting, which causes an ML model to lose previously learned information when learning new tasks. The improvement was intrinsic to the computational operation of the ML system. In the present application, the alleged improvement is not to how the machine learning model itself operates. The machine learning model recited in claim 1 is a conventional Random Forests Regression model, as confirmed by the specification at paragraph [0038], and is trained in a conventional manner the training step does not reflect any improvement to how the ML model functions. Rather, the alleged improvement is to a separate downstream component the "data engine" which is described in the specification as a component that "further processes the total audience data" output by the max entropy engine. The present claims thus describe a pipeline in which a conventional ML model and a mathematical optimization process produce numerical outputs, and those outputs are then processed by a downstream data engine. The asserted improvement is that the outputs are logically consistent, which allegedly makes the data engine's processing more efficient. This is not an improvement to the machine learning model or to the computing system that performs the claimed steps it is a benefit to a downstream processing component that receives the output of the abstract mathematical process, and it is achieved entirely through the enforcement of mathematical constraints, not through any unconventional computer architecture, data structure, or processing mechanism.
Applicant correctly notes that the Desjardins Memo adds new example xiv to MPEP 2106.05(a): "Improvements to computer component or system performance based upon adjustments to parameters of a machine learning model associated with tasks or workstreams." The Examiner observes that this example, consistent with the Desjardins decision itself, describes improvements that arise from specific adjustments to ML model parameters that cause the ML model itself to perform better for example, parameter adjustments that protect previously learned knowledge (task performance) while enabling new task learning, thereby reducing storage, complexity, and computational overhead inherent in the ML model's operation. The present claims do not recite adjustments to ML model parameters that improve the ML model's own performance, computational efficiency, or operational characteristics. The ML model in the present claims is trained to predict duplication factors a conventional use of a machine learning model and no parameter adjustments are claimed that alter how the ML model itself functions to achieve a technical improvement to the model. Accordingly, example xiv does not support eligibility of the present claims.
The Desjardins Memo further instructs that "examiners should not dismiss additional elements as mere 'generic computer components' without considering whether such elements confer a technological improvement to a technical problem." The Examiner has not dismissed the recited elements the computing system, the max entropy engine, the numeric solver, and the machine learning model without consideration. Rather, the Examiner has fully considered each of these elements and the claim as a whole as an ordered combination, and has concluded that they do not confer a technological improvement to a technical problem because: (i) the machine learning model is a conventional Random Forests Regression model used in its conventional capacity; (ii) the max entropy engine utilizing a numeric solver implements a standard mathematical optimization methodology; and (iii) the asserted improvement to the data engine is a downstream consequence of performing the abstract mathematical optimization, not a technical mechanism inherent to how the claimed computing system itself operates. The instruction not to oversimplify or generalize claims has been heeded but careful, element-by-element and combination analysis of the specific claim language confirms that the claimed steps do not reflect an improvement to a computer or computing system within the meaning of the Desjardins/Enfish/McRO line of cases.
For all of the foregoing reasons, the Desjardins decision and the accompanying MPEP updates, while fully considered and applied, do not support a finding of eligibility for the present claims. The present claims are distinguishable from Desjardins because the specification's disclosure of a computing improvement is conclusory, the claimed steps do not include the components or mechanisms through which any such improvement is technically achieved, and the asserted improvement is to a downstream data processing component achieved through conventional mathematical optimization rather than an improvement to how the claimed computing system itself operates. The 101 rejection is therefore maintained.
The Applicant argues The Reminders also state that "Examiners are reminded that if it is a 'close call' as to whether a claim is eligible, they should only make a rejection when it is more likely than not (i.e., more than 500 0) that the claim is ineligible under 35 U.S.C. 101." See Reminders, p. 5. Amended claim I is not an "idea of a solution." Amended claim I reflects a specific improvement in the technology of media audience measurement by requiring the use of a max entropy engine that minimizes cross-entropy subject to consistency constraints to achieve computationally efficient results. This integration of specific technical constraints distinguishes it from ineligible "apply it" claims. Accordingly, Applicant submits that it is not more likely than not that amended claim I is ineligible under 35 U.S.C. 101. Therefore, a subject matter rejection of amended claim I under 101 is not appropriate.
For at least the above-referenced reasons, amended claim I qualifies as eligible subject matter under 35 U.S.C. 101. And for largely the same reasons, amended claims 8 and 1 5 also qualify as eligible subject matter under 35 U.S.C. 101. Claims 3, 4, 7, 10, I l, 14, 17, 18, and 20 depend from claims l, 8, and 15 and are likewise patent eligible.
The Examiner respectfully disagrees.
With respect to the argument the Examiner notes that the Applicant's invocation of the "more likely than not" standard for 101 rejections, along with the contention that amended claim 1 presents a close call that should resolve in favor of eligibility, has been fully considered but is unpersuasive, and the 101 rejection is hereby maintained as to all pending claims for the reasons set forth below and as previously established in the Examiner's responses to Arguments 1 through 5, which are incorporated herein by reference in their entireties.
The Examiner acknowledges the USPTO's guidance that examiners should make a 101 rejection only when it is more likely than not i.e., more than fifty percent likely that the claim is directed to ineligible subject matter. This standard is intended to prevent examiners from making speculative, marginal, or inadequately supported rejections where genuine ambiguity exists as to whether a claim falls within the scope of a judicial exception. It is not intended to require certainty before a rejection may issue, and it is not a tiebreaker that automatically resolves in an applicant's favor whenever a claim involves any degree of technical character. Critically, the standard does not alter the substantive legal framework established by Alice, Mayo, the 2019 PEG, or the Desjardins precedent it operates as a threshold confidence requirement for the examiner's legal conclusion, not as an independent basis for eligibility.
Having fully applied the two-step Alice/Mayo framework, including the updated guidance under Ex parte Desjardins as addressed in the response to Argument 5, the Examiner is well beyond the "more likely than not" threshold and is firmly of the view that the claims are directed to patent-ineligible subject matter. This is not a close call. The analysis is not speculative, not marginal, and not based on an oversimplified characterization of the claims. As established in the responses to Arguments 1 through 5, the Examiner has: (i) carefully identified the specific judicial exceptions recited in each step of the independent claims mathematical operations, mathematical relationships, and mathematical optimization techniques without collapsing the claims to an impermissible level of generality; (ii) fully considered each of the claim limitations as an ordered combination, including the amended language reciting cross-entropy minimization, consistency constraints, and improvement to computational efficiency of the data engine; (iii) applied the Desjardins framework in full, including both prongs of the Step 2A Prong 2 analysis, and determined that neither the specification's disclosure nor the claim language satisfies that framework; and (iv) evaluated each additional element individually and in combination for whether it amounts to significantly more than the judicial exception. At each stage of this multi-layered analysis, the conclusion has been the same: the claims are directed to mathematical concepts applied to media campaign data on a generic computing platform, and neither the amended claim language nor the specification's characterizations establish a technical improvement to a computer or computing system that would integrate the judicial exception into a practical application. Where the analysis is this consistent and this thoroughly grounded in applicable legal authority, the "more likely than not" threshold is clearly satisfied.
Applicant argues that amended claim 1 does not merely claim an "idea of a solution" but rather reflects a specific improvement by requiring the use of a max entropy engine that minimizes cross-entropy subject to consistency constraints. The Examiner has given careful consideration to this argument in the context of the full claim language and finds it unpersuasive for reasons already articulated in detail in the responses to Arguments 2, 3, and 5. The distinction between an "idea of a solution" and a "particular solution" under McRO and its progeny turns on whether the claim recites specific technical means or mechanisms that achieve a technical result in a way that is meaningfully different from prior approaches, such that the claim covers a particular way of achieving the outcome rather than the outcome itself. McRO, Inc. v. Bandai Namco Games Am. Inc. Here, the claimed "specific" mechanism a max entropy engine utilizing a numeric solver that minimizes cross-entropy subject to consistency and logical constraints is, as established throughout this action, a specific mathematical optimization methodology applied to numerical data. The specificity is mathematical, not technical. Claiming a particular mathematical algorithm to solve a mathematical problem is not the same as claiming a particular technical solution to a technical problem, and the Federal Circuit has made this clear. SAP Am., Inc. v. InvestPic, ("No matter how impressive the mathematical execution or clever the actual mechanism, its core is still an abstract idea and is not transformed into something concrete by the particular mathematical implementation chosen"). The max entropy engine does not reconfigure the computer, does not implement an unconventional data structure, does not establish a novel hardware architecture, and does not alter how the processor or memory operates it implements a well-established mathematical optimization technique on a generic computing platform. Mathematical specificity does not substitute for technical specificity in the 101 analysis.
The Dependent Claims: With respect to the dependent claims, which Applicant addresses collectively as depending from eligible independent claims, the Examiner maintains the rejection as to each of these claims. Because the independent claims 1, 8, and 15 remain rejected under 101, the dependent claims, which add further limitations to those independent claims, are likewise rejected. The additional limitations recited in the dependent claims training the machine learning model using respective features of media campaigns (claims 3, 10, 17); features including demographics, media campaign time step, media platform reach, or digital duplicated reach (claims 4, 11, 18); and distance as a Euclidean distance based on a KD tree (claims 7, 14, 20) each recite additional mathematical concepts and well-known computational techniques, as established in the rejection. None of these additional limitations introduce a technological improvement to a computer or computing system that would cure the eligibility deficiency of the independent claims. Similarly, new claims 21, 22, and 23, which recite linear interpolation to common time points, threshold distance identification, and application of consistency and logical requirements respectively, add further mathematical constraints and operations that do not transform the claims into patent-eligible subject matter for the reasons set forth in the rejection and throughout this action.
For the foregoing reasons, and for all of the reasons set forth in the rejection and in the Examiner's responses to Arguments 1 through 5, the 101 rejection is maintained as to claims 1, 3, 4, 7, 8, 10, 11, 14, 15, 17, 18, 20, 21, 22, and 23. The Examiner has conducted a thorough, multi-layered eligibility analysis that fully accounts for the specific claim language as amended, the specification's disclosures, controlling Federal Circuit and Supreme Court precedent, the 2019 Revised Patent Subject Matter Eligibility Guidance, and the precedential Ex parte Desjardins decision and accompanying MPEP updates. The rejection is well-founded and clearly exceeds the "more likely than not" threshold. Applicant is encouraged to consider whether further amendment to the claims particularly amendments directed to reciting specific, unconventional technical mechanisms through which the claimed computing system itself is improved in its operation, beyond asserting the downstream benefit of performing a mathematical optimization could establish eligibility under the applicable legal standards. The rejection is therefore maintained.
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, 3, 4, 7, 8, 10, 11, 14, 15, 17, 18, 20, 21, 22, and 23 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea specifically, mathematical concepts and certain methods of organizing human activity without significantly more. The claims recite mathematical operations, mathematical relationships, and mathematical calculations directed to training and applying a machine learning model, performing nearest-neighbor campaign selection via a KD tree using Euclidean distance, and applying cross-entropy minimization subject to mathematical consistency and logical constraints to estimate audience duplication factors across media platforms. This judicial exception is not integrated into a practical application because the additional elements do not impose meaningful limits on the practice of the judicial exception and do not reflect a specific improvement to the functioning of a computer or to any other technology or technical field. The claims do not include additional elements sufficient to amount to significantly more than the judicial exception because they recite only generic computing components (a computing system comprising a processor and memory) and well-understood, routine, and conventional mathematical algorithms and software modules executing on that generic platform.
Step 1
Regarding Step 1 of the Subject Matter Eligibility Test, the pending claims fall within the following statutory categories of invention:
• Claims 15, 17, 18, and 20 are directed to a process/method and fall within the process category of 35 U.S.C. 101.
• Claims 8, 10, 11, and 14 are directed to a non-transitory computer-readable medium a manufacture and fall within the manufacture category of 35 U.S.C. 101.
• Claims 1, 3, 4, 7, 21, 22, and 23 are directed to a computing system and fall within the machine category of 35 U.S.C. 101.
All pending claims satisfy at least one of the four statutory categories of invention. Accordingly, the claims pass Step 1 (Step 1: YES) and the analysis proceeds to Step 2A.
STEP 2A, Prong One
The claims recite an abstract idea. Specifically, the independent claims 1, 8, and 15 from which all dependent claims depend recite the following limitations that constitute a judicial exception:
(1) Training a machine learning model to predict duplication factors for a media campaign based on total exposure metrics associated with respective media platforms, wherein the machine learning model is trained using training data comprising a plurality of media campaigns having known total exposure metrics and known duplication factors;
(2) Predicting, using the machine learning model, estimated duplication factors for a query media campaign based on total exposure metrics associated with respective media platforms, wherein duplication of media exposure across different possible combinations of the media platforms is unknown for the query media campaign;
(3) Selecting a reference media campaign from the plurality of media campaigns based on a comparison of the estimated duplication factors with reference duplication factors for the reference media campaign; and
(4) Determining, by a max entropy engine utilizing a numeric solver, final duplication factors for the query media campaign by minimizing a cross-entropy between the reference duplication factors of the selected reference media campaign and the final duplication factors, wherein the determining is based on (a) a distance between the final duplication factors and the reference duplication factors and (b) constrained by consistency between i) total exposure metrics associated with the respective media platforms for the query media campaign and ii) unique audiences for the different possible combinations of the media platforms derived from the final duplication factors.
Abstract Idea Grouping Analysis
Mathematical Concepts [Primary Grouping]
Each of the four identified limitations recites a mathematical concept within the meaning of MPEP 2106.04(a)(2), subsection I, specifically mathematical operations, mathematical relationships, and mathematical calculations, as follows:
Limitation (1) Training a machine learning model: Training a machine learning model is a mathematical operation. Model training involves iterative updating of mathematical parameters (weights, coefficients, regression tree split values) through statistical optimization applied to training data. The specification confirms at paragraph [0038] that the specific model is a Random Forests Regression model a well-known statistical/mathematical ensemble technique. Mapping input feature vectors of total exposure metrics to output numerical values of duplication factors is a mathematical relationship.
Limitation (2) Predicting estimated duplication factors: Predicting numerical output values (estimated duplication factors) by applying a trained mathematical function (the machine learning model) to a numerical input set (total exposure metrics) is a mathematical calculation. The prediction step is the evaluation of a mathematical function it is purely numerical in character.
Limitation (3) Selecting a reference media campaign by comparison: Selecting a campaign based on comparison of numerical duplication factor vectors is a mathematical operation. The specification at paragraphs [0043]-[0056] confirms this step is implemented via a KD tree nearest-neighbor search using Euclidean distance an explicit mathematical algorithm. Claim 7 (and parallel claims 14 and 20) further confirm this by explicitly reciting that 'the distance is a Euclidean distance based on a KD tree.' Identifying the minimum Euclidean distance among candidate numerical vectors is a mathematical calculation.
Limitation (4) Determining final duplication factors via cross-entropy minimization: Cross-entropy minimization is a mathematical optimization technique. The consistency constraints expressed as a linear system Ax = b (specification paragraph [0060]), the logical requirements establishing numerical bounds (paragraphs [0061]-[0063]), and the deviation requirements minimizing cross-entropy (paragraph [0065]) are all mathematical relationships and mathematical operations. The specification at paragraph [0067] states the max entropy engine 'utilizes numeric solvers that optimize the dual and minimize the duality gap' an explicit mathematical optimization methodology.
Certain Methods of Organizing Human Activity [Secondary Grouping]
The claims additionally fall within the 'commercial or legal interactions' sub-grouping of certain methods of organizing human activity (MPEP 2106.04(a)(2), subsection II). The overarching purpose of the claimed process estimating deduplicated audience overlap across advertising media platforms to determine total audience metrics for a media campaign is an audience measurement and commercial campaign analysis activity. Collecting, processing, and reporting audience data for advertising campaigns constitutes a method of organizing human commercial activity. See Electric Power Group, LLC v. Alstom S.A., (collecting, analyzing, and displaying information, even limited to a specific commercial environment, is an abstract idea).
Accordingly, the claims recite a judicial exception mathematical concepts (mathematical operations, mathematical relationships, mathematical calculations) and certain methods of organizing human activity (commercial audience measurement and campaign analytics). The analysis proceeds to Step 2A, Prong Two.
STEP 2A, Prong Two:
Identification of Additional Elements Beyond the Abstract Idea
The claims recite the following additional elements beyond the identified abstract idea:
• A computing system comprising a processor and a memory (independent claims 1, 8, 15 structural preamble)
• A max entropy engine utilizing a numeric solver (recited within the determining step of independent claims 1, 8, 15)
• The assertion that 'the determining improves a computational efficiency of a data engine by removing illogical data prior to determination of a total audience size' (recited as a functional result clause in independent claims 1, 8, 15 as amended)
• A KD tree-based Euclidean distance for campaign selection (dependent claims 7, 14, 20)
• Interpolation to common time points to normalize campaign durations (new claim 21)
• Identification of the reference media campaign as having a distance to the query campaign within a threshold distance (new claim 22)
• Application of consistency requirements and logical requirements by the max entropy engine (new claim 23)
Analysis of Additional Elements
Improvement to Technology or Technical Field (MPEP 2106.05(a)) As Updated by Ex parte Desjardins DESJARDINS COMPLIANCE NOTE: The Examiner has fully considered Ex parte Desjardins, Appeal No. 2024-000567 (PTAB September 26, 2025) (precedential) and the Deputy Commissioner's Memorandum of December 5, 2025 ('Desjardins Memo') updating MPEP §2106.04(d), 2106.05(a), and 2106.05(f). The claims have been evaluated as a whole, without oversimplification, and without dismissing additional elements as generic components without first considering whether they confer a technological improvement. The following analysis reflects that full consideration.
The claims do not recite an improvement to the functioning of a computer or to any other technology or technical field sufficient to integrate the judicial exception into a practical application. The Desjardins Memo instructs that the eligibility determination involves a two-part inquiry: (1) whether the specification provides sufficient, non-conclusory details that would cause one of ordinary skill in the art to recognize an improvement in the functioning of a computer or other technology, and (2) whether the claim reflects that disclosed improvement through the components or steps recited in the claim. Neither prong is satisfied here.
First Prong (Specification Disclosure): The specification asserts at paragraph [0070] that the max entropy engine processes reduce the operating efficiency of the data engine 128 by 'removing illogical data that would require additional processing cycles to analyze.' However, this statement is a bare, conclusory assertion without the technical detail necessary to be apparent to a person of ordinary skill in the art, as the Desjardins Memo expressly warns against. The specification does not describe the architecture of the data engine, does not specify what computational mechanisms are affected by receiving logically consistent data versus inconsistent data, does not quantify the processing cycles reduced, and does not explain how the data engine itself operates differently at a technical level. The disclosure amounts to the general proposition that a downstream processor benefits from receiving accurate inputs a truism applicable to any data processing system that does not constitute a technically specific disclosure of a computer improvement. Contrast this with the Desjardins specification, which described with technical specificity how the machine learning model itself was trained to overcome the known technical problem of 'catastrophic forgetting,' with specific improvements to storage capacity, system complexity, and preservation of prior task performance improvements that were intrinsic to how the ML model itself operated. No analogous technical specificity is present here.
Second Prong (Claim Reflection): Even assuming the specification sufficiently disclosed an improvement, the claim must include components or steps that provide the improvement. The determining step which purportedly provides the computational efficiency improvement is, in its entirety, a mathematical optimization process: applying mathematical constraints (linear system Ax = b; logical bounds per specification paragraphs [0059]-[0064]) and minimizing cross-entropy via a numeric dual solver (paragraph [0067]) to produce numerically consistent outputs. The 'removal of illogical data' is not a separate technical mechanism it is the natural mathematical result of enforcing the mathematical constraints, which prevent the solution from producing values violating those constraints. There is no additional hardware structure, data architecture, or processing mechanism that achieves the improvement; the mathematical constraints themselves enforce logical consistency. This is distinct from Desjardins, where the specific claim limitation regarding adjusting ML model parameters to protect prior task performance directly recited the mechanism by which the technical improvement overcoming catastrophic forgetting was achieved. The amended functional result clause ('wherein the determining improves a computational efficiency of a data engine') asserts an outcome of the abstract mathematical process, not a technical mechanism. See Two-Way Media Ltd. v. Comcast Cable Commc'ns, LLC.
Nature of the Improvement Distinguished from Desjardins: In Desjardins, the improvement was to how the machine learning model itself operates the model was technically configured to address a known problem intrinsic to continual learning AI systems (catastrophic forgetting), producing improvements to the model's own storage efficiency, complexity, and cross-task performance. In the present application, the alleged improvement is not to how the machine learning model operates (the model here is a conventional Random Forests Regression model trained conventionally) but rather to a separate, downstream data engine that receives the output of the mathematical optimization. An improvement to a downstream processing component achieved by producing better-quality mathematical outputs rather than by reconfiguring the computing system itself does not constitute an improvement to computer functionality. Enfish, LLC v. Microsoft Corp.
New MPEP Example xiv Distinguished: The Desjardins Memo adds new MPEP 2106.05(a) example xiv: 'Improvements to computer component or system performance based upon adjustments to parameters of a machine learning model associated with tasks or workstreams.' This example, consistent with the Desjardins decision, describes improvements arising from specific parameter adjustments that cause the ML model itself to perform better. The present claims do not recite parameter adjustments that improve the ML model's own performance characteristics. The ML model is trained conventionally to predict duplication factors, and no parameter adjustments are claimed that alter how the model itself functions to achieve a technical improvement to the model or computing system.
Particular Machine (MPEP 2106.05(b))
The claims do not recite use of a particular machine that imposes meaningful limits on the claim scope. The recited 'computing system comprising a processor and a memory' is a generic computing platform. The specification confirms at paragraph [0085] that the processor platform encompasses conventional computing devices including servers, personal computers, workstations, and mobile devices. The 'max entropy engine utilizing a numeric solver' is a software module implementing a mathematical optimization algorithm, not a specialized hardware device. The specification further confirms at paragraph [0080] that the machine learning engine, including the max entropy engine, 'may be implemented by hardware, software, firmware and/or any combination,' encompassing conventional programmable processors, GPUs, DSPs, and ASICs. A generic processor running a mathematical algorithm is not a 'particular machine' within the meaning of MPEP 2106.05(b).
Mere Instructions to Apply the Exception (MPEP 2106.05(f))
The additional elements amount to no more than mere instructions to implement the abstract idea on a generic computer. Reciting 'a computing system comprising a processor and a memory' as the platform for performing the claimed mathematical operations is tantamount to adding 'apply it on a computer' to the judicial exception. See Alice Corp. Pty. Ltd. v. CLS Bank Int'l. The functional result clause ('wherein the determining improves a computational efficiency of a data engine') is merely an instruction that the abstract mathematical process should be applied to achieve a particular outcome, without specifying the technical mechanism by which that outcome is achieved beyond the abstract mathematical steps already recited. Under the Desjardins Memo's updated MPEP 2106.05(f), a claim reciting a 'technological solution to a technological problem' such as DDR Holdings (overriding the conventional sequence of events triggered by a hyperlink) or Desjardins (overcoming catastrophic forgetting through specific ML parameter adjustment) does not constitute mere 'apply it' language. The present claims do not qualify because, as established above, the claimed process addresses a mathematical/statistical problem (estimating consistent duplication factors) through mathematical optimization, not a technological problem through a technological solution.
Insignificant Extra-Solution Activity (MPEP 2106.05(g))
To the extent any additional elements could be viewed as separate from the core mathematical process, they constitute insignificant extra-solution activity. The training step's use of 'training data comprising a plurality of media campaigns having known total exposure metrics and known duplication factors' is mere data gathering the collection of input data necessary to perform the abstract mathematical process. The output of the determining step total audience size determination is the mere reporting of the mathematical result. Mere data gathering and data reporting steps that are incidental to the core abstract process do not integrate a judicial exception into a practical application. MPEP 2106.05(g).
Considering the additional elements individually and in combination, and evaluating the claim as an ordered combination without oversimplification and without dismissing elements as generic without consideration as required by the Desjardins Memo, the claim as a whole does not integrate the judicial exception into a practical application. The additional elements do not impose meaningful limits on practicing the abstract idea, do not reflect a specific improvement to computer functionality or other technology as required by the two-prong Desjardins/Enfish/McRO framework, and do not constitute more than instructions to apply the exception on a generic computing platform. Accordingly, the claims are directed to an abstract idea. The analysis proceeds to Step 2B.
STEP 2B
As discussed with respect to Step 2A, Prong Two, the additional elements in the claims amount to no more than mere instructions to apply the exception using generic computer components performing well-understood, routine, and conventional activities. The same analysis applies in Step 2B mere instructions to apply an exception using generic computer components cannot provide an inventive concept. See MPEP 2106.05(f); Alice Corp., 573 U.S. at 225-26.
Well-Understood, Routine, and Conventional Activity Analysis
The additional elements, when considered individually and in combination, are well-understood, routine, and conventional activities in the field. Specifically:
Generic computing system (processor and memory): The courts have recognized that a generic computer processor, memory, and associated hardware is well-understood, routine, and conventional. See Alice Corp.; Mayo Collaborative Servs. v. Prometheus Labs., Inc.,. The specification confirms at paragraph [0085] that the processor platform encompasses conventional computing devices.
Machine learning model (Random Forests Regression): The specification confirms at paragraph [0038] that the machine learning model is a Random Forests Regression model an explicitly identified, well-known statistical learning technique. Random Forests, KD trees, and cross-entropy minimization are all established, conventional computational tools widely deployed in data science and statistical analysis prior to the filing of this application. The courts have recognized that applying conventional machine learning techniques is well-understood, routine, and conventional activity. See Customedia Techs., LLC v. Dish Network Corp.
KD tree nearest-neighbor algorithm: A KD tree is a well-known, conventional data structure for space-partitioning and nearest-neighbor search, described in computer science literature since the 1970s. The specification itself at paragraph [0043] describes the KD tree as a known method ('The KD tree can be interpreted as a method to find the neighboring points by using a binary tree...'), confirming it is not novel. Using a conventional algorithm does not provide an inventive concept. BSG Tech LLC v. BuySeasons, Inc.
Cross-entropy minimization via numeric solver: Cross-entropy minimization is a well-established mathematical optimization methodology. The specification at paragraph [0067] confirms use of numeric solvers that 'optimize the dual and minimize the duality gap' standard convex optimization techniques. Implementing well-known mathematical optimization using conventional numeric solvers is a well-understood, routine, and conventional activity. See SAP Am., Inc. v. InvestPic, LLC.
Linear interpolation (claim 21): Linear interpolation using the algebraic equation of a line as described in the specification at paragraphs [0071]-[0078] with the explicit algebraic derivation y = mx + b is a fundamental, well-understood mathematical technique that does not provide an inventive concept.
Threshold distance identification (claim 22) and consistency/logical requirements (claim 23): Limiting a nearest-neighbor search to candidates within a threshold distance and applying mathematical consistency and logical bounds constraints to an optimization problem are standard, well-understood algorithmic and mathematical techniques that do not provide an inventive concept.
Considered as an ordered combination, these elements do not provide significantly more. The combination of (i) training a Random Forests Regression machine learning model, (ii) performing a KD tree Euclidean nearest-neighbor search, and (iii) applying cross-entropy minimization with linear and logical constraints is a combination of well-known mathematical and computational techniques. Adding one abstract idea or mathematical concept to another does not render the claim non-abstract. RecogniCorp, LLC v. Nintendo Co.
Step 2B Conclusion
Considering the additional elements individually and in combination, the claims do not include additional elements sufficient to amount to significantly more than the judicial exception. The claims are not patent-eligible (Step 2B: NO).
Dependent Claims Analysis
The dependent claims do not add limitations that integrate the judicial exception into a practical application or provide an inventive concept:
Claims 3, 10, and 17 (training using respective features of media campaigns): These claims specify that the machine learning model is trained using respective features of the plurality of media campaigns. Identifying and using training features is a standard, conventional step in machine learning model development. This additional limitation further narrows the mathematical training process but does not add any element that integrates the exception into a practical application or provides significantly more.
Claims 4, 11, and 18 (features include demographics, time step, platform reach, digital duplicated reach): These claims further specify the types of features used in model training demographics (gender x age groups), media campaign time steps, media platform reach, and digital duplicated reach. Selecting the input features for a machine learning model is a well-understood, routine, and conventional aspect of machine learning. The features themselves demographic data, platform exposure metrics are standard data inputs in audience measurement. These limitations further specify the mathematical inputs to the abstract process but do not add elements that impose meaningful limits or provide an inventive concept.
Claims 7, 14, and 20 (Euclidean distance based on KD tree): These claims specify that the distance used in campaign selection is a Euclidean distance based on a KD tree. As established above, KD tree nearest-neighbor search using Euclidean distance is a well-known, conventional mathematical algorithm. These limitations further specify the mathematical method of campaign comparison but do not add elements that integrate the exception into a practical application or provide significantly more.
Claim 21 (interpolating to common time points): This claim specifies that training comprises interpolating total exposure metrics and known duplication factors to correspond to common time points to normalize durations. Linear interpolation to normalize temporal data is a well-understood, routine mathematical technique. This limitation further specifies a conventional pre-processing mathematical step but does not add elements sufficient to amount to significantly more.
Claim 22 (selecting reference campaign within threshold distance): This claim specifies that selecting the reference media campaign comprises identifying a campaign having a distance to the query campaign within a threshold distance. Limiting a search to candidates within a threshold distance metric is a standard algorithmic technique. This limitation further specifies the mathematical campaign selection step but does not integrate the exception into a practical application.
Claim 23 (max entropy engine applies consistency and logical requirements): This claim specifies that the max entropy engine applies: (i) consistency requirements to ensure final duplication factors add up to total exposure metrics, and (ii) logical requirements to ensure final duplication factors are self-consistent. As established in the main rejection, these requirements are mathematical constraints the linear system Ax = b and the inequality bounds that are standard features of constrained mathematical optimization. They do not add elements that integrate the exception into a practical application or provide significantly more.
For the foregoing reasons, claims 1, 3, 4, 7, 8, 10, 11, 14, 15, 17, 18, 20, 21, 22, and 23 are rejected under 35 U.S.C. 101 as being directed to non-statutory subject matter. The Examiner has conducted a thorough, multi-layered eligibility analysis fully accounting for: (i) the specific claim language as amended; (ii) the specification's disclosures; (iii) controlling Supreme Court and Federal Circuit precedent (Alice Corp., Mayo, Enfish, McRO, Electric Power Group, Two-Way Media, SAP America, RecogniCorp); (iv) the 2019 Revised Patent Subject Matter Eligibility Guidance and October 2019 Update; and (v) the precedential Ex parte Desjardins decision (Appeal No. 2024-000567, PTAB September 26, 2025) and the accompanying Deputy Commissioner's Memorandum of December 5, 2025 updating MPEP 2106.04(d), 2106.05(a), and 2106.05(f).
The rejection is well-founded and clearly exceeds the 'more likely than not' threshold required for a 101 rejection. The claims are directed to mathematical concepts (machine learning model training, nearest-neighbor KD tree search using Euclidean distance, cross-entropy minimization subject to linear and logical mathematical constraints) and certain methods of organizing human activity (commercial audience measurement and campaign deduplication analytics) judicial exceptions to patent eligibility. The claims do not integrate the judicial exception into a practical application under the Desjardins/Enfish/McRO framework, and the additional elements generic computing hardware, a conventional Random Forests Regression model, and well-known mathematical algorithms do not amount to significantly more than the judicial exception.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to STEPHEN S SWARTZ whose telephone number is (571)270-7789. The examiner can normally be reached Mon-Fri 9:00 - 6:00.
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, Boswell Beth can be reached at 571 272-6737. 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.
/S.S.S/Examiner, Art Unit 3625
/BETH V BOSWELL/Supervisory Patent Examiner, Art Unit 3625