DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA . The Preliminary Amendment filed 9 September 2024 has been entered and considered. Claim 1 has been canceled. Claims 2-21 have been added and are all the claims pending in the application. Claims 2-21 are rejected.
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 2-20 are rejected on the ground of nonstatutory double patenting as being unpatentable over claims 1-19 of U.S. Patent No. 12,106,548 (hereinafter “the ‘548 patent”). Although the claims at issue are not identical, they are not patentably distinct from each other because the patent claims anticipate the claims of the subject application.
As to independent claim 2, claim 1 of the ‘548 patent requires a method for training a generative image model, comprising (“A method for training a generative image model, comprising”): defining a plurality of sensitive categories associated with a plurality of training images (“defining a plurality of sensitive categories associated with a plurality of training images”); defining a plurality of protected attributes associated with the plurality of training images (“defining a plurality of protected attributes associated with the plurality of training images”); calculating for each image in the plurality of training images a corresponding image debiasing weight value associated with at least one protected attribute in the plurality of protected attributes (“calculating for each image in the plurality of training images a corresponding image debiasing weight value associated with the at least one protected attribute”); generating annotated training data comprising the plurality of training images and further comprising, for each image in the plurality of training images, (1) the corresponding image debiasing weight value associated with the at least one protected attribute and (2) a corresponding descriptive text caption (“generating annotated training data comprising the plurality of training images and further comprising, for each image in the plurality of training images, (1) the corresponding image debiasing weight value associated with the at least one protected attribute and (2) a corresponding descriptive text caption”); and performing a training process using the annotated training data to train a generative image model resulting in a trained model, wherein a contribution of each image in the plurality of training images to an optimization loss of the training process is weighted during the training process using the corresponding image debiasing weight value (“performing a training process using the annotated training data to train a generative image model resulting in a trained model, wherein a contribution of each image in the plurality of training images to an optimization loss of the training process is weighted during the training process using the corresponding image debiasing weight value”).
As to claim 3, claim 2 of the ‘548 patent requires providing as an input to the trained model a prompt comprising a description of a particular sensitive category; and receiving as an output from the trained model a plurality of output images associated with the particular sensitive category, wherein the plurality of output images has an unbiased distribution of the at least one protected attribute for the particular sensitive category (“providing as an input to the trained model a prompt comprising a description of the particular sensitive category; and receiving as an output from the trained model a plurality of output images associated with the particular sensitive category, wherein the plurality of output images has an unbiased distribution of the at least one protected attribute for the particular sensitive category”).
As to claim 4, claim 3 of the ‘548 patent requires that each image in the plurality of training images comprises metadata tags describing at least one sensitive category associated with the image (“wherein each image in the plurality of training images comprises metadata tags describing at least one sensitive category associated with the image”).
As to claim 5, claim 4 of the ‘548 patent requires that each training image in the annotated training data comprises the corresponding image debiasing weight value stored as a metadata tag (“wherein each training image in the annotated training data comprises the corresponding image debiasing weight value stored as a metadata tag”).
As to claim 6, claim 5 of the ‘548 patent requires that each training image in the annotated training data comprises the corresponding descriptive text caption stored as a metadata tag (“wherein each training image in the annotated training data comprises the corresponding descriptive text caption stored as a metadata tag”).
As to claim 7, claim 6 requires that calculating each corresponding image debiasing weight value comprises calculating a bias weight value that characterizes a bias associated with a particular sensitive category and the at least one protected attribute (“wherein calculating each corresponding image debiasing weight value based on the distribution comprises calculating from the distribution a bias weight value that characterizes a bias associated with the particular sensitive category and the at least one protected attribute”).
As to claim 8, claim 7 requires that a particular image in the plurality of training images is characterized by a particular protected attribute, the method further comprising: calculating a particular debiasing weight value that balances a particular bias associated with a particular sensitive category and the particular protected attribute; and calculating for the particular image the corresponding image debiasing weight value based on the particular debiasing weight value (“wherein a particular image in the plurality of training images is characterized by a particular protected attribute, the method further comprising: calculating from the distribution a particular debiasing weight value that balances a particular bias associated with the particular sensitive category and the particular protected attribute; and calculating for the particular image the corresponding image debiasing weight value based on the particular debiasing weight value”).
As to claim 9, claim 8 of the ‘548 patent requires that the particular protected attribute is a first protected attribute, the particular debiasing weight value is a first debiasing weight value, the particular bias is a first bias, and the particular image is further characterized by a second protected attribute, the method further comprising: calculating a second debiasing weight value that balances a second bias associated with the particular sensitive category and the second protected attribute; and calculating the corresponding image debiasing weight value by averaging the first debiasing weight value and the second debiasing weight value (“wherein the particular protected attribute is a first protected attribute, the particular debiasing weight value is a first debiasing weight value, the particular bias is a first bias, and the particular image is further characterized by a second protected attribute, the method further comprising: calculating from the distribution a second debiasing weight value that balances a second bias associated with the particular sensitive category and the second protected attribute; calculating for the particular image a corresponding second debiasing weight value based on the debiasing weight value; and calculating the corresponding image debiasing weight value by averaging the first debiasing weight value and the second debiasing weight value”).
As to claim 10, claim 9 of the ‘548 patent requires that the plurality of sensitive categories comprise a plurality of occupations (“wherein the plurality of sensitive categories comprise a plurality of occupations”).
As to claim 11, claim 10 of the ‘548 patent requires that the plurality of sensitive categories comprise a plurality of income levels (“wherein the plurality of sensitive categories comprise a plurality of income levels”).
As to claim 12, claim 11 of the ‘548 patent requires that the plurality of protected attributes comprise gender, skin color, ethnicity, race, age, and religion (“wherein the plurality of protected attributes comprise gender, skin color, ethnicity, race, age, and religion”).
As to claim 13, claim 12 of the ‘548 patent requires that the generative image model is one of a Generative Adversarial Network (GAN), a Variational Autoencoder (VAE), an autoregressive model, a diffusion model, and a transformer-based architecture (“wherein the generative image model is one of a Generative Adversarial Network (GAN), a Variational Autoencoder (VAE), an autoregressive model, a diffusion model, and a transformer-based architecture”).
As to independent claim 14, claim 13 of the ‘548 patent requires a non-transitory computer-readable medium storing a program for training a generative image model, which when executed by a computer, configures the computer to (“A non-transitory computer-readable medium storing a program for training a generative image model, which when executed by a computer, configures the computer to”): define a plurality of sensitive categories associated with a plurality of training images (“define a plurality of sensitive categories associated with a plurality of training images”); define a plurality of protected attributes associated with the plurality of training images (“define a plurality of protected attributes associated with the plurality of training images”); calculate for each image in the plurality of training images a corresponding image debiasing weight value associated with at least one protected attribute in the plurality of protected attributes (“calculate for each image in the plurality of training images a corresponding image debiasing weight value associated with the at least one protected attribute”); generate annotated training data comprising the plurality of training images and further comprising, for each image in the plurality of training images, (1) the corresponding image debiasing weight value associated with the at least one protected attribute and (2) a corresponding descriptive text caption (“generate annotated training data comprising the plurality of training images and further comprising, for each image in the plurality of training images, (1) the corresponding image debiasing weight value associated with the at least one protected attribute and (2) a corresponding descriptive text caption”); and perform a training process using the annotated training data to train a generative image model resulting in a trained model, wherein a contribution of each image in the plurality of training images to an optimization loss of the training process is weighted during the training process using the corresponding image debiasing weight value (“perform a training process using the annotated training data to train a generative image model resulting in a trained model, wherein a contribution of each image in the plurality of training images to an optimization loss of the training process is weighted during the training process using the corresponding image debiasing weight value”).
As to claim 15, claim 14 of the ‘548 patent requires that the program, when executed by the computer, further configures the computer to: provide as an input to the trained model a prompt comprising a description of a particular sensitive category; and receive as an output from the trained model a plurality of output images associated with the particular sensitive category, wherein the plurality of output images has an unbiased distribution of the at least one protected attribute for the particular sensitive category (“the program, when executed by the computer, further configures the computer to: provide as an input to the trained model a prompt comprising a description of the particular sensitive category; and receive as an output from the trained model a plurality of output images associated with the particular sensitive category, wherein the plurality of output images has an unbiased distribution of the at least one protected attribute for the particular sensitive category”).
As to claim 16, claim 15 of the ‘548 patent requires that calculating each corresponding image debiasing weight value comprises calculating a bias weight value that characterizes a bias associated with a particular sensitive category and the at least one protected attribute (“wherein calculating each corresponding image debiasing weight value based on the distribution comprises calculating from the distribution a bias weight value that characterizes a bias associated with the particular sensitive category and the at least one protected attribute”).
As to claim 17, claim 16 of the ‘548 patent requires that a particular image in the plurality of training images is characterized by a particular protected attribute, and the program, when executed by the computer, further configures the computer to: calculate a particular debiasing weight value that balances a particular bias associated with a particular sensitive category and the particular protected attribute; and calculate for the particular image the corresponding image debiasing weight value based on the particular debiasing weight value (“wherein a particular image in the plurality of training images is characterized by a particular protected attribute, and the program, when executed by the computer, further configures the computer to: calculate from the distribution a particular debiasing weight value that balances a particular bias associated with the particular sensitive category and the particular protected attribute; and calculate for the particular image the corresponding image debiasing weight value based on the particular debiasing weight value”).
As to claim 18, claim 17 of the ‘548 patent requires that the particular protected attribute is a first protected attribute, the particular debiasing weight value is a first debiasing weight value, the particular bias is a first bias, and the particular image is further characterized by a second protected attribute, and the program, when executed by the computer, further configures the computer to: calculate a second debiasing weight value that balances a second bias associated with the particular sensitive category and the second protected attribute; and calculate the corresponding image debiasing weight value by averaging the first debiasing weight value and the second debiasing weight value (“wherein the particular protected attribute is a first protected attribute, the particular debiasing weight value is a first debiasing weight value, the particular bias is a first bias, and the particular image is further characterized by a second protected attribute, and the program, when executed by the computer, further configures the computer to: calculate from the distribution a second debiasing weight value that balances a second bias associated with the particular sensitive category and the second protected attribute; calculate for the particular image a corresponding second debiasing weight value based on the debiasing weight value; and calculate the corresponding image debiasing weight value by averaging the first debiasing weight value and the second debiasing weight value”).
As to claim 19, claim 18 of the ‘548 patent requires that the plurality of sensitive categories comprise one of a plurality of occupations and a plurality of income levels, and the plurality of protected attributes comprise gender, skin color, ethnicity, race, age, and religion (“wherein the plurality of sensitive categories comprise one of a plurality of occupations and a plurality of income levels, and the plurality of protected attributes comprise gender, skin color, ethnicity, race, age, and religion”).
As to claim 20, claim 19 of the ‘548 patent requires that the generative image model is one of a Generative Adversarial Network (GAN), a Variational Autoencoder (VAE), an autoregressive model, a diffusion model, and a transformer-based architecture (“wherein the generative image model is one of a Generative Adversarial Network (GAN), a Variational Autoencoder (VAE), an autoregressive model, a diffusion model, and a transformer-based architecture”).
Claim 21 is rejected on the ground of nonstatutory double patenting as being unpatentable over claims 20 of U.S. Patent No. 12,106,548 (hereinafter “the ‘548 patent”) in view of U.S. Patent Application Publication No. 2023/0094954 to Sinha et al. (hereinafter “Sinha”).
As to claim 21, claim 20 of the ‘548 patent requires A system for training a generative image model, comprising: a processor; and a non-transitory computer readable medium storing a set of instructions, which when executed by the processor, configure the processor to (“ A system for training a generative image model, comprising: a processor; and a non-transitory computer readable medium storing a set of instructions, which when executed by the processor, configure the processor to”): define a plurality of sensitive categories associated with a plurality of training images; define a plurality of protected attributes associated with the plurality of training images (“define a plurality of sensitive categories associated with a plurality of training images; define a plurality of protected attributes associated with the plurality of training images”); calculate for each image in the plurality of training images a corresponding image debiasing weight value associated with at least one protected attribute in the plurality of protected attributes (“calculate for each image in the plurality of training images a corresponding image debiasing weight value associated with the at least one protected attribute”); generate annotated training data comprising the plurality of training images and further comprising, for each image in the plurality of training images, (1) the corresponding image debiasing weight value associated with the at least one protected attribute and (2) a corresponding descriptive text caption (“generate annotated training data comprising the plurality of training images and further comprising, for each image in the plurality of training images, (1) the corresponding image debiasing weight value associated with the at least one protected attribute and (2) a corresponding descriptive text caption”); perform a training process using the annotated training data to train a generative image model resulting in a trained model, wherein a contribution of each image in the plurality of training images to an optimization loss of the training process is weighted during the training process using the corresponding image debiasing weight value (“perform a training process using the annotated training data to train a generative image model resulting in a trained model, wherein a contribution of each image in the plurality of training images to an optimization loss of the training process is weighted during the training process using the corresponding image debiasing weight value”).
The claims of the ‘548 patent do not expressly require, within the context of claim 20 of the ‘548 patent, provide as an input to the trained model a prompt comprising a description of a particular sensitive category; and receive as an output from the trained model a plurality of output images associated with the particular sensitive category, wherein the plurality of output images has an unbiased distribution of the at least one protected attribute for the particular sensitive category.
Sinha, like the claims of the ‘548 patent, is directed to “an image-generating application that generates simulated images that enhance socio-demographic diversity”, wherein the image-generating application applies “a machine-learning model” which is trained to output the generated images ([0005] and Figs. 1 and 11). Sinha discloses that a user device submits a query with a text description of a particular occupation in order to retrieve a set of “socially-diverse images” generated by the trained model, wherein the results mitigate bias between the sensitive category (e.g., occupation – plumber) and the attribute (e.g., gender – female) (Figs. 1, 6, 9, and 12 and [0039-0044, 0082-0090]).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the claims of the ‘548 patent to provide an image-generation application which receives an input prompt and outputs from a trained model a plurality of images that reduce bias, as taught by Sinha, to arrive at the claimed invention discussed above. Such a modification is the result of combining prior art elements according to known methods to yield predictable results. It is predictable that the proposed modification would have reduced bias in the generative model, as desired by Sinha ([0083]).
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.
Claims 2-8, 10-17, and 19-21 are rejected under 35 U.S.C. 103 as being unpatentable over U.S. Patent Application Publication No. 2023/0094954 to Sinha et al. (hereinafter “Sinha”) in view of “Bias Correction of Learned Generative Models using Likelihood-Free Importance Weighting” by Grover et al. (cited in the IDS filed 11/27/24; hereinafter “Grover”).
As to independent claim 2, Sinha discloses a method for training a generative image model ([0005] and Figs. 1 and 11 discloses that Sinha is directed to “an image-generating application that generates simulated images that enhance socio-demographic diversity”, wherein the image-generating application applies “a machine-learning model” which is trained to output the generated images), comprising: defining a plurality of sensitive categories associated with a plurality of training images ([0039-0058] discloses “various occupations” associated with a plurality of stored images); defining a plurality of protected attributes associated with the plurality of training images ([0039-0058] discloses socio-demographic attributes including gender, age, and race associated with the set of training images); calculating for each image in the plurality of training images a corresponding image debiasing value associated with at least one protected attribute in the plurality of protected attributes; generating annotated training data comprising the plurality of training images and further comprising, for each image in the plurality of training images, (1) the corresponding image debiasing value associated with the at least one protected attribute and (2) a corresponding descriptive text caption; and performing a training process using the annotated training data to train a generative image model resulting in a trained model ([0043, 0071-0081] discloses generating a synthetically balanced set of training images for training the generative model by augmenting each real training image in the training set of under-represented socio-demographic attributes and labeling that generated image accordingly (e.g., “female plumber”); by generating a transformed counterfactual image to augment the training dataset, bias in the finally-trained model is reduced).
Sinha discloses that the synthetically balanced set of training images generated for training the generative image model includes augmented images generated from real training images in the training set ([0043, 0071-0081]). Thus, Sinha does not expressly disclose that the debiasing value (i.e., the augmented training image) is a weight. Further, although Sinha discloses that the loss function used to train the generative model is weighted ([0116-0128]), Sinha does not expressly disclose that the weights are applied on a per-training-image basis. That is, Sinha does not expressly disclose that a contribution of each image in the plurality of training images to an optimization loss of the training process is weighted during the training process using the corresponding image debiasing weight value.
Grover, like Sinha, is directed to correcting the bias of a trained generative image model (Abstract). Grover contemplates the solution disclosed by Sinha in which “various combinations of the real training data Dcl and generated training data Dg” are used to debias the model, but notes that “weighting the generated points” with the novel LFIW importance weighting algorithm yields “significant improvements” (Section 5.2). Specifically, Grover discloses “reweight[ing] each sample” xi with an “importance weight” w in order to “reduce the bias of a generative” model when training on “loss l” (Sections 2-3 and equations 2-3).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify Sinha to not only augment the training set with synthetically generated training images, but to also weight each training image to variably contribute to the loss function during training of the generative image model, as taught by Grover, to arrive at the claimed invention discussed above. Such a modification is the result of combining prior art elements according to known methods to yield predictable results. It is predictable that the proposed modification would have “obtain[ed] significant improvements” in bias correction over Sinha’s augmentation of the training image set alone (Section 5.2 of Grover).
As to claim 3, Sinha as modified above further teaches providing as an input to the trained model a prompt comprising a description of a particular sensitive category; and receiving as an output from the trained model a plurality of output images associated with the particular sensitive category, wherein the plurality of output images has an unbiased distribution of the at least one protected attribute for the particular sensitive category (Figs. 1, 6, 9, and 12 and [0039-0044, 0082-0090] of Sinha discloses that a user device submits a query with a text description of a particular occupation in order to retrieve a set of “socially-diverse images” generated by the trained model, wherein the results mitigate bias between the sensitive category (e.g., occupation – plumber) and the attribute (e.g., gender – female)).
As to claim 4, Sinha as modified above further teaches that each image in the plurality of training images comprises metadata tags describing at least one sensitive category associated with the image ([0082, 0094-0103] discloses that each training image is labeled with its ground truth socio-demographic attributes, wherein such ground truth labels are metadata).
As to claim 5, Sinha as modified by Grover further teaches that each training image in the annotated training data comprises the corresponding image debiasing weight value stored as a metadata tag (Sections 2-3 and equations 2-3 of Grover discloses “reweight[ing] each sample” xi with an “importance weight” w in order to “reduce the bias of a generative” model when training on “loss l”, wherein the importance weight must be stored and constitutes metadata since it describes information about the image; the reasons for combining the references are the same as those discussed above in conjunction with claim 2).
As to claim 6, Sinha as modified above further teaches that each training image in the annotated training data comprises the corresponding descriptive text caption stored as a metadata tag ([0082, 0094-0103] discloses that each training image is labeled with its textual ground truth socio-demographic attributes, wherein such ground truth labels are metadata).
As to claim 7, Sinha as modified above further teaches that calculating each corresponding image debiasing weight value comprises calculating a bias weight value that characterizes a bias associated with a particular sensitive category and the at least one protected attribute ([0116-0129] of Sinha discloses that scores for different classes are weighted to handle the imbalance of classes in the training dataset, for example by generating a score for generated “combination of race, gender, and occupation”; Sections 2-3 and equations 2-3 of Grover discloses “reweight[ing] each sample” xi with an “importance weight” w in order to “reduce the bias of a generative” model when training on “loss l”; the reasons for combining the references are the same as those discussed above in conjunction with claim 2).
As to claim 8, Sinha as modified above further teaches that a particular image in the plurality of training images is characterized by a particular protected attribute ([0039-0058] of Sinha discloses “various occupations” and various socio-demographic attributes including gender, age, and race associated with each training image), the method further comprising: calculating a particular debiasing weight value that balances a particular bias associated with a particular sensitive category and the particular protected attribute; and calculating for the particular image the corresponding image debiasing weight value based on the particular debiasing weight value ([0116-0129] of Sinha discloses that scores for different classes are weighted to handle the imbalance of classes in the training dataset, for example by generating a score for generated “combination of race, gender, and occupation”; Sections 2-3 and equations 2-3 of Grover discloses “reweight[ing] each sample” xi with an “importance weight” w in order to “reduce the bias of a generative” model when training on “loss l”; the reasons for combining the references are the same as those discussed above in conjunction with claim 2).
As to claim 10, Sinha as modified above further teaches that the plurality of sensitive categories comprise a plurality of occupations ([0039-0058] discloses “various occupations” associated with a plurality of stored images).
As to claim 11, Sinha as modified above further teaches that the plurality of sensitive categories comprise a plurality of income levels ([0039-0058] discloses “various occupations” associated with a plurality of stored images, wherein different occupations are necessarily associated with different income levels).
As to claim 12, Sinha as modified above further teaches that the plurality of protected attributes comprise gender, race, age ([0027]). Sinha as modified above does not expressly disclose that the protected attributes also comprise skin color, ethnicity, and religion. However, official notice is taken to note that the uses and benefits of labeling images according to the claimed attributes are known and expected within the image processing arts. It would have been obvious to the ordinarily-skilled artisan at the time of invention to include among the attributes skin color, ethnicity, and religion within the system of Sinha, to achieve the known and expected uses and benefits of providing a generative model that is more robust to a variety of protected classes.
As to claim 13, Sinha as modified above further teaches that the generative image model is one of a Generative Adversarial Network (GAN), a Variational Autoencoder (VAE), an autoregressive model, a diffusion model, and a transformer-based architecture ([0028] discloses that the model may be a GAN).
Independent claim 14 recites a non-transitory computer-readable medium storing a program for training a generative image model, which when executed by a computer, configures the computer ([0045] of Sinha discloses “a non-transitory computer-readable medium, is executed by one or more processing devices to cause a server system to perform one or more operations described herein”) to perform the method recited in claim 2. Accordingly, claim 14 is rejected for reasons analogous to those discussed above in conjunction with claim 2.
Claims 15-17 and 19-20 recite features nearly identical to those recited in claims 3, 7-8, 10+12, and 13, respectively. Accordingly, claims 15-17 and 19-20 are rejected for reasons analogous to those discussed above in conjunction with claims 3, 7-8, 10+12, and 13, respectively.
Independent claim 21 recites a system for training a generative image model, comprising: a processor; and a non-transitory computer readable medium storing a set of instructions, which when executed by the processor, configure the processor ([0045] of Sinha discloses “a non-transitory computer-readable medium, is executed by one or more processing devices to cause a server system to perform one or more operations described herein”) to perform the method recited in claims 2+3. Accordingly, claim 21 is rejected for reasons analogous to those discussed above in conjunction with claims 2 and 3.
Allowable Subject Matter
Claims 9 and 18 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims and if the double patenting rejections were overcome.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SEAN M CONNER whose telephone number is (571)272-1486. The examiner can normally be reached 10 AM - 6 PM Monday through Friday, and some Saturday afternoons.
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, Greg Morse can be reached at (571) 272-3838. 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.
/SEAN M CONNER/Primary Examiner, Art Unit 2663
Prosecution Insights