DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is responsive to the original application filed on 11/28/2022 and the Remarks and Amendments filed on 3/9/2026.
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.
Claims 1-4, 6-11, 13-18, and 20 are rejected under 35 U.S.C. § 103 as being obvious over Norouzi et al. (Norouzi et al., “Exemplar VAE: Linking Generative Models, Nearest Neighbor Retrieval, and Data Augmentation”, Nov. 24, 2020, arXiv:2004.04795v, pp. 1-20, hereinafter “Norouzi”) in view of Li et al. (Li et al., “ON THE LATENT HOLES OF VAES FOR TEXT GENERATION”, Oct. 7, 2021, arXiv:2110.03318v1, pp. 1-18, hereinafter “Li”), and Abdelaal et al. (US 20240070465 A1, hereinafter “Abdelaal”).
Regarding claim 1, Norouzi discloses [a] computer-implemented method for utilizing a variational autoencoder for neighborhood sampling, the method comprising: (Abstract; “We propose retrieval augmented training (RAT) as a way to speed up Exemplar VAEtraining by using approximate nearest neighbor search in the latent space to define a lower bound on log marginal likelihood”)
training said variational autoencoder to generate in-distribution neighborhood samples; (Abstract; “Exemplar VAE is a variant of VAE with a non-parametric prior in the latent space based on a Parzen window estimator. To sample from it, one first draws a random exemplar from a training set, then stochastically transforms that exemplar into a latent code and a new observation”, which discloses the claimed neighborhood sampling in that an exemplar is sampled, a latent code is generated, and a new point is decoded; and Page 2, §3; “The generative process of an Exemplar VAE is summarized in three steps: 1. Sample n Uniform(1N)toobtain a random exemplar xn from the training set, X xn N n=1. 2. Sample z r ( xn)using an exemplar based prior, r , to transform an exemplar xn into a distribution over latent codes, from which z is drawn. 3. Sample x p ( z)usingadecoder to transform z into a distribution over observations, from which x is drawn”, which discloses generating in-distribution neighborhood or exemplar samples; and Page 2; “We propose retrieval augmented training (RAT), using approximate nearest neighbor search in the latent space, to speed up training based on a novel log-likelihood lower bound”; and Page 3, ¶4; “one can train this model with one set of exemplars and perform generation with another, potentially much larger set”)
generating, using said trained variational autoencoder, in-distribution neighborhood samples of an instance of a dataset in a latent space (Abstract; “Exemplar VAE is a variant of VAE with a non-parametric prior in the latent space based on a Parzen window estimator. To sample from it, one first draws a random exemplar from a training set, then stochastically transforms that exemplar into a latent code and a new observation”, which discloses the claimed neighborhood sampling in that an exemplar is sampled, a latent code is generated, and a new point is decoded; and Page 2, §3; “The generative process of an Exemplar VAE is summarized in three steps: 1. Sample n Uniform(1N)toobtain a random exemplar xn from the training set, X xn N n=1. 2. Sample z r ( xn)using an exemplar based prior, r , to transform an exemplar xn into a distribution over latent codes, from which z is drawn. 3. Sample x p ( z)using a decoder to transform z into a distribution over observations, from which x is drawn”, which discloses generating in-distribution neighborhood or exemplar samples of an instance of a dataset in a latent space)
generating a set of interpretable examples for said in-distribution neighborhood samples using a k-nearest neighbors algorithm (Abstract; “We propose retrieval augmented training (RAT) as a way to speed up Exemplar VAEtraining by using approximate nearest neighbor search in the latent space to define a lower bound on log marginal likelihood”, which discloses using approximate K-nearest neighbors retrieval in a latent space to generate interpretable examples; and Page 4, §3.1; “This can be mitigated with fast, approximate nearest neighbor search in the latent space to find a subset of exemplars that exert the maximum influence on the generation of each data point. Interesting, as shown below, the use of approximate nearest neighbor for training Exemplar VAEs is mathematically justified based on a lower bound on the log marginal likelihood”, which discloses generating a set of interpretable examples or exemplars for in-distribution neighborhood samples using a k-nearest neighbors algorithm).
Norouzi fails to explicitly disclose but Li discloses a latent space that satisfies a distortion constraint (Abstract; “we provide the first focused study on the discontinuities (aka. holes) in the latent space of Variational Auto-Encoders (VAEs), a phenomenon which has been shown to have a detrimental effect on model capacity. When investigating latent holes, existing works are exclusively centered around the encoder network and they merely explore the existence of holes. We tackle these limitations by proposing a highly efficient Tree-based Decoder-Centric (TDC) algorithm for latent hole identification, with a focal point on the text domain”, which discloses a latent space that satisfies a distortion constraint by identifying latent holes in a latent space; and Page 2, ¶2; “we provide, for the first time, an in-depth empirical analysis that examines three important aspects: (i) how the holes impact VAE models’ performance on text generation; (ii) whether the holes are really vacant, i.e., useful information is not captured by the holes at all; and (iii) how the holes are distributed in the latent space”; and §2.2; and §3.1; and Algorithm 1; and Page 7, §4.2; “As for the density estimation of latent holes, we utilise the average number of paths traversed before the number of identified holes reaches the algorithm halting threshold Nhole = 200”, which suggests that the distortion constraint is a density estimation for a latent hole in a latent space).
Norouzi and Li are analogous art because all are concerned with the analysis and use of autoencoders. Before the effective filing date of the claimed invention, it would have been obvious to one skilled in autoencoders and machine learning to combine the latent holes/distortion constraints of Li and the generating of neighborhood samples of and method of Norouzi to yield to the predictable result of generating, using said trained variational autoencoder, in-distribution neighborhood samples of an instance of a dataset in a latent space that satisfies a distortion constraint. The motivation for doing so would be to provide a highly efficient tree-based decoder-centric (TDC) algorithm for latent hole identification (Li; Conclusion).
Norouzi fails to explicitly disclose but Abdelaal discloses jointly training an encoder and a decoder to minimize a reconstruction error between an input reconstruction and an original input ([0053]; "The encoder and decoder are trained jointly such that the output minimizes reconstruction error and the KL divergence between the parametric posterior (i.e., distribution of the generated data) and the true posterior (i.e., distribution of the original data)", which discloses, under a broadest reasonable interpretation of the claim language, jointly training an encoder and decoder to minimize a reconstruction error between an input reconstruction or parametric posterior or distribution of generated data and an original input or true posterior or distribution of the original data).
Li, Norouzi, and Abdelaal are analogous art because all are concerned with the use of variational autoencoders. Before the effective filing date of the claimed invention, it would have been obvious to one skilled in autoencoder technology to combine the the latent holes/distortion constraints of Li and the generating of neighborhood samples of and method of Norouzi and the jointly training of the auto encoder and decoder to minimize reconstruction errors between an input reconstruction and an original input of Abdelaal to yield the predictable result of jointly training an encoder and a decoder to minimize a reconstruction error between an input reconstruction and an original input. The motivation for doing so would be to produce more effective ML models without having the need to repair data with errors (Abdelaal; Abstract).
Regarding claim 8, it is a computer program product claim corresponding to the steps of claim 1, and is rejected for the same reasons as claim 1.
Regarding claim 15, it is a system claim corresponding to the steps of claim 1, and is rejected for the same reasons as claim 1.
Regarding claims 2, 9, and 16, the rejection of claims 1, 8, and 15 are incorporated and Norouzi further discloses encoding an input as a distribution over said latent space, wherein said input comprises an instance of said dataset; sampling a point of said distribution from said latent space; decoding said sampled point which corresponds to an input reconstruction, wherein said input reconstruction satisfies a minimum distortion level (Figure 1; and Page 3, §3; and Equation 5).
Regarding claims 3, 10, and 17, the rejection of claims 1, 8, and 15 are incorporated and Norouzi further discloses encoding said instance of said dataset; and computing a mean and a standard deviation of said encoded instance (Page 3, towards the bottom; “The two encoders use the same parametric mean function µφ, but they differ in their covariance functions. The variational posterior uses a data dependent diagonal covariance matrix Λφ, while the exemplar based prior uses an isotropic Gaussian (per exemplar), with a shared, scalar parameter σ 2”).
Regarding claims 4, 11, and 18, the rejection of claims 1, 3, 8, 10, 15, and 17 are incorporated and Norouzi further discloses sampling a set of latent vectors from a Gaussian distribution with said mean and said standard deviation of said encoded instance (Page 3, §3; “The two encoders use the same parametric mean function µφ, but they differ in their covariance functions. The variational posterior uses a data dependent diagonal covariance matrix Λφ, while the exemplar based prior uses an isotropic Gaussian (per exemplar), with a shared, scalar parameter σ 2”; and see generally §3).
Regarding claims 6, 13, and 20, the rejection of claims 1, 8, and 15 are incorporated Norouzi discloses sampling a set of representative examples from said generated in-distribution neighborhood samples in said latent space that belong to a class; computing a set of k-nearest neighbors to said instance of said dataset in said latent space with respect to said set of representative examples; and generating said set of interpretable examples for said in-distribution neighborhood samples from said sampled set of representative examples corresponding to said k-nearest neighbors (Abstract; “We propose retrieval augmented training (RAT) as a way to speed up Exemplar VAEtraining by using approximate nearest neighbor search in the latent space to define a lower bound on log marginal likelihood”, which discloses using approximate K-nearest neighbors retrieval in a latent space to generate interpretable examples; and Page 2, §3; and Page 4, §3.1; “This can be mitigated with fast, approximate nearest neighbor search in the latent space to find a subset of exemplars that exert the maximum influence on the generation of each data point. Interesting, as shown below, the use of approximate nearest neighbor for training Exemplar VAEs is mathematically justified based on a lower bound on the log marginal likelihood”, which discloses generating a set of interpretable examples or exemplars for in-distribution neighborhood samples using a k-nearest neighbors algorithm).
Regarding claims 7 and 14, the rejection of claims 1, 8, and 15 are incorporated and Norouzi further discloses wherein said dataset comprises time-series data or image data (Page 3, Figure 1; and page 6, §5; “gray-scale image data”).
Response to Arguments
Applicant’s arguments and amendments, filed on 3/9/2026, with respect to the 35 USC § 101 rejection of the pending claims have been fully considered and are persuasive. The 35 USC §101 rejection of the pending claims is withdrawn.
Applicant’s arguments and amendments, filed on 3/9/2026, with respect to the 35 USC § 103 rejection of the pending claims have been fully considered but are moot because the arguments do not apply to any of the references used to presently reject the independent claims. Li, Norouzi, and Abdelaal are now being used to render the independent claims obvious under 35 USC § 103.
Conclusion
Claims 5, 12, and 19 have been searched, but no prior art was uncovered.
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Brent Hoover whose telephone number is (303)297-4403. The examiner can normally be reached Monday - Friday 9-5 MST.
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, Abdullah Kawsar can be reached on 571-270-3169. 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.
/BRENT JOHNSTON HOOVER/Primary Examiner, Art Unit 2127