DETAILED ACTION
Status of Claims
The present application, filed on or after 3/16/2013, is being examined under the first inventor to file provisions of the AIA .
This action is in reply to the Remarks and Amendments filed 03/20/2026.
Claims 1, 8, 15 have been amended.
Claims 1-21 have been examined and are pending.
(AIA ) Examiner Note
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 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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned at the time any inventions covered therein were effectively filed absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned at the time a later invention was effectively filed in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), first paragraph:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-21 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claims contain subject matter which is not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor(s), at the time the application was filed, had possession of the claimed invention.
Independent claim 1 recites limitations directed towards the following: A method implemented on at least one processor, a memory, and a communication platform for generating combination distributions for ads via machine learning, comprising: training, by an ad recommendation backend server…; estimating, by the ad recommendation backend server…; generating, by the ad recommendation backend server…; determining…; sending, by the ad recommendation backend server… updating, by the ad recommendation backend server…; updating, by the ad recommendation backend server the prediction model based on the updated training data”
Independent claim 8 is similar to claim 1 albeit being directed towards: Machine readable and non-transitory medium having information stored thereon… the information, when read by the [sic] machine, causes the machine to perform the following steps: training, by an ad recommendation backend server…; estimating, by the ad recommendation backend server…; generating, by the ad recommendation backend server…; determining…; sending, by the ad recommendation backend server… updating, by the ad recommendation backend server…; updating, by the ad recommendation backend server the prediction model based on the updated training data”
Independent claim 15 is also similar to claim 1 albeit being directed towards: A system for generating combination distributions for ads via machine learning, comprising: a frontend ad serving engine including an explore/exploit layer (EEL); and an ad recommendation backend server comprising: a machine learning engine configured for training a prediction model…; estimating…; generating…; determining…; sending…; and a training data processor configure for updating the training data…; wherein the machine learning engine is further configured for updating the prediction model based on the updated training data.”
These limitations recite functionality attributed to a generic “machine learning”; i.e. the method appears to be recited as being accomplished “via machine learning” per the preamble of claim 1 and apparently as executed by “the ad recommendation backend server” as noted per the steps of the recited method. Claim 15 appears to confirm this interpretation of the claimed invention whereby it is “an ad recommendation backend server comprising: a machine learning engine configured for training a prediction model…; estimating…; generating…; determining…; sending…; etc…”;i.e. a generically recited “machine learning engine” for achieving the recited functionality.
However, the broadly claimed functions (i.e. the steps as claimed) are not a general functions that any general purpose computer server (e.g. the generically referenced “ad recommendation backend server”) can perform without special programming. Nonetheless, Applicant asserts that these broadly claimed functions/steps are his invention and alludes to some “machine learning” and/or “machine learning engine” which is not further illuminated in the original disclosure (Spec, drawings, original claims). The Original disclosure does not provide sufficient written description support for the recited functionality – there is no description of any special or particular machine learning capable of performing all of applicant’s recited steps nor is there an algorithm provided by which applicant’s generically recited “ad recommendation backend server” may be programmed to achieve the functionality ascribed to it.
This deficiency further indicates that the Specification describes only "a mere wish or plan for" achieving the functionality, as now claimed, by some generic “machine learning” and/or “ad recommendation backend server”. See Eli Lilly, 119 F .3d at 1566 (citation omitted). As such, the broadly recited limitation "merely recite[s] a description of the problem to be solved," and leaves to future inventors to "complete an unfinished invention." See Ariad, 598 F.3d at 1353. In this case, without the Specification describing any particular algorithm to achieve the claimed function of generating a de-duplicated advertising campaign, one of ordinary skill in the art would not have reasonably concluded that the inventors invented the claimed invention or that they possessed the claimed subject matter at the time of filing of the application. See Vasudevan, 782 F.3d at 683; see also Regents, 119 F.3d at 1566; § 112 Guidance at 61.
To satisfy the written description requirement of 35 U.S.C. § 112, first paragraph, the Specification must reasonably convey to an artisan of ordinary skill that Appellant had possession of the claimed invention at the time the application was filed. Vasudevan Software, Inc. v. MicroStrategy, Inc., 782 F.3d 671, 682 (Fed. Cir. 2015) (citing Ariad Pharm., Inc. v. Eli Lilly & Co., 598 F.3d 1336, 1351 (Fed. Cir. 2010) (en banc)). Functional claim language that merely describes an intended result and fails to support the scope of the claimed invention is insufficient to show possession, even when the claim recitations are found word-for-word in the Specification. Vasudevan, 782 F.3d at 682 ("[t]he written description requirement is not met if the specification merely describes a 'desired result"') (citing Ariad, 598 F.3d at 1349); Enzo Biochem, Inc. v. Gen-Probe, Inc., 323 F.3d 956, 968 (Fed. Cir. 2002) ("[t]he appearance of mere indistinct words in a specification or a claim, even an original claim, does not necessarily satisfy" the written description requirement). The Specification must explain, for example, how Appellant intended to achieve the claimed function to satisfy the written description requirement. Vasudevan, 782 F.3d at 683. While "[t]here is no rigid requirement that the disclosure contain 'either examples or an actual reduction to practice,"' due to the written description requirement, the Specification must set forth "an adequate description that 'in a definite way identifies the claimed invention' in sufficient detail such that a person of ordinary skill would understand that the inventor had made the invention at the time of filing." Allergan, Inc. v. Sandoz Inc., 796 F.3d 1293, 1308 (Fed. Cir. 2015) (citing Ariad, 598 F.3d at 1352); see also Examining Computer-Implemented Functional Claim Limitations for Compliance with 35 US.C. 112, 84 F.R. 57, 61- 62 (January 7, 2019) ("112 Guidance").
Here, the Specification does not sufficiently support any particular “machine learning” and/or algorithm for programming the “ad recommendation backend server” which is capable of performing the recited functionality as recited in the aforementioned limitations; i.e. as noted supra and recited as noted per claims 1, 8, and/or 15.
The specification does not provide a single technique nor method nor algorithm regarding how the recited functionality is intended to be accomplished based upon the given generic input. Instead, the Specification at [0008] merely asserts an “…A prediction model is obtained via machine learning with respect to a criterion…" and per [0009]: “…The machine learning engine is configured for obtaining a prediction model based on training data with respect to a criterion…”; paragraph [0039] makes another general assertion: “[0039] The ad recommendation back end server 23 0 herein is also provided to obtain asset combination prediction models 260 via machine learning based on training data …” The remainder of the Specification is notably void of any particular discussion or claim to a novel or inventive machine learning which is capable of performing the steps now attributed to it.
Under these circumstances, the Examiner has determined the Specification merely states a wish that machine learning could be used to perform the claimed steps as recited in the claims, but the Specification does not demonstrate the Applicant was in possession of such a capable machine learning nor how applicant intended to create such machine learning capable of performing the steps now claimed. The Specification does not describe the claimed invention in sufficient detail such that an ordinarily skilled artisan would understand that the inventor had made the invention at the time of filing. Thus, the Examiner rejects claims 1, 8, 15 under 35 U.S.C. § 112(a) for a lack of written description support.
Dependent claims 1-7, 9-14, and 16-21 inherit the deficiencies of their parent claim and are also rejected under 35 U.S.C. 112(a) as failing to comply with the written description requirement.
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 1-21 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea (i.e. a judicial exception) without significantly more.
Per step 1 of the 2019 Revised Patent Subject Matter Eligibility Guidance, the claims are directed towards a process, machine, or manufacture.
Per step 2A Prong One, the claims recite specific limitations which fall within at least one of the groupings of abstract ideas enumerated in the 2019 PEG, as follows:
Per Independent claims 1, 8, 15:
estimating, … based on the prediction model, predicted performance of each of the combinations associated with each of the plurality of ads;
generating, … based on the prediction model, combination distributions for each of the plurality of ads based on the predicted performance for each of the combinations for the ad, wherein each of the combination distributions is associated with a corresponding one of a plurality of traffic segments, and each of the plurality of traffic segments corresponds to one of a plurality of groups of users sharing one or more common characteristics that impact return of a displayed ad;
determining, based on a selection criteria configuration that maps each subset of the combination distributions to a respective one of a plurality of frontend ad serving engines each distributed in a respective one of a plurality of geographic jurisdictions and serving ads for products allocated to the respective geographic jurisdiction, different subsets of the combination distributions to be sent to different ones of the plurality of frontend ad serving engines;
As noted supra, these limitations fall within at least one of the groupings of abstract ideas enumerated in the 2019 PEG. Specifically, these limitations fall within the group Certain Methods Of Organizing Human Activity (e.g. fundamental economic principles or practices (including hedging, insurance, mitigating risk); commercial or legal interactions (including agreements in the form of contracts; legal obligations; advertising, marketing or sales activities or behaviors; business relations); managing personal behavior or relationships or interactions between people (including social activities, teaching, and following rules or instructions).
That is, these steps, as drafted, are a business decision to use the output of a generic machine-learning model as an “estimate” to predict advertisement performance (e.g. estimating probability of Click through rate or Conversion of an ad based on an undisclosed “prediction model”) and then generate (i.e. select) combinations of ads to load onto a server based on an undisclosed algorithm/relationship to this predicted performance for the purpose of either exploiting predicted good combinations or exploring a lesser understood combination, and thus falling into Certain Methods of Organizing Human Activity. At this high-level of generality, there is no technical problem being solved and no technical solution being presented – instead, only the very high-level general ideas are being claimed. Allusions to use of machine learning to explore and exploit ads is recited only at the highest level of generality. Furthermore, the mere nominal recitation of “a least one processor, a memory, and a communication platform”, although it generally ties the idea to a computing environment, does not take the claim limitation out of the enumerated grouping. Thus, the claims recite an abstract idea.
Per step 2A Prong 2, the Examiner finds that the judicial exception is not integrated into a practical application. Although there are additional elements, other than those noted supra, recited in the claims, none of these additional element(s) or a combination of elements as recited in the claims apply, rely on, or use the judicial exception in a manner that imposes a meaningful limit on the judicial exception, such that it is more than a drafting effort designed to monopolize the exception. As drafted, the claims as a whole merely describe how to generally “apply” the aforementioned concepts using generic computer components, “link” them to a field of use (i.e. in this case exploring and exploiting advertising combinations within generic confines of machine learning), or serve as insignificant extra-solution activity (e.g. data transfer such as between computers and servers). The claimed computer components are recited at a high level of generality and are merely invoked as tools to implement the idea but are not technical in nature. Simply implementing the abstract idea on or with generic computer components is not a practical application of the abstract idea.
These additional limitations (exemplified in the features of claim 1) are as follows:
PNG
media_image1.png
376
732
media_image1.png
Greyscale
PNG
media_image2.png
532
754
media_image2.png
Greyscale
However, these elements do not present a technical solution to a technical problem; i.e. Applicant’s invention is not a technique nor technical solution for “training” a prediction model when recited at this high-level of generality. Instead, at this high-level of generality, this feature is akin to data-collection – i.e. the feature is providing context which describes that some undisclosed “prediction model” is trained but no specific “training” is apparently necessary outside of what is already known in the art (e.g. see prior art of Kagian US 20180232660 A1 at [0063]-[0070] upon which applicant appears to have drawn this idea). Furthermore, applicant’s invention is not a technique of “sending” / “transmitting” data (regardless of its intended purpose, e.g. “to enable” functionality of an undisclosed “EEL”, explore exploit layer algorithm), nor is applicant’s invention a technique of “updating” training data, nor “updating” an undisclosed machine learning model with such data. Instead, these features merely serve to generally “apply” the aforementioned concepts, or link them to a field of use (e.g. within generic confines of machine learning), or are insignificant extra-solution activity as relates to the already identified abstract idea and do not integrate the abstract idea into a practical application thereof.
Per Step 2B, the Examiner does not find that the claims provide an inventive concept, i.e., the claims do not recite additional element(s) or a combination of elements that amount to significantly more than the judicial exception recited in the claim. As discussed with respect to Step 2A Prong Two, the additional elements in the independent claims were considered as merely serving to generally “apply” the aforementioned concepts via generically described computer components (e.g. by at least one processor, a memory, and a communication platform) and “link” them to a field of use (i.e. advertising performance estimation), or as insignificant extra-solution activity. For the same reason these elements are not sufficient to provide an inventive concept; i.e. the same analysis applies here in 2B. Mere instructions to apply an exception using a generic computer component and conventional data gathering cannot integrate a judicial exception into a practical application at Step 2A or provide an inventive concept in Step 2B. So, upon revaluating here in step 2B, these elements are determined to amount to no more than mere instructions to apply the exception using generic computer components and/or gather and transmit data which is well-understood, routine, conventional activity in the field; i.e. note the Symantec, TLI, and OIP Techs Court decisions cited in MPEP 2106.05(d)(ll) indicate that mere receipt or transmission of data over a network is a well-understood, routine, and conventional function when it is claimed in a merely generic manner (as it is here).
Accordingly, alone and in combination, these elements do not integrate the abstract idea into a practical application, as found supra, nor provide an inventive concept, and thus the claims are not patent eligible.
As for the dependent claims, the dependent claims do recite a combination of additional elements. However, these claims as a whole, considered either independently or in combination with the parent claims, do not integrate the identified abstract idea into a practical application thereof nor do they provide an inventive concept.
For example, dependent claims 2, 9, 16 recite the following: “…each of the plurality of attributes of each of the plurality of ads is associated with multiple assets, any of which can be used to render the attribute; and the criterion related to performance includes a click through rate (CTR) and a conversion rate (CVR)..” However, these are merely descriptions of abstract features of the already identified abstract idea and further describing a performance metric to be a Click-through rate or Conversion rate does not illuminate how the machine learning aspects of such estimation of these metrics is intended to be performed. Therefore, these limitations are part of the abstract idea but not significantly more.
Therefore, the Examiner does not find that these additional claim limitations integrate the abstract idea into a practical application nor provide an inventive concept. Instead, these limitations, as a whole and in combination with the already recited claim elements of the parent claims, are not significantly more than the already identified abstract idea. A similar finding is found for the remaining dependent claims.
For these reasons, the claims are not found to include additional elements that are sufficient to amount to significantly more than the judicial exception and are therefore patent ineligible.
Please see the 2019 Revised Patent Subject Matter Eligibility Guidance published in the Federal Register (84 FR 50) on January 7, 2019 (found at http://www.uspto.gov/patent/laws-and-regulations/examination-policy/examination-guidance-and-training-materials).
Claim Rejections - 35 USC § 103 (AIA )
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 of this title, 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 factual inquiries set forth in Graham v. John Deere Co., 383 U.S. 1, 148 USPQ 459 (1966), that are applied for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or non-obviousness.
Claims 1-21 are rejected under 35 U.S.C. 103 as obvious over Kagian et al. (U.S. 2018/0232660 A1; hereinafter, "Kagian") in view of non-Patent Literature of Aharon, M., et al. (2019). Carousel Ads Optimization in Yahoo Gemini Native. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining (hereinafter, NPL) and Ellis et al. (U.S. 2004/0015608 A1; hereinafter, “Ellis”).
Claims 1, 8, 15: (Currently amended)
Pertaining to claims 1, 8, 15 exemplified in the method step limitations of claim 1, Kagian as shown teaches the following:
A method implemented on at least one processor, a memory, and a communication platform for generating combination distributions for ads […], comprising:
training, by an ad recommendation backend server with respect to performance according to a criterion, a prediction model based on training data associated with a plurality of ads, each of which has a plurality of attributes (Kagian, see at least [0063]-[0070] teaching e.g.:
PNG
media_image3.png
302
436
media_image3.png
Greyscale
Also note per [0032]-[0035]: “…The pCTR and pCONV [prediction models] may be calculated using models that are periodically updated by an algorithm, e.g. OFFSET (One-pass Factorization of Feature Sets), which is a feature enhanced collaborative-filtering (CF) based event prediction algorithm [i.e. via machine learning] that updates its latent factor model for every new batch of logged data using stochastic gradient descent (SGD) [with respect to performance according to criterion]. OFFSET may be implemented on the grid using map-reduce architecture, where every new batch of logged data is preprocessed and parsed by many mappers and the ongoing update of a model is conducted as a centralized process on a single reducer. As many other learning algorithms [again, OFFSET is machine learning], OFFSET includes several hyper-parameters that can be tuned to provide best performance for given system conditions. The architecture of OFFSET makes it possible to do a parallel grid-search to find an optimal set of hyper-parameters (or configuration) and its resulting model, for boosting system performance. It usually takes a few days to train the system using a few weeks of logged data in order to get a mature model which can be pushed to production and start serving ads to users…”; see also at least Fig. 1, [0043] and [0063]-[0070], e.g. “…The adaptive model training engine 140 [ad recommendation backend server] may adaptively and continuously train an ad related model stored in the ad related model database 150 for advertisement selection at the web service provider 130…”), wherein the training data include combinations with past performance thereof for each of the plurality of ads (Kagian, see citations noted supra, e.g. again at least [0033]-[0034] in view of at least [0063]-[0070], teaching e.g.: “…To learn the model parameters ‘Theta’, the system minimizes the logistic loss (LogLoss) of the training data set:T (e.g., past impressions and clicks) using one-pass stochastic gradient descent (SGD) based algorithm…”) and each of the combinations includes a plurality of assets representing respectively [a] plurality of attributes of an ad (Kagian, again see citations noted supra at least [0063]-[0066] e.g.: “…Both ad and user vectors are constructed using their features, which enable dealing with the data sparsity issues (when native ad CTR is less than I%). For ads, one can use a simple summation between the vectors of the unique creative id, campaign id, and advertiser id (currently 3 feature vectors, all in dimension D). The combination between the different user feature vectors may be a bit more complex to allow non-linear dependencies between feature pairs….”);
estimating, by the ad recommendation backend server, based on the prediction model, predicted performance of each of the combinations associated with each of the plurality of ads (Kagian, see citations noted supra and again per at least [0063]-[0066], pCTR(u,a) is an estimated predicted performance of the combo of “u,a” e.g.:
PNG
media_image4.png
471
681
media_image4.png
Greyscale
);
generating, by the ad recommendation backend server based on the prediction model, combination distributions for each of the plurality of ads based on the predicted performance for each of the combinations for the ad, (Kagian, see at least Fig. 13, and [0042]-[0043] and [0111], teaching: “…In general, the models in the ad related model database 150 may be generated based on a training process, and used by an application (not shown) in the web service provider 130, at the backend of the web service provider 130…” and “…Based on the disclosed model [the prediction model], each web service provider 130 [ad recommendation backend server] may select which ads [combination distributions for each of the plurality of ads] to show a user in a certain context…”)
Although Kagian teaches the above limitations including e.g. per [0042] in view of [0052]: “…an adaptive model training engine 140 [ad recommendation backend server] may adaptively and continuously train [update] an ad related model stored in the ad related model database 150 for advertisement selection at the web service provider 130…”, etc… and has been shown to teach, e.g. per [0063]-[0070], estimating predicted performance of an advertisement, e.g. pCTR or pCONV via an OFFSET algorithm, and teaches e.g. per [0032]-[0033] his ads may be Gemini native ads, each of which ad has its own latent factor vector of assets, and per [0111] teaches: “In practice, there may be an ad serving system, e.g. a Gemini native serving platform [frontend ad serving engine] that serves ads across many web service providers.”, and the Examiner notes that it is understood in the art that such Gemini native “carousel” ads, include several assets (each of which include variations [attributes], e.g. variations of title, variations of images, and also variations of descriptions) and several slots to which the different assets may be rendered into, Kagian may not explicitly teach the below nuance. However, Kagian in view of NPL teaches the following:
wherein each of the combination distributions is associated with a corresponding one of a plurality of traffic segments, and each of the plurality of traffic segments corresponds to one of a plurality of groups of users sharing one or more common characteristics that may impact return of a displayed ad (NPL, see at least “7 Concluding Remarks”, teaching: e.g. “Another direction may be to perform CASE on smaller traffic segments [group of users] composed of ad cross user features (e.g., ad x gender or, ad x device) [sharing one or more common characteristics that may impact return of displayed ad]. This may improve personalization but may require longer exploration periods due to sparsity issues…”; Applicant’s own specification at [0069] and [0098] read on these teachings; e.g. segmentation by a common “gender” characteristic is expected to impact return according to Applicant’s own specification at [0069]. Furthermore, Specification at [0098] notes that segments may be defined as “gender and device”.);
Therefore, the Examiner understands that the limitations in question are merely applying known techniques of NPL (directed towards a successive elimination based approach, e.g. the CASE algorithm [EEL layer], for ranking assets such as Gemini Native ads and performing CASE on smaller traffic segments [group of users] composed of ad cross user features (e.g., ad x gender or, ad x device) [sharing one or more common characteristics that may impact return of displayed ad], e.g. which may improve personalization) which are applicable to a known base device/method of Kagian (who already teaches a system/method directed towards predicting CTR, click through rate, for Gemini native ads, such as Gemini native carousel ads) to yield predictable results. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply the techniques of NPL to the device/method of Kagian such that Kagian’s adaptive model training engine 140 [ad recommendation backend server], which Kagian teaches adaptively and continuously trains [updates] an ad related model stored in the ad related model database 150 for advertisement selection at the web service provider 130, may also perform the techniques taught by NPL because Kagian and NPL are analogous art in the same field of endeavor (at least G06Q30/02) and because according to MPEP 2143(I) (C) and/or (D), the use of known technique to improve a known device, methods, or products in the same way (or which is ready for improvement) is obvious.
Although the combination of the teachings of Kagian/NPL teaches the above limitations and Kagian, e.g. per [0032] teaches: “…A native marketplace, e.g. the Gemini native marketplace, can serve users with ads that are rendered to resemble the surrounding native content...” and per [0111], as shown teaches: “…In practice, there may be an ad serving system, e.g. a Gemini native serving platform that serves ads [i.e. via front end ad serving engines] across many web service providers… each web service provider may select [selection criteria] which ads to show a user in a certain context…”, which Examiner finds suggests a need for ad servers which are each dedicated to serve such ads related to a web service providers’ specific selection of “certain contexts” and therefore this also suggests a need to determine, based on a mapping of such “certain context” [a selection criteria] each subset of a combination distribution of Kagian’s Gemini native ads to one of these dedicated ad servers [front end ad serving engine], Kagian may not explicitly teach this nuance as recited below. However, regarding the below recited features, Kagian/NPL in view of Ellis teaches the following:
determining, based on a selection criteria configuration that maps each subset of the combination distributions to a respective one of a plurality of frontend ad serving engines each distributed in a respective one of a plurality of geographic jurisdictions and serving ads for products allocated to the respective geographic jurisdictions different subsets of the combination distributions to be sent to different ones of the plurality of frontend ad serving engines (Ellis, see at least Figs. 1B, 4, 5, and 7, [0050]-[0055], [0069], and [0079]-[0090] teaching the determination of a subset of ads which are to be sent to various dedicated ad servers, and then to frontend ad engines such as Dynamic Ad Inserter 421 of Fig. 4, based on selection criteria such as “type of video game”, “genre”, “timing considerations, etc…” to which an ad is intended to be served, each of which is a type of mapping of a subset of ads to each dynamic inserter; e.g.: “…When scheduling [determining] which advertisement [different subsets of the combination distributions] to send [to be sent] to the game client system [different ones of the plurality of frontend ad serving engines], the ad scheduler reviews which ads are available to be inserted to the game with respect to [based on a selection criteria] ad type, genre, timing considerations, etc. Because these requirements may be continuously changing, embodiments of the ad scheduler are capable of operating in a mode where all of the limitations are checked prior to sending an ad to the game client system…” and as noted per [0069]: “…because all types of ads may not be suitable for all game users, embodiments of the dynamic inserter allow the game client system to specify certain genres, or classes of ads that are not acceptable to that game client. Similar to the list of types described above, the list of all of the genres in a game could be sent in step 601, or the individual genre matching an ad tag could be sent as the ad tags are individually found in the game. An example genre of ads may be beer or other alcohol type ads, which should not be targeted to those too young to purchase such products. In that case, the game client system 106 would indicate to the ad server that such a genre of advertisements is unacceptable; and such ads would not be inserted into the game. (Note, as above, that such control can take place at the server or the client side of the communication.)…”; Examiner notes that applicant’s description each distributed in a respective one of a plurality of geographic jurisdictions and serving ads for products allocated to the respective geographic jurisdictions is non-functional descriptive material which does not alter the determining step nor the selection criteria which only is recited as “maps each subset of the combination distributions to a respective one of a plurality of frontend ad serving engines”; as long as determination is made based on selection criteria configuration that maps each subset to a respective one of the plurality of frontend ad serving engines, it is immaterial where such engines are distributed, etc…
PNG
media_image5.png
596
434
media_image5.png
Greyscale
);
Therefore, the Examiner understands that the limitation in question is merely applying a known technique of Ellis (directed towards a technique of determining which advertisements [different subsets of the combination distributions] to send [to be sent] to the game client system [different ones of the plurality of frontend ad serving engines], where the ad scheduler makes such determination based on a review of which ads are available to be inserted to the game with respect to [based on a selection criteria] ad type, genre, timing considerations, etc.) which is applicable to a known base device/method of Kagian/NPL (already directed towards a system/method by which an ad serving system, e.g. a Gemini native serving platform that serves ads [subset of the combination distributions] across many web service providers, where each web service provider may select which ads to show a user in a certain context) to yield predictable results. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply the technique of Ellis to the device/method of Kagian/NPL in order to perform the limitation in question such that the ads which Kagian’s web service providers select to show a user in a certain context, e.g. Ellis’ video game context, are also selected per Ellis’ technique and sent to the game client system based on selection criteria which maps each ad to a particular game being played by the client and because Kagian/NPL and Ellis are analogous art in the same field of endeavor (at least G06Q30/02) and because according to MPEP 2143(I) (C) and/or (D), the use of known technique to improve a known device, methods, or products in the same way (or which is ready for improvement) is obvious.
Furthermore, as already shown supra, Kagian/Ellis has been shown to teach the aforementioned step of “determining” and applicant’s “the ad recommendation backend server”. Furthermore, NPL teaches the following:
sending […] each of the subsets of the combination distributions for at least some of the plurality of ads to an explore/exploit layer (EEL) at a respective one of the plurality of frontend ad serving engines; (NPL, see at least “Section 3, Related Work”, in view of “Sections 5.3 CASE, and 5.5 Why CASE?”, teaching e.g.: “…Explore-exploit techniques are widely used to optimize web content. For example, [Agarwal]1 describes an explore-exploit based system for selecting news items for Yahoo Homepage. There are also many commercial tools that conduct various ad related optimizations on social platforms…” and “CASE algorithm [explore/exploit layer (EEL)]…On average the model includes 86.7% carousels in exploration state (or unresolved carousels), 10.9% in exploitation state (or resolved carousels), and 2.5% in idle state10 . We note that while only 10.9% of the carousels are in exploitation state, their traffic consists of 92.2% of the total CAO carousel traffic. Out of the exploring carousels, 14.0% have at least one eliminated asset (partially resolved carousels). Out of the exploiting carousels, 62.5% have declared a "best''2 asset different than the one denoted by the advertiser. Assuming our algorithm provides ground truth, advertisers identify their carousels' "best" assets in advance, only 37.5% of the time…”; CASE means Carousel Assets Successive Elimination Algorithm)
Therefore, the Examiner understands that the limitation sending, by the ad recommendation backend server based on the determining each of the subsets of the combination distributions for at least some of the plurality of ads to an explore/exploit layer (EEL) at a respective one of the plurality of frontend ad serving engines is merely applying a known techniques of NPL (directed towards a sending ads to a CASE algorithm [EEL layer], for ranking assets according to their click through rate (CTR), and gradually eliminating assets with lower CTR than that of the “best” asset, and exploring the “surviving” assets with uniform distribution until only one remains and then rendering the carousel ads accordingly, placing higher CTR assets in more conspicuous slots) which are applicable to a known base device/method of Kagian/Ellis (who already teaches a system/method directed towards determining different subsets of the combination distributions of ads to be sent to different ones of the plurality of frontend ad serving engines) to yield predictable results. Therefore, it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply the techniques of NPL to the device/method of Kagian/Ellis such that Kagian/Ellis’ adaptive model training engine 140 [ad recommendation backend server], may send, based on the determining, each of the subsets of the combination distributions for at least some of the plurality of ads to an explore/exploit layer (EEL) at a respective one of the plurality of frontend ad serving engines because Kagian/Ellis and NPL are analogous art in the same field of endeavor (at least G06Q30/02) and because according to MPEP 2143(I) (C) and/or (D), the use of known technique to improve a known device, methods, or products in the same way (or which is ready for improvement) is obvious.
Furthermore, Kagian is found to teach the following:
updating, by the ad recommendation backend server, the training data based on the first user's action on an auction winning ad, rendered on a webpage viewed by the first user on a user device, the auction winning ad being rendered using a combination from combination distributions for the auction winning ad, the combination being drawn by one of the plurality of frontend ad serving engines directly from an EEL at the frontend ad serving engine in response to a determination that the combination distributions for the auction winning ad are available in the EEL (Note the clause which is not in bold type face is interpreted as non-functional descriptive material; i.e. applicant’s step of “updating” appears to be agnostic regarding ads except that the updating must occur in response to user action on auction winning ad. Therefore, this description of the auction winning ad does not affect nor limit the step of “updating”. Kagian, teaches at least the updating step as recited. See again citations noted supra in view of at least Kagian at [0043], teaching e.g.: “…The adaptive model training engine 140 [ad recommendation backend server] may adaptively and continuously train [update] an ad related model stored in the ad related model database 150 for advertisement selection at the web service provider 130. With fresh [updated training data based on user’s action] logged ad related data, the adaptive model training engine 140 can make use of latest trained model to tune the hyper parameters in the model to make it adapted to temporal changes...”; note that hyper parameters, i.e. model parameters, include user’s actions on rendered ad which won auction, such as a user’s “click” on an such an ad, etc… e.g. see [0063]-[0067]);
updating, by the ad recommendation backend server the prediction model based on the updated training data (Kagian, see again citations noted supra in view of at least Figs. 5B and Fig. 7 and [0054]-[0055] and [0061]-[0064], e.g.: “…the hyper parameter tuner 560 may generate a model parameter update to update the ad related model 505 based on newly tuned parameters. The predetermined condition may be related to a level of convergence of the model training…”).
Claims 2, 9, 16: (Original)
Kagian/NPL/Ellis teaches the limitations upon which these claims depend. Furthermore, Kagian in view of NPL teaches the following:
The method of claim 1, wherein each of the plurality of attributes of each of the plurality of ads is associated with multiple assets, any of which can be used to render the attribute (Kagian, see at least [0002] regarding “Technical Background” teaching advertisements of various types, including “Native ads” which are “…a type of advertising that matches the form and function of the platform upon which it appears. For example, the Gemini native marketplace serves users with ads that are rendered to resemble the surrounding native content...”; Applicant’s “attributes3 of each of the plurality of ads is associated with multiple assets” reads on the “assets” of a Gemini native “carousel” advertisement; i.e. Examiner notes, although Gemini native “carousel” ads are perhaps not explicitly expounded upon by Kagian, he nonetheless notes it is Gemini native ads which are contemplated and it was within the knowledge of a person of ordinary skill in the art before the effective filing date of the claimed invention, e.g. as noted per NPL, “Section 2.4 native Carousel Ads”, discussing that the native ad marketplace has also introduced a variant of the "traditional" native ads which are referred to as “carousel ads”. A Gemini native carousel ad includes several assets (which are native-like ads with a title [asset], image [asset] and also a description [asset] in certain cases) and several slots to which the different assets may be rendered into. Therefore, whether explicitly stated by Kagian or not, the Examiner finds that it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to have use Gemini native carousel ads which have a plurality of attributes which are associated with multiple assets, such as variations of images, variations of titles, and variations of descriptions, any of which can be used to render the title attribute, or the image attribute, or the description attribute because per MPEP 2143(I) (G) Some teaching, suggestion, or motivation in the prior art that would have led one of ordinary skill to modify the prior art reference to arrive at the claimed invention is obvious. The motivation may be implicit and may be found in the knowledge of one of ordinary skill in the art, or, in some cases, from the nature of the problem to be solved. Id. at 1366, 80 USPQ2d at 1649.); and
the criterion related to performance includes a click through rate (CTR) and a conversion rate (CVR) (Kagian, again per at least [0033]-[0034]: “…The pCTR and pCONV may be calculated using models that are periodically updated by an algorithm, e.g. OFFSET…”).
Claims 3, 10, 17: (previously presented)
Kagian/NPL/Ellis teaches the limitations upon which these claims depend. Furthermore, Kagian teaches the following:
… wherein each of the combinations in the training data associated with an ad in the plurality of ads further includes information about a corresponding display environment in which the combination is used to render the ad, and the information about the display environment indicates at least one of the first user to whom the ad is rendered and a webpage in which the ad is rendered using the combination (Kagian, see at least [0042]-[0044], in view of [0107] e.g.: “The ad request analyzer 1310 in this example may receive and analyze an ad request from a user. Based on the ad request, the ad request analyzer 1310 may retrieve a user profile of the user from the user profile database 1315 and determine personal information of the user and ad related information with respect to the request. For example, the personal information may include user ID [indicates at least one of a user to whom the ad is rendered], user location, and demographic information of the user; the ad related information may include …environment with respect to the ad request.”; where per [0043]-[0044], such “fresh logged ad related data” is used to train the ad selection models.)
Claims 4, 11, 18: (Original)
Kagian/NPL/Ellis teaches the limitations upon which these claims depend. Furthermore, Kagian in view of NPL teaches the following:
… wherein the plurality of attributes of each of the plurality of ads include at least some of: a title of the ad[,] an image that visually conveys information about the ad[,] and a description that textually summarizes content of the ad, wherein each of the combinations includes a plurality of assets, each of which instantiates one corresponding one of the plurality of attributes of the ad, for rendering the plurality of attributes of the ad (NPL, see at least Figs. 1 and 2, associated disclosures, and at least “Section 2.4 Native Carousel Ads”: “…the native ad marketplace introduced a variant of the "traditional" native ads referred to as carousel ads. A Gemini native carousel ad includes several assets (which are native-like ads with a title, image and also a description in certain cases) and several slots to which the different assets may be rendered into (see Figure 2 for a schematic example of Gemini native carousel ad on mobile and desktop devices).
PNG
media_image6.png
206
600
media_image6.png
Greyscale
).
Therefore, the Examiner understands that the limitations in question are merely applying known techniques of NPL (directed towards Gemini native carousel ads which each include a plurality of attributes such as a title of the ad, etc…) which are applicable to a known base device/method of Kagian (who already teaches a system/method directed towards predicting CTR for Gemini native ads), to yield predictable results. Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply the techniques of NPL, and use Gemini carousel ads, etc…, to the device/method of Kagian because Kagian and NPL are analogous art in the same field of endeavor (at least G06Q30/02) and because according to MPEP 2143(I) (C) and/or (D), the use of known technique to improve a known device, methods, or products in the same way (or which is ready for improvement) is obvious.
Claims 5, 12, 19: (Original)
Kagian/NPL/Ellis teaches the limitations upon which these claims depend. Furthermore, Kagian in view of NPL teaches the following:
… wherein the step of generating combination distributions for each of the plurality of ads is via a combination successive elimination (CSE) process, which comprises:
identify a list of combinations associated with the ad; initializing a combination distribution over the list of combinations using a uniform distribution; generating a surviving set of combinations; identifying a best combination in the surviving set that has a best predicted performance; removing each combination in the surviving set that has a performance below the best predicted performance to generate an updated surviving set; updating the combination distribution based on the updated surviving set; repeating the steps of identifying, removing, and updating until a predetermined criterion is satisfied; and allocating probability mass with respect to combinations in the updated surviving set based on their respective predicted performance (NPL, see at least Abstract and Introduction, teaching, e.g.: “…The asset distribution, calculated for each carousel, is updated every training period by our carousel assets successive elimination (CASE) algorithm. Roughly speaking, we start with a uniform distribution, measure the CTR of each asset on slot 1 (the most conspicuous slot), and "eliminate" all assets that have lower CTR than the best asset with a predefined confidence probability (e.g., 90%). We continue this until one asset remains and then declare this carousel as "resolved". The CASE algorithm is a practical and robust version of the celebrated successive elimination approach…”; See also “Sections 5.3 CASE, and 5.5 Why CASE?”).
Therefore, the Examiner understands that the limitations in question are merely applying known techniques of NPL (directed towards a successive elimination based approach, e.g. the CASE algorithm to rank and select Gemini native ads based measurements of the CTR of each asset) which are applicable to a known base device/method of Kagian (who already teaches a system/method directed towards predicting CTR for Gemini native ads with assets, such as Gemini native carousel ads) to yield predictable results. Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply the techniques of NPL to the device/method of Kagian because Kagian and NPL are analogous art in the same field of endeavor (at least G06Q30/02) and because according to MPEP 2143(I) (C) and/or (D), the use of known technique to improve a known device, methods, or products in the same way (or which is ready for improvement) is obvious.
Claims 6, 13, 20: (Original)
Kagian/NPL/Ellis teaches the limitations upon which these claims depend. Furthermore, Kagian in view of NPL teaches the following:
…wherein the combination distributions associated with each of the plurality ads provide different ways to render the ad, wherein each combination in the combination distributions[:]
represents one way to render the ad with respect to a display environment; (NPL, see at least “Section 2.4: Native Carousel Ads” which include multiple assets, each representing variations of attributes, e.g. variations in “title”, or variations in “image” or “description”, etc…; and see disclosure of CASE algorithm, e.g. Section 5); is provided with an indication of a likelihood to be drawn to render the ad the indication of the likelihood is determined based on the predicted performance of the combination in the display environment estimated by the prediction model (NPL, see at least “Section 4.1”: “popularity trends also causes ad (and assets within ads) CTR variations...”, etc…; and Section 5)
Therefore, the Examiner understands that the limitations in question are merely applying known techniques of NPL (directed towards application of a CASE algorithm on Gemini native carousel ads which each include a plurality of attributes such as a title of the ad, etc…) which are applicable to a known base device/method of Kagian (who already teaches a system/method directed towards predicting CTR for Gemini native ads), to yield predictable results. Therefore it would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to apply the techniques of NPL, to the device/method of Kagian because Kagian and NPL are analogous art in the same field of endeavor (at least G06Q30/02) and because according to MPEP 2143(I) (C) and/or (D), the use of known technique to improve a known device, methods, or products in the same way (or which is ready for improvement) is obvious.
Claims 7, 14: (previously presented)
Kagian/NPL/Ellis teaches the limitations upon which these claims depend including the prediction model and predications of performance, e.g. pCTR, for combinations distributions for each ad. Furthermore, regarding updating of these aforementioned features, Kagian teaches the following:
…further comprising: retraining the prediction model when updated training data having an updated set of combinations are available to obtain updated prediction model; providing updated predicted performance based on the updated prediction model for each of the combination in the updated set of combinations in the updated training data; updating the combination distributions for each of the plurality of ads based on updated predicted performance for each of the combinations for the ad to generate updated combination distributions; sending the updated combination distributions for at least some of the plurality of ads to the EEL at the frontend ad serving engine so that the combination for the auction winning ad can be drawn from the updated combination distributions for rendering (Kagian, see at least Abstract, Figs. 3b and 12, in view of at least [0033] and [0052], e.g.: “The present teaching relates to generating an updated model related to advertisement selection. In one example, a request is obtained for updating a model to be utilized for selecting an advertisement… The model is pre-selected based on a performance metric related to advertisement selection… The steps of generating, creating, and selecting are repeated until a predetermined condition is met. The model is updated with the latest selected candidate model when the predetermined condition is met…” Therefore, the Examiner finds that in view of the combination of teachings of Kagian/NPL it would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to have performed the various updating of already claimed steps, shown to be obvious in view of Kagian/NPL as noted supra, because Kagian explicitly teaches updating his models and repeating until predetermined conditions are met and because per MPEP 2143(I) (C) Use of known technique to improve similar devices (methods, or products) in the same way is obvious; Furthermore, the Examiner notes that rigid preventative rules that deny factfinders recourse to common sense are neither necessary under our case law nor consistent with it. KSR Int'l Co. v. Teleflex, Inc., No 04-1350 (U.S. Apr. 30, 2007).)
Claim 21: (previously presented)
Kagian/NPL teaches the limitations upon which these claims depend. Furthermore, Kagian teaches the following:
The method of claim 1, wherein the one or more common characteristics that impact return of a displayed ad comprise one or more of a common profession, a common interest, a common age range, and a common geographical region shared by one of the plurality of groups of users (NPL, see at least “7 Concluding Remarks”, teaching: e.g. “Another direction may be to perform CASE on smaller traffic segments composed of ad cross user features [group of users sharing one or more common characteristics] (e.g., ad x gender or, ad x device [an interest]). This may improve personalization but may require longer exploration periods due to sparsity issues…”; device which a user uses represents an interest of the user; e.g. user is interested in Android device instead of iPhone by virtue of user using such device instead of an iPhone).
Response to Arguments
Applicant amended Claims 1, 8, 15 on 03/20/2026. Applicant's arguments (hereinafter “Remarks”) also filed 03/20/2026, have been fully considered but are moot in view of the new grounds of rejection necessitated by applicant’s amendments. Note the new 35 USC 112 and 103 rejection with updated citations to Kagian, Ellis, and NPL teaching applicant’s amended and argued features.
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).
The following prior art is made of record although not relied upon as it is considered pertinent to applicant's disclosure:
US 10,438,114 B1 to Blundell et al. Blundell teaches a system/method where advertisements to be served with Internet resources can be effectively selected and media content can be effectively recommended by use of Bayesian neural network layers. Blundell teaches techniques regarding training a neural network to learn approximations of probability distributions for Bayesian neural network layers rather than exact probability distributions. Furthermore, he notes exploration/exploitation trade-off may be adjusted automatically to approximately match the inherent uncertainty in advertisement data. At each training step, the neural network can be trained on a respective mini-batch of data to adjust the approximations of the probability distributions, resulting in faster training and more accurate action selections.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MICHAEL J SITTNER whose telephone number is (571)270-3984. The examiner can normally be reached M-F; ~9:30-6:30. 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, Waseem Ashraf can be reached on (571) 270-3948. 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.
/Michael J Sittner/
Primary Examiner, Art Unit 3621
1 Deepak Agarwal, Bee-Chung Chen, Pradheep Elango, Nitin Motgi, Seung-Taek Park, Raghu Ramakrishnan, Scott Roy, and Joe Zachariah. Online models for content optimization. In Advances in Neural Information Processing System; pages 17-24, 2009.
2 The asset with highest slot 1 CTR.
3 Drawings: Fig. 3A; and Specification [0016]: “Fig. 3A shows an example ad with multiple attributes, each of which has a set of assets that can be used to assemble the ad, in accordance with an embodiment of the present teaching”
Prosecution Insights