Prosecution Insights
Last updated: November 04, 2026
Application No. 19/185,390

GENERATING CONTENT USING A GENERATIVE MODEL WITHOUT RELYING ON SELECTED TRAINING EXAMPLES

Non-Final OA §103§DOUBLEPATENT
Filed
Apr 22, 2025
Priority
Feb 10, 2023 — provisional 63/444,805 +4 more
Examiner
CRESPO FEBLES, HECTOR J
Art Unit
2657
Tech Center
2600 — Communications
Assignee
Bria Artificial Intelligence Ltd.
OA Round
1 (Non-Final)
100%
Grant Probability
Favorable
1-2
OA Rounds
7m
Est. Remaining
99%
With Interview

Examiner Intelligence

Grants 100% — above average
100%
Career Allowance Rate
1 granted / 1 resolved
+38.0% vs TC avg
Minimal +0% lift
Without
With
+0.0%
Interview Lift
resolved cases with interview
Fast prosecutor
2y 1m
Avg Prosecution
15 currently pending
Career history
25
Total Applications
across all art units

Statute-Specific Performance

§101
9.1%
-30.9% vs TC avg
§103
68.2%
+28.2% vs TC avg
§102
9.1%
-30.9% vs TC avg
§112
6.8%
-33.2% vs TC avg
Black line = Tech Center average estimate • Based on career data from 1 resolved cases

Office Action

§103 §DOUBLEPATENT
DETAILED ACTION Notice of Pre-AIA or AIA Status The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . Double Patenting The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969). A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b). The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13. The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/patents/apply/applying-online/eterminal-disclaimer. Claims 1-20 are rejected on the ground of nonstatutory double patenting as being unpatentable over claims 1-20 of U.S. Patent No. 12,223,281. Although the claims at issue are not identical, they are not patentably distinct from each other because the claims are obvious variations of each other. Current application U.S. Patent No. 12,223,281 1. A non-transitory computer readable medium storing a software program comprising data and computer implementable instructions that when executed by at least one processor cause the at least one processor to perform operations for generating content using a generative model without relying on selected training examples, the operations comprising: receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using a plurality of training examples, each training example of the plurality of training examples is associated with a respective content; obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples; based on the indication, using the input and the generative model to generate the new content, abstaining from basing the generation of the new content on any training example included in the particular subgroup; and providing the new content. 1. A non-transitory computer readable medium storing a software program comprising data and computer implementable instructions that when executed by at least one processor cause the at least one processor to perform operations for generating content using a generative model, the operations comprising receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using all training examples of a plurality of training examples, each training example of the plurality of training examples is associated with a respective content, the generative model includes a plurality of artificial neurons; obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples; identifying a specific subgroup of the plurality of artificial neurons associated with the particular subgroup of at least one but not all of the plurality of training examples; based on the indication, using the input and the generative model while disregarding the artificial neurons included in the specific subgroup to generate the new content, abstaining, after the training of the machine learning model is completed, from basing the generation of the new content on any training example included in the particular subgroup; and providing the new content. 19. A system for generating content using a generative model without relying on selected training examples, the system comprising: at least one processor configured to perform the operations of: receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using a plurality of training examples, each training example of the plurality of training examples is associated with a respective content; obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples; based on the indication, using the input and the generative model to generate the new content, abstaining from basing the generation of the new content on any training example included in the particular subgroup; and providing the new content. 19. A system for generating content using a generative model without relying on selected training examples, the system comprising: at least one processor configured to perform the operations of: receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using all training examples of a plurality of training examples, each training example of the plurality of training examples is associated with a respective content, the generative model includes a plurality of artificial neurons; obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples; identifying a specific subgroup of the plurality of artificial neurons associated with the particular subgroup of at least one but not all of the plurality of training examples; based on the indication, using the input and the generative model while disregarding the artificial neurons included in the specific subgroup to generate the new content, abstaining, after the training of the machine learning model is completed, from basing the generation of the new content on any training example included in the particular subgroup; and providing the new content. 20. A method for generating content using a generative model without relying on selected training examples, the method comprising: receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using a plurality of training examples, each training example of the plurality of training examples is associated with a respective content; obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples; based on the indication, using the input and the generative model to generate the new content, abstaining from basing the generation of the new content on any training example included in the particular subgroup; and providing the new content. 20. A method for generating content using a generative model without relying on selected training examples, the method comprising: receiving an input indicative of a desire to generate a new content using a generative model, the generative model is a result of training a machine learning model using all training examples of a plurality of training examples, each training example of the plurality of training examples is associated with a respective content, the generative model includes a plurality of artificial neurons; obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples; identifying a specific subgroup of the plurality of artificial neurons associated with the particular subgroup of at least one but not all of the plurality of training examples; based on the indication, using the input and the generative model while disregarding the artificial neurons included in the specific subgroup to generate the new content, abstaining, after the training of the machine learning model is completed, from basing the generation of the new content on any training example included in the particular subgroup; and providing the new content. It is clear that all the elements of the application claims 1, 19, and 20 are to be found in patent claims 1, 19, and 20 of U.S. Patent No. 12,223,281, as the application claims 1, 19, and 20 are completely taught by patent claims 1, 19, and 20 of U.S. Patent No. 12,223,281. The difference between the application claims 1, 19, and 20 and the patent claims 1, 19, and 20 of U.S. Patent No. 12,223,281 lies in the fact that the patent claims includes more elements and is thus more specific. Thus, the invention of claims 1, 19, and 20 of U.S. Patent No. 12,223,281 is in effect a "species" of the "generic" invention of the application claims 1, 19, and 20 of the current application. It has been held that the generic invention is "anticipated" by the "species". See In re Goodman, 29 USPQ2d 2010 (Fed. Cir. 1993). Furthermore: Claim 2 of the current application corresponds to claim 2 of U.S. Patent No. 12,223,281. Claim 3 of the current application corresponds to claim 3 of U.S. Patent No. 12,223,281. Claim 4 of the current application corresponds to claim 4 of U.S. Patent No. 12,223,281. Claim 5 of the current application corresponds to claim 5 of U.S. Patent No. 12,223,281. Claim 6 of the current application corresponds to claim 6 of U.S. Patent No. 12,223,281. Claim 7 of the current application corresponds to claim 7 of U.S. Patent No. 12,223,281. Claim 8 of the current application corresponds to claim 8 of U.S. Patent No. 12,223,281. Claim 9 of the current application corresponds to claim 9 of U.S. Patent No. 12,223,281. Claim 10 of the current application corresponds to claim 10 of U.S. Patent No. 12,223,281. Claim 11 of the current application corresponds to claim 11 of U.S. Patent No. 12,223,281. Claim 12 of the current application corresponds to claim 12 of U.S. Patent No. 12,223,281. Claim 13 of the current application corresponds to claim 13 of U.S. Patent No. 12,223,281. Claim 14 of the current application corresponds to claim 14 of U.S. Patent No. 12,223,281. Claim 15 of the current application corresponds to claim 15 of U.S. Patent No. 12,223,281. Claim 16 of the current application corresponds to claim 16 of U.S. Patent No. 12,223,281. Claim 17 of the current application corresponds to claim 17 of U.S. Patent No. 12,223,281. Claim 18 of the current application corresponds to claim 18 of U.S. Patent No. 12,223,281. Claim Rejections - 35 USC § 103 In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status. The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action: A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made. The factual inquiries 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 nonobviousness. 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 as of the effective filing date of the claimed invention(s) 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 as of the effective filing date of the later invention 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(s) 1-5 and 14-18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Zafar; Muhammad Bilal et al. (US 20210209507 A1), hereinafter ZAFAR, in view of Ciminelli; Nikolas Louis et al. (US 20230409298 A1), hereinafter CIMINELLI, in view of INAKOSHI; Hiroya et al. (US 20200160119 A1), hereinafter INAKOSHI. Regarding claim 1, ZAFAR teaches: A non-transitory computer readable medium storing a software program comprising data and computer implementable instructions that when executed by at least one processor cause the at least one processor to perform operations for generating content using a generative model without relying on selected training examples, the operations comprising: ZAFAR [Abstract] “A system for processing a model. The model provides a model output given an input instance. The model has been trained on a training dataset by iteratively optimizing an objective function including losses according to a loss function for training instances of the training dataset. Upon receiving a removal request message identifying one or more undesired training instances of the training dataset, the model is made independent from the one or more undesired training instances. To this end, the one or more undesired training instances are removed from the training dataset to obtain a remainder dataset, and an adapted model is determined for the remainder dataset. The parameters of the adapted model are first initialized based on the set of parameters of the trained model, and then iteratively adapted by optimizing the objective function with respect to the remainder dataset.” ZAFAR [0056] “In general, the system 100 may be embodied as, or in, a single device or apparatus, such as a workstation, e.g., laptop or desktop-based, or a server. The device or apparatus may comprise one or more microprocessors which execute appropriate software. For example, the processor subsystem may be embodied by a single Central Processing Unit (CPU), but also by a combination or system of such CPUs and/or other types of processing units. The software may have been downloaded and/or stored in a corresponding memory, e.g., a volatile memory such as RAM or a non-volatile memory such as Flash. Alternatively, the functional units of the system, e.g., the data interface and the processor subsystem, may be implemented in the device or apparatus in the form of programmable logic, e.g., as a Field-Programmable Gate Array (FPGA) and/or a Graphics Processing Unit (GPU). In general, each functional unit of the system may be implemented in the form of a circuit. It is noted that the system 100 may also be implemented in a distributed manner, e.g., involving different devices or apparatuses, such as distributed servers, e.g., in the form of cloud computing.” ZAFAR [0097] “The method(s) may be implemented on a computer as a computer implemented method, as dedicated hardware, or as a combination of both. As also illustrated in FIG. 5, instructions for the computer, e.g., executable code, may be stored on a computer readable medium 500, e.g., in the form of a series 510 of machine-readable physical marks and/or as a series of elements having different electrical, e.g., magnetic, or optical properties or values. The executable code may be stored in a transitory or non-transitory manner. Examples of computer readable mediums include memory devices, optical storage devices, integrated circuits, servers, online software, etc. FIG. 5 shows an optical disc 500.” obtaining an indication of a particular subgroup of at least one but not all of the plurality of training examples; ZAFAR [Abstract] “A system for processing a model. The model provides a model output given an input instance. The model has been trained on a training dataset by iteratively optimizing an objective function including losses according to a loss function for training instances of the training dataset. Upon receiving a removal request message identifying one or more undesired training instances of the training dataset, the model is made independent from the one or more undesired training instances. To this end, the one or more undesired training instances are removed from the training dataset to obtain a remainder dataset, and an adapted model is determined for the remainder dataset. The parameters of the adapted model are first initialized based on the set of parameters of the trained model, and then iteratively adapted by optimizing the objective function with respect to the remainder dataset.” ZAFAR [0051] ”System 100 may also comprise a removal request interface 160 configured for receiving a removal request message 124. The removal request message 124 may identify one or more undesired training instances of the training dataset. Removal request interface 160 may internally communicate with processor subsystem 140 via data communication 123. Removal request interface 160 may be arranged for direct communication with other systems from which removal request messages may be received, e.g., user devices, e.g., using USB, IEEE 1394, or similar interfaces. Removal request interface 160 may also communicate over a computer network, for example, a wireless personal area network, an internet, an intranet, a LAN, a WLAN, etc. ...” ZAFAR [0012] “... If relatively few training instances are removed, e.g., a single instance, at most ten, at most fifty, or at most 1% or at most 5% of the training dataset, then the objective function for the remainder dataset may mostly comprise the same losses as the objective function for the original dataset. Accordingly, an optimum of the objective function for the remainder dataset may be expected to be relatively close to an optimum of the objective function for the original training dataset. As a consequence, relatively few iterations may suffice to adapt the set of parameters. Such iterative adapting of the parameters of the adapted model is referred to throughout as “iterative detraining” with respect to the undesired training instances.” ZAFAR does not teach but CIMINELLI teaches: receiving an input indicative of a desire to generate a new content using a generative model, CIMINELLI [0075] “In some examples, step 404 may comprise analyzing a textual input (such as textual input, the textual input received by step 402, the textual input received by step 802, textual data 102, etc.) and/or a sketch (such as the sketch received by step 402, the sketch received by step 802, sketch data 104, etc.) to obtain user interface elements (for example, a single user interface element, at least two user interface elements, at least ten user interface elements, and so forth). In one example, step 404 may comprise analyzing the textual input and/or the sketch to select the user interface elements of a group of alternative user interface elements. In one example, step 404 may comprise analyzing the textual input and/or the sketch to generate the user interface elements, for example using a generative model. In one example, step 404 may comprise using a machine learning model (such as a unimodal machine learning model, a multimodal machine learning model, a generative model, etc.) to analyze the textual input and/or the sketch and/or additional information and obtain the at least two user interface elements. The machine learning model may be a machine learning model trained using training examples to generate user interface elements and/or to select user interface elements of a plurality of alternatives based on textual information and/or visual information and/or additional information. ...” the generative model is a result of training a machine learning model using a plurality of training examples, each training example of the plurality of training examples is associated with a respective content; CIMINELLI [0078] “In some examples, the user interface elements obtained by step 404 may include at least one image, and step 404 may further comprise analyzing the textual input and/or the sketch to determine the at least one image. The at least one image may differ from the sketch. For example, step 404 may select, based on the textual input and/or the sketch, the at least one image of a plurality of alternative images. For example, step 404 may select the image most similar to a selected portion of the sketch of the plurality of alternative images. In another example, For example, step 404 may select the image compatible with a description included in the textual input of the plurality of alternative images. In some examples, step 404 may use a machine learning model to analyze the textual input and/or the sketch to generate the at least one image. The machine learning model may be a machine learning model trained using training examples to generate images based on texts and/or sketches. An example of such training example may include a sample textual input and/or a sample sketch, together with a sample image. In some examples, step 404 may identify a first mathematical object in a mathematical space, wherein the first mathematical object may correspond to a first word of the textual input, for example using module 302. Further, step 404 may identify a second mathematical object in the mathematical space, wherein the second mathematical object may correspond to a second word of the textual input, for example using module 302. Further, step 404 may calculate a function of the first mathematical object and the second mathematical object to obtain a third mathematical object in the mathematical space, for example using module 304. Further, step 404 may generate the at least one image based on the third mathematical object. For example, step 404 may use a function of the third mathematical object as a random seed or an input of a generative model for generating the at least one image.” It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR the capability to receive indication of desire to generate new content and to disclose a generative model trained in a plurality of examples. The benefit and motivation of such modification is discussed by CIMINELLI in the following portion: CIMINELLI [0078] “… In some examples, step 404 may use a machine learning model to analyze the textual input and/or the sketch to generate the at least one image. The machine learning model may be a machine learning model trained using training examples to generate images based on texts and/or sketches. An example of such training example may include a sample textual input and/or a sample sketch, together with a sample image. …” ZAFAR in view of CIMINELLI does not teach, but INAKOSHI teaches: based on the indication, using the input and the generative model to generate the new content, abstaining from basing the generation of the new content on any training example included in the particular subgroup; INAKOSHI [0137] “The generation unit 121c_3 receives the discrimination result by the discrimination unit 122c_3, and performs the learning by making the update seeking the parameter that makes it difficult for the discrimination unit 122c_3 to discriminate the inputted image from the true image. For example, the generation unit 121c_3 learns to generate an image identified as the image of the wild birds of Mexico (MX) or Canada (CA) and enters the state in which the influence of the image of the wild birds of America (US) is removed.” INAKOSHI [0141] “As described above, in the sequential learning device 10 according to Embodiment 3, if the influence of the specific learning data is to be removed from the obtained generation model and discrimination model, the already created generation model is selectively used. The generation model and the discrimination model from which the learning data is removed may be created. In the examples illustrated in FIGS. 10 and 11, each of a discrimination model and a generation model that identify and generate an image from which the influence of the wild birds of America (US) is removed may also be recreated, with respect to the generation model and the discrimination model about the image of the wild birds of Mexico (MX), the image of the wild birds of America (US), or the image of the wild birds of Canada (CA).” INAKOSHI [0129] “FIG. 11 is a diagram illustrating the action to generate a model from which the influence of the specific learning data is removed after the learning in Embodiment 3, the action in a case in which there is no real image. As illustrated in FIG. 11, description is given of the action to remove the influence of the learning data of America (real image of the wild birds of America) from the already created model, the action in a case in which the real image of the wild birds of Mexico (learning data) used in the action to generate the model in the sequential learning device 10c does not remain." It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI the capability to generate new content in which the model trained exclude some training examples. The benefit and motivation of such modification is discussed by INAKOSHI in the following portion: INAKOSHI [0119] “In the sequential learning device 10 according to Embodiment 2, as described above with reference to the example illustrated in FIG. 6, lastly, the discrimination model is obtained that learns to identify the image of the wild birds of Mexico (MX), the image of the wild birds of America (US), or the wild birds of Canada (CA). In this case, any influence of the learning data from the obtained model may have to be removed because the real image of the wild birds, which is the learning data provided from a specific vendor, includes an error or because the quality is poor. In Embodiment 3, description is given of an action to recreate a model from which any influence of the learning data is removed, assuming that of the learning data reflected in respective models of the last obtained generation unit 121c and the discrimination unit 122c, it is found out that the learning data on America has a problem.” Regarding claim 2, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI does not teach but INAKOSHI teaches: The non-transitory computer readable medium of claim 1, wherein the new content is at least one of a visual content, an audio content or a textual content. INAKOSHI [0137] “The generation unit 121c_3 receives the discrimination result by the discrimination unit 122c_3, and performs the learning by making the update seeking the parameter that makes it difficult for the discrimination unit 122c_3 to discriminate the inputted image from the true image. For example, the generation unit 121c_3 learns to generate an image identified as the image of the wild birds of Mexico (MX) or Canada (CA) and enters the state in which the influence of the image of the wild birds of America (US) is removed.” It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI the capability to generate new content in the form of images, audio or textual content. The benefit and motivation of such modification is discussed by INAKOSHI in the following portion: INAKOSHI [0029] “The sequential learning device 10 illustrated in FIG. 1 is a device that learns (sequential learning) new learning data, utilizing GAN and a model, using output (generation data) of a generation model that is already created. As illustrated in FIG. 1, the sequential learning device 10 includes a communication unit 11, a control unit 12, and a storage unit 13. In the following, description is given assuming that learning data as well as data or the like outputted by a generation model are “images”.” Regarding claim 3, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI does not teach but INAKOSHI teaches: The non-transitory computer readable medium of claim 1, wherein the particular subgroup is selected based on a geographical region. INAKOSHI [0137] “The generation unit 121c_3 receives the discrimination result by the discrimination unit 122c_3, and performs the learning by making the update seeking the parameter that makes it difficult for the discrimination unit 122c_3 to discriminate the inputted image from the true image. For example, the generation unit 121c_3 learns to generate an image identified as the image of the wild birds of Mexico (MX) or Canada (CA) and enters the state in which the influence of the image of the wild birds of America (US) is removed.” It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI the capability to discriminate regarding certain aspects of the data as location. The benefit and motivation of such modification is discussed by INAKOSHI in the following portion: INAKOSHI [0119] “In the sequential learning device 10 according to Embodiment 2, as described above with reference to the example illustrated in FIG. 6, lastly, the discrimination model is obtained that learns to identify the image of the wild birds of Mexico (MX), the image of the wild birds of America (US), or the wild birds of Canada (CA). In this case, any influence of the learning data from the obtained model may have to be removed because the real image of the wild birds, which is the learning data provided from a specific vendor, includes an error or because the quality is poor. In Embodiment 3, description is given of an action to recreate a model from which any influence of the learning data is removed, assuming that of the learning data reflected in respective models of the last obtained generation unit 121c and the discrimination unit 122c, it is found out that the learning data on America has a problem.” Regarding claim 4, The rejection of claim 1 is incorporated, furthermore ZAFAR teaches: The non-transitory computer readable medium of claim 1, wherein the particular subgroup is selected based on a regulatory requirement. ZAFAR [0006] “Although noise may be added to model outputs of ERM-type models to make the model outputs more or less independent from a single training record, doing so provides only statistical guarantees that may be negated at least in part by collecting many model outputs and cancelling out their noise. Moreover, adding noise necessarily decreases the accuracy of the model outputs. Also, the approach of Chaudhuri and the framework of differential privacy more generally concerns the influence of single records on model outputs, and so may not be able to sufficiently limit the dependence of model outputs on multiple training records. Fundamentally, the more records the model would need to be made more independent of, the more noise would need to be added and thus the more accuracy would have to be sacrificed. Effectively, adding noise provides a trade-off in which making model outputs more independent from training records results in a lower accuracy of the model outputs that are obtained. Moreover, in Chaudhuri's approach, although adding noise reduces the privacy impact of model outputs, the existing model that is stored is still a function of the personal information and so may still allow model inversion and/or may still represent personal information. For example, in various situations, applying noise to model outputs may not be regarded as a sufficient measure to satisfy right-to-be-forgotten request arising due to the GDPR and similar other privacy regulations.” ZAFAR [0013] “Generally, a removal request message may be sent for various reasons. For example, a removal request message may represent an absence of consent to further use a training instance, e.g., a withdrawal of consent to use the training instance. This can be the case when the training instance comprises personal information about a certain user. For example, the user itself may send the withdrawal of consent. Such a withdrawal of consent is sometimes also known as a right-to-be-forgotten request or right-to-erasure request. The withdrawal of consent can also be automatic, for example, the user may have provided a conditional consent, e.g., a time-limited consent or a consent dependent on another type of condition, and/or consent may be withdrawn by another party than the user: for example, another party with which a data sharing contract is in place. In these and other cases, the removal request message may be received from a consent management system configured to send the removal request message upon detecting that consent for using a training instance from the training dataset is missing. Such a consent management system can be combined with the system for processing a model, for example, in a single device." Regarding claim 5, The rejection of claim 1 is incorporated, furthermore ZAFAR teaches: The non-transitory computer readable medium of claim 1, wherein the particular subgroup is selected based on a source associated with the contents associated with the particular subgroup. ZAFAR [0065] “Interestingly, however, training dataset TD may still be stored or accessed along with the trained model in order to deal with removal request messages. The figure shows a removal request message RRM, 210. A removal request message may identify one or more undesired training instances UTI, 245, of the training dataset TD. For example, removal request message RRM shown in the figure indicates training instances TIi, 242 up to TIj, 243. The undesired training instances may be indicated in various ways, e.g., by including the instances in the message, or by including indices in the message. As another example, the training dataset TD may comprise multiple training instances collected from respective users, in which case removal request message RRM may indicate a user whose training instances are to be removed from the training dataset TD, for example, in the form of a user identifier or the like.” Regarding claim 14, The rejection of claim 1 is incorporated, furthermore ZAFAR does not teach but CIMINELLI teaches: The non-transitory computer readable medium of claim 1, wherein the input includes a textual input indicative of the desire to generate the new content using the generative model. CIMINELLI [0075] “In some examples, step 404 may comprise analyzing a textual input (such as textual input, the textual input received by step 402, the textual input received by step 802, textual data 102, etc.) and/or a sketch (such as the sketch received by step 402, the sketch received by step 802, sketch data 104, etc.) to obtain user interface elements (for example, a single user interface element, at least two user interface elements, at least ten user interface elements, and so forth). In one example, step 404 may comprise analyzing the textual input and/or the sketch to select the user interface elements of a group of alternative user interface elements. In one example, step 404 may comprise analyzing the textual input and/or the sketch to generate the user interface elements, for example using a generative model. In one example, step 404 may comprise using a machine learning model (such as a unimodal machine learning model, a multimodal machine learning model, a generative model, etc.) to analyze the textual input and/or the sketch and/or additional information and obtain the at least two user interface elements. The machine learning model may be a machine learning model trained using training examples to generate user interface elements and/or to select user interface elements of a plurality of alternatives based on textual information and/or visual information and/or additional information. ...” It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR the capability to add textual request specifying desired content. The benefit and motivation of such modification is discussed by CIMINELLI in the following portion: CIMINELLI [0078] “… In some examples, step 404 may use a machine learning model to analyze the textual input and/or the sketch to generate the at least one image. The machine learning model may be a machine learning model trained using training examples to generate images based on texts and/or sketches. An example of such training example may include a sample textual input and/or a sample sketch, together with a sample image. …” Regarding claim 15, The rejection of claim 14 is incorporated, furthermore ZAFAR does not teach, but CIMINELLI teaches: The non-transitory computer readable medium of claim 14, wherein the textual input includes a noun and an adjective adjacent to the noun, wherein for each training example of the particular subgroup, the respective content associated with the training example includes a respective object of a category of objects selected based on the noun wherein at least one aspect of the respective object matches the adjective. CIMINELLI [0091] “In some examples, a textual input (such as textual input, the textual input received by step 402, textual data 102, etc.) may include an adjective. Further, a characteristic of a particular user interface element in the new user interface of method 400 (such as a particular user interface element of the at least two user interface elements obtained by step 404) may be based on the adjective. In one example, the adjective may be an adjective that describes size, and the characteristic of the particular user interface element may be a size of the particular user interface element. For example, the particular user interface element may be a text box, the textual input may include either ‘The user may enter long descriptions’ or ‘The user may enter short descriptions’, and the size of the text box may be bigger or smaller based on the adjective. In one example, the adjective may be an adjective that describes texture, and the characteristic of the particular user interface element may be a texture associated with at least part of the particular user interface element. For example, the particular user interface element may be a background, the textual input may include either ‘Let's present everything over a smooth wall’ or ‘Let's present everything over a grainy wall’, and the textual of the background may be either smooth or grainy based on the adjective. In one example, the adjective may be an adjective that describes color, and the characteristic of the particular user interface element may be a color associated with at least part of the particular user interface element. For example, the particular user interface element may be a textual element associated with a notification, the textual input may include either ‘Let's show notification in dark red’ or ‘Let's show notification in light red’, and the color of the textual element may be either dark or light red based on the adjective. In one example, the adjective may be an adjective that describes shape, and the characteristic of the particular user interface element may be a shape of the particular user interface element. For example, the particular user interface element may be a button, the textual input may include either ‘Let's have round buttons’ or ‘Let's have oblong buttons’, and the shape of the button may be selected based on the adjective.” It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR the capability to include adjective interpretation. The benefit and motivation of such modification is discussed by CIMINELLI in the following portion: CIMINELLI [0078] “… In one example, the adjective may be an adjective that describes shape, and the characteristic of the particular user interface element may be a shape of the particular user interface element. For example, the particular user interface element may be a button, the textual input may include either ‘Let's have round buttons’ or ‘Let's have oblong buttons’, and the shape of the button may be selected based on the adjective.” Regarding claim 16, The rejection of claim 14 is incorporated, furthermore ZAFAR does not teach but CIMINELLI teaches: The non-transitory computer readable medium of claim 14, wherein the textual input includes a verb and an adverb adjacent to the verb, wherein for each training example of the particular subgroup, the respective content associated with the training example includes a respective event of a category of events selected based on the verb wherein at least one aspect of the respective event matches the adverb. CIMINELLI [0104] “In some examples, the textual input may include a verb. Further, based on the verb, step 510 may select a type of change. For example, step 510 may use a LLM to analyze the textual input including the verb and selecting the type of change. Further, step 510 may include a change of the selected type in the at least one change. In one example, the verb may be associated with a noun and/or a pronoun in the textual input, step 510 may select a particular element of the elements included in the selected portion of the user interface (of step 506) based on the noun and/or the pronoun (for example as described above), and step 510 may include a change of the selected type to the particular element in the at least one change. For example, the textual input may include either ‘Quiver the notification to draw attention to it’ or ‘Flicker the notification to draw attention to it’, and step 510 may make the notification in the selected portion either quiver or flicker based on the verb. In one example, the verb may be associated with at least one of a user or a viewer of the user interface. For example, when the textual input includes ‘Most important to our clients is the ability to easily see the status’, step 510 may include in the at least one change a size change to enlarge the status indicator, and when the textual input includes ‘Most important to our clients is the ability to easily change the status’, step 510 may include in the at least one change an addition change to add of a user interface element for modifying the status.” CIMINELLI [0107] “In some examples, the textual input may include an adverb. Further, based on the adverb, step 510 may determine a change to a particular animation associated with a particular element of the elements included in the selected portion of the user interface. For example, step 510 may use a LLM to analyze the textual input including the adverb and determining the change to the particular animation. Further, step 510 may include the determined change to the particular animation in the at least one change. For example, the change to the particular animation may include at least one of a change to a pace associated with the particular animation, a change to a frequency associated with the particular animation, a change to timing associated with the particular animation, a change to a type of the particular animation, a change to a location associated with the particular animation, or a change to an extent associated with the particular event. In one example, the textual input may include either ‘The notification should move slowly to the designated area’ or ‘The notification should move quickly to the designated area’, and step 510 may modify the motion speed of the notification based on the adverb." It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR the capability to include verb/adverb interpretation. The benefit and motivation of such modification is discussed by CIMINELLI in the following portion: CIMINELLI [0107] “… For example, the change to the particular animation may include at least one of a change to a pace associated with the particular animation, a change to a frequency associated with the particular animation, a change to timing associated with the particular animation, a change to a type of the particular animation, a change to a location associated with the particular animation, or a change to an extent associated with the particular event. In one example, the textual input may include either ‘The notification should move slowly to the designated area’ or ‘The notification should move quickly to the designated area’, and step 510 may modify the motion speed of the notification based on the adverb.” Regarding claim 17, The rejection of claim 14 is incorporated, furthermore ZAFAR does not teach, but CIMINELLI teaches: The non-transitory computer readable medium of claim 14, wherein the textual input includes a preposition, wherein for each training example of the particular subgroup, the respective content associated with the training example includes a respective pair of objects wherein a spatial relation between the pair of objects in the respective content matches the preposition. CIMINELLI [0105] “In some examples, the textual input may include a preposition. Further, based on the preposition, step 510 may select a new location for a particular element of the elements included in the selected portion of the user interface. For example, step 510 may use a LLM to analyze the textual input including the preposition and selecting the new location. Further, step 510 may include a change of a location of the particular element to the new location in the at least one change. For example, the textual input may include either ‘Put the status above the notifications’ or ‘Put the status below the notifications’, and step 510 may include in the at least one change moving the status indicator to a position above or below the notifications based on the preposition.” It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR the capability to include preposition interpretation. The benefit and motivation of such modification is discussed by CIMINELLI in the following portion: CIMINELLI [0105] “In some examples, the textual input may include a preposition. Further, based on the preposition, step 510 may select a new location for a particular element of the elements included in the selected portion of the user interface. For example, step 510 may use a LLM to analyze the textual input including the preposition and selecting the new location. Further, step 510 may include a change of a location of the particular element to the new location in the at least one change. For example, the textual input may include either ‘Put the status above the notifications’ or ‘Put the status below the notifications’, and step 510 may include in the at least one change moving the status indicator to a position above or below the notifications based on the preposition.” Regarding claim 18, The rejection of claim 1 is incorporated, furthermore ZAFAR teaches: The non-transitory computer readable medium of claim 1, wherein the input includes input image data. ZAFAR [0015] “The techniques described herein in accordance with example embodiments of the present invention can be used for various types of trained model. In particular, interestingly, the techniques described herein may be applied to non-linear trained models. In various embodiments, the trained model is a classifier, e.g., a model configured to classify input instances into one or more classes. For example, the model may be a binary classifier for two classes or a multiclass classifier for three or more classes. In various embodiments, the trained model is a regression model, for example, a model configured to predict values for one or more output quantities, e.g., a real-valued output, given an input instance. In various embodiments, the trained model is an object detection model, for example, a model configured to detect one or more objects in an input image, e.g., to output a location of an object of a given type detected in the input instance. In various embodiments, the trained model is a segmentation model, e.g., a model configured to associate features of an input instance, e.g., pixels of an input image, with respective class labels. In various embodiments, the trained model is a generative model, e.g., a model configured to generate instances such as images based on latent feature vectors. The model can also be a time-series model, e.g., for time-series modelling or forecasting.” Regarding claim 19, Arguments analogous to claim 1 are applicable (see the rejections of the corresponding limitations in the rejection of claim 1 above), furthermore ZAFAR teaches: A system for generating content using a generative model without relying on selected training examples, the system comprising: at least one processor configured to perform the operations of: ZAFAR [Abstract] “A system for processing a model. The model provides a model output given an input instance. The model has been trained on a training dataset by iteratively optimizing an objective function including losses according to a loss function for training instances of the training dataset. Upon receiving a removal request message identifying one or more undesired training instances of the training dataset, the model is made independent from the one or more undesired training instances. To this end, the one or more undesired training instances are removed from the training dataset to obtain a remainder dataset, and an adapted model is determined for the remainder dataset. The parameters of the adapted model are first initialized based on the set of parameters of the trained model, and then iteratively adapted by optimizing the objective function with respect to the remainder dataset.” ZAFAR [0056] “In general, the system 100 may be embodied as, or in, a single device or apparatus, such as a workstation, e.g., laptop or desktop-based, or a server. The device or apparatus may comprise one or more microprocessors which execute appropriate software. For example, the processor subsystem may be embodied by a single Central Processing Unit (CPU), but also by a combination or system of such CPUs and/or other types of processing units. The software may have been downloaded and/or stored in a corresponding memory, e.g., a volatile memory such as RAM or a non-volatile memory such as Flash. Alternatively, the functional units of the system, e.g., the data interface and the processor subsystem, may be implemented in the device or apparatus in the form of programmable logic, e.g., as a Field-Programmable Gate Array (FPGA) and/or a Graphics Processing Unit (GPU). In general, each functional unit of the system may be implemented in the form of a circuit. It is noted that the system 100 may also be implemented in a distributed manner, e.g., involving different devices or apparatuses, such as distributed servers, e.g., in the form of cloud computing.” Regarding claim 20, Arguments analogous to claim 1 are applicable (see the rejections of the corresponding limitations in the rejection of claim 1 above), furthermore ZAFAR teaches: A system for generating content using a generative model without relying on selected training examples, the system comprising: at least one processor configured to perform the operations of: ZAFAR [Abstract] “A system for processing a model. The model provides a model output given an input instance. The model has been trained on a training dataset by iteratively optimizing an objective function including losses according to a loss function for training instances of the training dataset. Upon receiving a removal request message identifying one or more undesired training instances of the training dataset, the model is made independent from the one or more undesired training instances. To this end, the one or more undesired training instances are removed from the training dataset to obtain a remainder dataset, and an adapted model is determined for the remainder dataset. The parameters of the adapted model are first initialized based on the set of parameters of the trained model, and then iteratively adapted by optimizing the objective function with respect to the remainder dataset.” ZAFAR [0007] “In accordance with a first aspect of the present invention, a system for processing a model trained based on a loss function is provided. In accordance with another aspect of the present invention, a corresponding computer-implemented method is provided. In accordance with an aspect of the present invention, a computer-readable medium is provided.” Claim(s) 6 and 11-13 is/are rejected under 35 U.S.C. 103 as being unpatentable over ZAFAR, in view of CIMINELLI, in view of INAKOSHI in further view of SAGAR; Mark et al. (US 20220358369 A1), hereinafter SAGAR. Regarding claim 6, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI in view of INAKOSHI does not teach but SAGAR teaches: The non-transitory computer readable medium of claim 1, wherein the operations further comprise analyzing the input to select the particular subgroup. SAGAR [0083] A stimulus whose associated memories to be forgotten is input. In the example above, a gun shot is provided on audio Input Field, the ASOM Alpha Weight for video is set to 0 (to retrieve all videos associated with the gun shot). The resulting Activation Map, can be used directly as the Reset Mask. Alternatively, a discrete Reset Mask may be created, e.g. setting the Reset Mask to 1 for all neurons whose activation is greater than a set threshold and to 0 for the rest. Or, set it to 1 for the winning neuron and to 0 for all the other neurons. The Reset Mask is set to 1−Training Record, or to its discretized version (Reset Mask[i]=1, if Training Record[i]<threshold, and 0 otherwise). Training Record Decay is set to a value<1 during training. This will cause Training Record to erode to zero over time for those neurons that are not “refreshed” by new training. Then Reset Mask is set to 1-Training Record, or to its discretized version (Reset Mask[i]=1, if Training Record[i]<Reset Threshold, and 0 otherwise). It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI in view of INAKOSHI the capability to receive an input to select a certain subgroup. The benefit and motivation of such modification is discussed by SAGAR in the following portion: SAGAR [0083] “A stimulus whose associated memories to be forgotten is input. In the example above, a gun shot is provided on audio Input Field, the ASOM Alpha Weight for video is set to 0 (to retrieve all videos associated with the gun shot). The resulting Activation Map, can be used directly as the Reset Mask.” Regarding claim 11, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI in view of INAKOSHI does not teach but SAGAR teaches: The non-transitory computer readable medium of claim 1, wherein the generative model includes a plurality of artificial neurons, and the operations further comprise: identifying a specific subgroup of the plurality of artificial neurons associated with the particular subgroup of at least one but not all of the plurality of training examples; SAGAR [0078] “Targeted forgetting is controlled by a mask (analogical to an Activation Mask), termed a “Reset Mask”. The Reset Mask is isomorphic to the SOM (i.e. with one Mask Value for each SOM Neuron). When replacing weight vectors of neurons with noise, only those neurons whose Reset Mask=1 will are reset, others (with Reset Mask=0) will are retained.” SAGAR [0045] “It may be desirable to reconstruct items stored in a SOM at random. This happens, for instance, in the construction of pseudo-training items during consolidation of long-term memories or in the random generation of motor movements during motor babbling. In these cases, the SOM's training record drives stochastic selection of SOM Neurons to reconstruct from. Sampling may be combined with an inhibition-of-return (IOR) process, to sample from the full set of trained values.” SAGAR [0083] “A stimulus whose associated memories to be forgotten is input. In the example above, a gun shot is provided on audio Input Field, the ASOM Alpha Weight for video is set to 0 (to retrieve all videos associated with the gun shot). The resulting Activation Map, can be used directly as the Reset Mask. Alternatively, a discrete Reset Mask may be created, e.g. setting the Reset Mask to 1 for all neurons whose activation is greater than a set threshold and to 0 for the rest. Or, set it to 1 for the winning neuron and to 0 for all the other neurons. The Reset Mask is set to 1−Training Record, or to its discretized version (Reset Mask[i]=1, if Training Record[i]<threshold, and 0 otherwise). Training Record Decay is set to a value<1 during training. This will cause Training Record to erode to zero over time for those neurons that are not “refreshed” by new training. Then Reset Mask is set to 1-Training Record, or to its discretized version (Reset Mask[i]=1, if Training Record[i]<Reset Threshold, and 0 otherwise)." and abstaining from basing the generation of the new content on outputs generated by any one of the specific subgroup of the plurality of artificial neurons. SAGAR [0080] “During resetting, the Training Record may be updated, such that reset Neurons' training records are cleared (i.e. in case of discrete Reset Mask, Training Record:=0 for those neurons whose Reset Mask=1). In case of continuous mixing (blurring the memory): new_training_record[i]=(1−reset_mask[i])*original_training_record[i]” SAGAR [0047] “To provide a cleaner reconstructed output, more weight may be given to Neurons that have received more training (disregarding untrained neurons). The amount of adaptation of each Neuron may be recorded as a value between 0 and 1, accessible in the SOM parameter “Training Record”." It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI in view of INAKOSHI the capability to add zero activation mask for neurons associated with selected memories/ training data or a certain subgroup and further preventing their output to contribute to the final output. The benefit and motivation of such modification is discussed by SAGAR in the following portion: SAGAR [0078] “Targeted forgetting is controlled by a mask (analogical to an Activation Mask), termed a “Reset Mask”. The Reset Mask is isomorphic to the SOM (i.e. with one Mask Value for each SOM Neuron). When replacing weight vectors of neurons with noise, only those neurons whose Reset Mask=1 will are reset, others (with Reset Mask=0) will are retained.” Regarding claim 12, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI in view of INAKOSHI does not teach but SAGAR teaches: The non-transitory computer readable medium of claim 1, wherein the generative model includes a plurality of weights, and the operations further comprise: identifying a specific subgroup of the plurality of weights associated with the particular subgroup of at least one but not all of the plurality of training examples; SAGAR [0078] “Targeted forgetting is controlled by a mask (analogical to an Activation Mask), termed a “Reset Mask”. The Reset Mask is isomorphic to the SOM (i.e. with one Mask Value for each SOM Neuron). When replacing weight vectors of neurons with noise, only those neurons whose Reset Mask=1 will are reset, others (with Reset Mask=0) will are retained.” SAGAR [0083] “A stimulus whose associated memories to be forgotten is input. In the example above, a gun shot is provided on audio Input Field, the ASOM Alpha Weight for video is set to 0 (to retrieve all videos associated with the gun shot). The resulting Activation Map, can be used directly as the Reset Mask. Alternatively, a discrete Reset Mask may be created, e.g. setting the Reset Mask to 1 for all neurons whose activation is greater than a set threshold and to 0 for the rest. Or, set it to 1 for the winning neuron and to 0 for all the other neurons. The Reset Mask is set to 1−Training Record, or to its discretized version (Reset Mask[i]=1, if Training Record[i]<threshold, and 0 otherwise). Training Record Decay is set to a value<1 during training. This will cause Training Record to erode to zero over time for those neurons that are not “refreshed” by new training. Then Reset Mask is set to 1-Training Record, or to its discretized version (Reset Mask[i]=1, if Training Record[i]<Reset Threshold, and 0 otherwise)." and abstaining from basing the generation of the new content on any one of the specific subgroup of the plurality of weights. SAGAR [0079] “Alternatively, Reset Mask values may be between 0 and 1, in which case the original weight vector will be mixed with random noise with a mixing coefficient determined by the Reset Mask value: new_weight[i]=(1−reset_mask[i])*original_weight[i]+reset_mask[i]*noise” SAGAR [0031] “In other embodiments, rather than clearing the entire Experience Memory Store, updated or modified experiences may be located in the Experience Memory Store and selectively deleted form the Experience Memory Store, by “unlearning” specific data points. In a model of “forgetting”, Experiences are time-stamped, or otherwise marked to indicate recency of the memory, and older events may be “forgotten”, by deleting these from the Experience Memory Store and/or Memory Database." It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI in view of INAKOSHI the capability to add complete resetting of selected weight vectors of selected data, eliminating their value while preserving the unselected vectors. The benefit and motivation of such modification is discussed by SAGAR in the following portion: SAGAR [0078] “Targeted forgetting is controlled by a mask (analogical to an Activation Mask), termed a “Reset Mask”. The Reset Mask is isomorphic to the SOM (i.e. with one Mask Value for each SOM Neuron). When replacing weight vectors of neurons with noise, only those neurons whose Reset Mask=1 will are reset, others (with Reset Mask=0) will are retained.” Regarding claim 13, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI in view of INAKOSHI does not teach but SAGAR teaches: The non-transitory computer readable medium of claim 1, wherein the generative model includes a memory unit that stores a plurality of patterns, and the operations further comprise: identifying a specific subgroup of the plurality of patterns associated with the particular subgroup of at least one but not all of the plurality of training examples; SAGAR [0027] “The Memory Database may be used to train the Experience Memory Store. Entries in the Memory Database are provided as training inputs to the Experience Memory Store during consolidation. Memories encoded in the Experience Memory Store allow the agent to recognize objects, concept, events and make predictions. As one example, a user can generate the set of input files for specific learning domains. For example, an Agent can become a “dog expert” without experiencing dogs during live operation, by being provided with a Memory Database with images of different dog breeds with associated Modalities, such symbols comprising the names of the dogs, spectrograms of the sounds of their barking, and emotional responses that the dogs would evoke.” SAGAR [0029] “Entries in the Memory Database can be re-read to reconstruct memories: for example they can train a short-term memory Experience Memory Store, creating “virtual events” or train a Long term memory Experience Memory Store during memory consolidation. It may be possible to reconstruct the raw sensor input (such as an image) that triggered a learning event; from the Experience Memory Store as the raw sensory input is stored in the weights of neurons of low-level maps. However, as potentially several different Input Vectors can modify the weights of a single neuron, the resulting weights in the neural-network may be a blend of several input instances. As the Memory Database explicitly stores individual Input Vectors and their constituent Input Fields as separate entries with associated attributes, the Memory Database provides a way to accurately reconstruct individual Experiences." and abstaining from basing the generation of the new content on any one of the specific subgroup of the plurality of patterns. SAGAR [0035] “Memories may be categorized, labelled or tagged in a manner which makes individual memories easy to locate, modify and/or delete. A user interface may be provided to facilitate users in viewing and editing the memory of Embodied Agents.” SAGAR [0031] “In other embodiments, rather than clearing the entire Experience Memory Store, updated or modified experiences may be located in the Experience Memory Store and selectively deleted form the Experience Memory Store, by “unlearning” specific data points. In a model of “forgetting”, Experiences are time-stamped, or otherwise marked to indicate recency of the memory, and older events may be “forgotten”, by deleting these from the Experience Memory Store and/or Memory Database." It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI in view of INAKOSHI the capability to add selective deletion of stored memory patterns and preventing them from contributing to subsequent generation. The benefit and motivation of such modification is discussed by SAGAR in the following portion: SAGAR [0031] “In other embodiments, rather than clearing the entire Experience Memory Store, updated or modified experiences may be located in the Experience Memory Store and selectively deleted form the Experience Memory Store, by “unlearning” specific data points. In a model of “forgetting”, Experiences are time-stamped, or otherwise marked to indicate recency of the memory, and older events may be “forgotten”, by deleting these from the Experience Memory Store and/or Memory Database." Claim(s) 7 is/are rejected under 35 U.S.C. 103 as being unpatentable over ZAFAR, in view of CIMINELLI, in view of INAKOSHI in further view of Zass; Ron et al. (US 20230048149 A1), hereinafter ZASS. Regarding claim 7, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI in view of INAKOSHI does not teach but ZASS teaches: The non-transitory computer readable medium of claim 1, wherein the operations further comprise: calculating a convolution of at least part of the content associated with a particular training example of the plurality of training examples to obtain a numerical result value; ZASS [0132] “In some examples, Step 474 may analyze the image data received by Step 422 to detect at least three events, and may select the plurality of events of the at least three events. In one example, for each event of the at least three events, Step 474 may calculate a convolution of at least part of the image data and thereby obtain a result value of the calculated convolution of the at least part of the image data, and may determine whether to include the event in the plurality of events based on the result value of the calculated convolution of the at least part of the image data.” and determining whether to include the particular training example in the particular subgroup based on the numerical result value. ZASS [0132] “In some examples, Step 474 may analyze the image data received by Step 422 to detect at least three events, and may select the plurality of events of the at least three events. In one example, for each event of the at least three events, Step 474 may calculate a convolution of at least part of the image data and thereby obtain a result value of the calculated convolution of the at least part of the image data, and may determine whether to include the event in the plurality of events based on the result value of the calculated convolution of the at least part of the image data.” It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI in view of INAKOSHI the capability to add convolution based content analysis producing a numerical recognition result. The benefit and motivation of such modification is discussed by ZASS in the following portion: ZASS [0105] “… In some examples, Step 424 may calculate a convolution of at least part of the image data and thereby obtain a result value of the calculated convolution of the at least part of the image data, and may identify the plurality of objects based on the result value. For example, when the result value is a first value, Step 424 may identify a first plurality of objects, and when the result value is a second value, Step 424 may identify a second plurality of objects. The second plurality of objects may differ from the first plurality of objects.” Claim(s) 8 is/are rejected under 35 U.S.C. 103 as being unpatentable over ZAFAR, in view of CIMINELLI, in view of INAKOSHI in further view of MADL; Tamas et al. (US 20230315899 A1), hereinafter MADL. Regarding claim 8, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI in view of INAKOSHI does not teach but MADL teaches: The non-transitory computer readable medium of claim 1, wherein the operations further comprise: for each training example of the plurality of training examples, analyzing the respective content associated with the training example to identify a respective mathematical object corresponding to the respective training example in a mathematical space; MADL [0021] “In some examples, when privacy is desired, datapoints are to be unidentifiable to protect against reidentification scenarios by minimizing the risk of singling out as estimated by the identifiability risk. Identifiability can be defined as follows. Let w be a user-defined weight vector encoding the identifiability of various features based on their domain knowledge. In the simplest case, w.sub.i can be set to 0 for non-identifiable features and to 1 for the rest. In the experiments below, we will use entropy as the weight vector, to capture the informativeness of individual features.” MADL [0022] “T and S denote a real and corresponding synthetic dataset, respectively. Let r.sub.i denote the minimum distance between datapoint x.sub.i and its nearest neighbor in weighted Euclidean distance space, such that and, similarly, let {circumflex over (r)}.sub.i denote the minimum distance between a datapoint x.sub.i and its next-closest synthetic datapoint {circumflex over (x)}.sub.j: A synthetic dataset S is ∈-identifiable from T if” MADL [0059] “Pairs of datapoints within the entire original dataset are determined at 810. This determination is made based on distance metrics such as a (weighted) Euclidean distance, Manhattan distance, etc. The metric(s) may be default or provided by a request for synthetic dataset generation.” MADL [0061] “For each datapoint in the original dataset several acts may be performed as detailed below. At 816, the datapoint and a number of nearest neighbors are set. Then, for each of the number of nearest neighbors, a plurality of acts is performed.” MADL [0062] “Local neighborhoods near the set datapoint and the number of pairs of datapoints are determined at 818. These local neighborhoods are projected onto a line connecting the initialized datapoint and the pairs of datapoints at 820. The projection is a 1-D projection." based on the identified mathematical objects, identifying a region in the mathematical space; MADL [0060] “In some examples, when privacy is desired, for each datapoint in the original dataset a zone for which a synthetic point cannot be inside at 712. This creates the no go zones. In particular, the datapoints are to be unidentifiable to protect against reidentification scenarios by minimizing the risk of singling out as estimated by the identifiability risk. The choosing of a set of synthetic datapoints between pairs of datapoints of a dataset along a line connecting them while sampling a likely value of a local probability distribution, and, in some examples, ensure that none of the synthetic datapoints in the set too closely matches the original data at 608 may also include several acts.” using the input to select a specific mathematical object in the mathematical space outside the identified region; MADL [0028] “In some examples, the provider network 100 includes a ML model training and/or testing service 106. This service may be used to test the synthetic dataset against testing data of the original dataset. In some examples the request (or a plurality of requests) allows a user to set parameters for the generation of the synthetic dataset by the synthetic dataset generation service that utilizes DDE 104 or the synthetic dataset generation service using privacy filtering 105. For example, the request may include one or more of an identification of what features to exclude from the original dataset, an identification of location of the original dataset, the original dataset itself, a number of synthetic datapoints to generate, whether privacy should be taken into consideration when generating the synthetic datapoints of the dataset, an identification of location of where to store the synthetic dataset, a number of pairs of datapoints to consider, a neighborhood size, an indication of metrics to use to minimize identifiability, etc.) and/or parameters to test the synthetic dataset such what data to keep from the original dataset as testing data, what accuracy metric to use, what model to use (and, in some examples, to train with both datasets), etc. In some examples, this information is provided via a graphical user interface, however, command line interface commands may also be used.” MADL [0063] “A probability distribution estimation (e.g., a density estimate (such as KDE)) is performed on the projection at 822. The probability distribution estimation value is then mapped into high-dimensional space (e.g., the high-dimensional space of the original dataset) at 824. When privacy is to be taken into account, the local neighborhoods are excluded if they are in the zone for which a synthetic point cannot be inside at 826.” MADL [0064] “A value for a datapoint to potentially added to the synthetic dataset is determined by sampling from the estimated probability at 828." and using the specific mathematical object to generate the new content. MADL [0063] “A probability distribution estimation (e.g., a density estimate (such as KDE)) is performed on the projection at 822. The probability distribution estimation value is then mapped into high-dimensional space (e.g., the high-dimensional space of the original dataset) at 824. When privacy is to be taken into account, the local neighborhoods are excluded if they are in the zone for which a synthetic point cannot be inside at 826.” MADL [0065] “The datapoint is added to the synthetic dataset if the determined value of 728 is greater than 0 at 830.” MADL [0055] “The synthetic dataset is provided at 712 as configured by the request. For example, in most instances it is stored at a designated location." It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI in view of INAKOSHI the capability to add datapoint based prohibited regions and sampling outside those regions. The benefit and motivation of such modification is discussed by MADL in the following portion: MADL [0039] “… In addition to high-fidelity data generation, nearest-neighbor based density approximation also allows explicitly minimizing identifiability risk, by avoiding synthetic data instances nearer to real instances than their next-nearest neighbors.” Claim(s) 9 and 10 is/are rejected under 35 U.S.C. 103 as being unpatentable over ZAFAR, in view of CIMINELLI, in view of INAKOSHI in further view of Hazard; Christopher James et al. (US 20210326652 A1), hereinafter HAZARD. Regarding claim 9, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI in view of INAKOSHI does not teach but HAZARD teaches: The non-transitory computer readable medium of claim 1, wherein the operations further comprise: using the input and the generative model to generate a plurality of alternative contents; HAZARD [0077] “Upon completion of determination and testing of the synthetic data case, optionally, it can be provided 170 as synthetic data. For example, the synthetic data case may be provided 170 in response to the received 110 request for data. In some embodiments, multiple synthetic data cases may be created in response to receiving 110 the original request, and may be provided 170 in response to that request. Providing the synthetic data case(s) in response to the request can take any appropriate form, including having them sent via HTTP, HTTPS, FTP, FTPS, via an API, a remote procedure call, a function or procedure call, etc., and/or in response to one of the foregoing.” for each content of the plurality of alternative contents, determining a likelihood that the generation of the content is based on any training example included in the particular subgroup; HAZARD [0035] “Some embodiments herein use a combination of metrics for analysis of the fitness or data quality of the generated dataset. The fitness or dataset quality metrics may include one or more statistical quality metrics that compare the statistical properties of the set of training data cases and the set of two or more synthetic data cases (where an example statistical property is mean, standard deviation, metrics for similarity of distribution, etc.); one or more model comparison metrics, which may quantify the machine learning model properties and performance of the set of training data cases and the set of two or more synthetic data cases; and at least one privacy metric, which may quantify the likelihood of identification of private data in the set of training data cases from the set of two or more synthetic data cases. In some embodiments, only data that meets certain fitness or data quality thresholds (e.g., measuring the fitness metric or data quality metric against a threshold) in order for the data to be fit for use. As discussed herein, the data may be resampled and replaced or discarded if a data quality or fitness metric does not meet a respective threshold.” HAZARD [0180] “In some embodiments, privacy metrics may quantify the likelihood of identification of private data in the set of training data cases from the set of two or more synthetic data cases. Not depicted in the figures, some of the embodiments herein may be performed based on the entire set of training data and/or based on clusters within the set of training data. For example, in some embodiments, the closeness percentile metric may be based on the percentile of the distance of the synthetic data case within a particular cluster in the set of training data (vs. within the entire distribution of set of training data cases). This may be useful when the set of training data cases has clusters of data within it, and the percentile distances for the synthetic data cases are more accurate when taken as percentiles within just that cluster of data." and selecting the new content of the plurality of alternative contents based on the determined likelihoods. HAZARD [0065] “In some embodiments, a single data case may be tested for fitness and/or similarity 160 (in the case of FIG. 1A, 1B, 1C, and FIG. 1D) and/or multiple data cases may be tested together for fitness and/or similarity (not depicted in FIG. 1A, 1B, 1C, and FIG. 1D). For example, the k-anonymity, validity, and/or similarity of the set of two or more data cases may be tested 160 together, and the decision whether to retest, discard, and/or replace the set of two or more data cases would be made as a whole. Further, in some embodiments, any of the metrics described herein for dataset quality may be used to test the similarity and/or fitness for any of one or more data case as part of testing fitness and/or similarity 160. For example, any of the metric or calculations used (together or separately) as part of determining 630 a dataset quality metric may be used to determine the fitness and/or similarity 160 of one or more data cases.” HAZARD [0096] “As noted above, similarity of newly-generated synthetic data cases may also be tested using certainty scores (e.g., any of the conviction measures discussed here). If the certainty score is beyond a certain threshold (e.g., prediction conviction being over a threshold), then the case may be modified (e.g., resampled) and retested, discarded, and/or replaced. In some embodiments, as discussed elsewhere herein, a function of the certainty score may be used to determine whether to modify (e.g., resample) and retest, discard, and/or replace the newly-generated synthetic data case, and this function may additionally test to see if the certainty score is beyond or within some threshold or thresholds." It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI in view of INAKOSHI the capability to add evaluation and selection of alternative outputs against excluded subgroups. The benefit and motivation of such modification is discussed by HAZARD in the following portion: HAZARD [0098] “… For example, it may be desirable that the synthetic data not represent any particular existing training data in order to ensure that a model can be made without including that training data. For example, a data owner may want to share data (e.g., data related to machine vibration analysis) with a collaborator company, but may not want to share data specific to their operation. As such, they may desire that the synthetic data not be overly similar to their existing training data, but still be useful for the collaborator company. The same can be true for a portion of the data. For example, the data owner may want to ensure that data related to a particular driver (for self-driving vehicles) not be included in the synthetic data. As such, that driver's data may be checked against the synthetic data for over similarity, and any overly-similar data may be modified or discarded.” Regarding claim 10, The rejection of claim 1 is incorporated, furthermore ZAFAR in view of CIMINELLI in view of INAKOSHI does not teach but HAZARD teaches: The non-transitory computer readable medium of claim 1, wherein the operations further comprise: using the input and the generative model to generate a preliminary content; HAZARD[0050] “FIG. 1A is a flow diagram depicting example processes for synthetic data generation in computer-based reasoning systems. In some embodiments, process 100 proceeds by receiving 110 a request for synthetic data. For example, a system or system operator may request additional or different training data in order to train a computer-based reasoning that will be used to control a system. In some cases, the system or operator may request anonymous data that is similar to a current training dataset (or different from, but still anonymized). In other cases, the system or operator may require more data than is in the current training dataset, and therefore may request additional data to augment the current training dataset. In some cases, synthetic data may be requested to direct sampling via a reinforcement learning process. The synthesized data (perhaps combined with original training data or by itself) may be used as part of a computer-based reasoning system to cause control of a system. Many controllable systems can be controlled with the techniques herein, such as controllable machinery, autonomous vehicles, lab equipment, etc. In some embodiments, the request for synthetic data may include a target surprisal and/or conviction for the target data. In some embodiments, if low target surprisal is requested, then the synthetic data may be close to and not differ much from existing data. If high target surprisal is requested, then the generated synthetic data may differ much from the existing data.” HAZARD[0077] “Upon completion of determination and testing of the synthetic data case, optionally, it can be provided 170 as synthetic data. For example, the synthetic data case may be provided 170 in response to the received 110 request for data. In some embodiments, multiple synthetic data cases may be created in response to receiving 110 the original request, and may be provided 170 in response to that request. Providing the synthetic data case(s) in response to the request can take any appropriate form, including having them sent via HTTP, HTTPS, FTP, FTPS, via an API, a remote procedure call, a function or procedure call, etc., and/or in response to one of the foregoing." and based on the indication, analyzing the preliminary content to modify the preliminary content. HAZARD[0097] “One or more newly generated synthetic data cases may be tested against at least a portion of the training data. For example, in some embodiments, it may be important that newly generated cases differ from all data in the training data. As such, newly generated data cases may be assessed for similarity with all data cases in the training data. If over-similarity is found, then the synthetic data case may be modified (and retested), discarded, or replaced (and the replacement synthetic data case tested for similarity). In some embodiments, it may only be important that the synthetic data case differ from a portion of the existing training data. As such, the newly generated data case may be tested for similarity to only that portion of the cases. As a more specific example, if portions of the training data have personal information and other portions do not (e.g., the latter may have been synthetically generated, and/or may be part of the public domain), then the test for similarity may only be with respect to the portion of concern—the former portion containing personal information. In some embodiments, determining similarity (or fitness) 160 of the generated or synthetic data case may include determining the k-anonymity of the data case (as discussed elsewhere herein).” It would have been obvious to someone of ordinary skill in the art before the effective filling date of the claimed invention to include in the teachings of ZAFAR in view of CIMINELLI in view of INAKOSHI the capability to add testing, modification and retesting of preliminary content against previous data, avoiding excessive similarity. The benefit and motivation of such modification is discussed by HAZARD in the following portion: HAZARD [0097] “One or more newly generated synthetic data cases may be tested against at least a portion of the training data. For example, in some embodiments, it may be important that newly generated cases differ from all data in the training data. As such, newly generated data cases may be assessed for similarity with all data cases in the training data. If over-similarity is found, then the synthetic data case may be modified (and retested), discarded, or replaced (and the replacement synthetic data case tested for similarity). In some embodiments, it may only be important that the synthetic data case differ from a portion of the existing training data. As such, the newly generated data case may be tested for similarity to only that portion of the cases. As a more specific example, if portions of the training data have personal information and other portions do not (e.g., the latter may have been synthetically generated, and/or may be part of the public domain), then the test for similarity may only be with respect to the portion of concern—the former portion containing personal information. In some embodiments, determining similarity (or fitness) 160 of the generated or synthetic data case may include determining the k-anonymity of the data case (as discussed elsewhere herein).” Summary of references used as prior art ZAFAR is used as a primary reference because it teaches a framework for removing selected training data influence in the output. CIMINELLI teaches request driven content generation and interpretation of different textual language instructions, such as nouns/adjectives, verbs/adverbs and spatial prepositions. INAKOSHI teaches selective removal of learning data from an image generator while retaining the rest of the learning-data. SAGAR teaches neuron-based analysis, supplies input driven identification of memories to forget and selective masking for removal or suppression of associated neuron, weight values and stored patterns. ZASS teaches convolution-based content analysis and numerical recognition. HAZARD teaches evaluation of multiple generated outputs and modification to identify overly similar preliminary outputs. MADL supplies prohibited regions around original datapoints and generation outside those regions. Pertinent art not relied upon on this rejection Keren; Yonatan et al. (US 20220269980 A1) RANDOM CLASSIFICATION MODEL HEAD FOR IMPROVED GENERALIZATION: Teaches reduce the training sample memorization and limiting sample-specific dependance. PACKALEN; Tom et al. (US 20190026601 A1) METHOD, SYSTEM AND TOOL FOR CONTENT MODERATION: Teaches the use of labels acceptable and unacceptable training content to establish a separation boundary, to identify undesirable content. Conclusion Any inquiry concerning this communication or earlier communications from the examiner should be directed to HECTOR J. CRESPO FEBLES whose telephone number is (571)272-4512. The examiner can normally be reached Mon - Fri 7:30 - 5:00. 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, Daniel Washburn can be reached at (571) 272-5551. 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. /HECTOR J. CRESPO FEBLES/ Examiner, Art Unit 2657 /DANIEL C WASHBURN/ Supervisory Patent Examiner, Art Unit 2657
Read full office action

Prosecution Timeline

Apr 22, 2025
Application Filed
Oct 19, 2026
Non-Final Rejection mailed — §103, §DOUBLEPATENT (current)

Precedent Cases

Applications granted by this same examiner with similar technology

Patent 12770638
SELECTING ISOLATED SPEAKER SIGNAL BY COMPARING TEXT OBTAINED FROM AUDIO AND VIDEO STREAMS
2y 5m to grant Granted Oct 20, 2026
Study what changed to get past this examiner. Based on 1 most recent grants.

Strategy Recommendation AI-generated — please review before filing

Get a prosecution strategy drawn from examiner precedents, rejection analysis, and claim mapping.
Typically takes 5-10 seconds — AI-generated, attorney review required before filing

Prosecution Projections

1-2
Expected OA Rounds
100%
Grant Probability
99%
With Interview (+0.0%)
2y 1m (~7m remaining)
Median Time to Grant
Low
PTA Risk
Based on 1 resolved cases by this examiner. Grant probability derived from career allowance rate.

Sign in with your work email

Enter your email to receive a magic link. No password needed.

Personal email addresses (Gmail, Yahoo, etc.) are not accepted.

Free tier: 3 strategy analyses per month