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 .
Response to Amendment and Arguments
Applicant’s amendment filed on July 25, 2025 has been entered and made of record. Claims 1-20 are pending and are being examined in this application.
In light of Applicant’s amendments to the claims, the 101 rejection is withdrawn.
Applicant’s arguments with respect to the 102 and 103 rejections have been considered, but are moot in view of the new ground(s) of rejection.
Allowable Subject Matter
Claims 8-11 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.
Claim Rejections - 35 USC § 103
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 1-3, 6, 13, 14, 17, 19, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (US Pub. 20160155136) in view of Yuan et al. (US Pub. 20220036890).
Referring to claim 1, Zhang discloses A system of machine learning architecture for out-of-distribution data set detection comprising: a processor; a memory coupled to the processor and storing processor-executable instructions that, when executed, configure the processor [pars. 55 and 56; features are implemented in one or more computer programs executable by at least one programmable processor coupled to a machine-readable medium] to:
receive an input data set [par. 36; an auto-encoder receives input];
...generate, via the autoencoder, an out-of-distribution prediction [pars. 36 and 37; outlier behaviors are predicted as departures from normality (i.e., the input is provided to the auto-encoder, and a reconstruction error in the auto-encoder is used to indicate deviation from the input)], the auto-encoder trained to encode a ... meaning of in-distribution data within one or more training data sets by performing a pretext task, the pretext task including [pars. 30, 36, and 38-40; the auto-encoder is trained to optimize a latent representation of the input so that the latent representation is sufficient to prevent loss of meaningful features]:
applying a transformation function to the one or more training data sets [pars. 30 and 36; the auto-encoder is trained by applying a logistic function with encoding weights and bias];
encoding the transformed one or more training data sets into a latent representation [pars. 30 and 36; the input is encoded into the latent representation]; and
reconstructing the latent representation of the transformed one or more training data sets [pars. 30 and 36; the latent representation of the input is reconstructed],
the trained auto-encoder trained for reducing a reconstruction error between the one or more training data sets and the reconstructed transformed one or more training data sets thereby learning to encode the ... meaning of the in-distribution data of the one or more training data sets [pars. 38-40; the auto-encoder learns (i.e., is trained) to minimize a loss function L (i.e., the reconstruction error) and optimize the latent representation of the input so that the latent representation is sufficient to prevent loss of meaningful features]; and
generate a signal for providing an indication of whether the input data set is an out-of-distribution data set [par. 42; when the reconstruction error reaches a threshold level, a signal is generated to indicate that production data (i.e., training data) and a production model (i.e., a model trained with the training data) are no longer suitable].
Zhang does not appear to explicitly disclose determine, via an autoencoder, a semantic representation of the input data set; that the out-of-distribution prediction is generated based at least on the semantic meaning of the input data set; and that the auto-encoder is trained to encode a semantic meaning of the in-distribution data.
However, Yuan discloses determine, via an autoencoder, a semantic representation of the input data set; that the out-of-distribution prediction is generated based at least on the semantic meaning of the input data set; and that the auto-encoder is trained to encode a semantic meaning of the in-distribution data by performing a pretext task [figs. 2, 3, and 23; pars. 184-187; pre-training is performed on the input data to output a semantic representation for use by an OOD model (e.g., to perform a OOD/IND classification task)].
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 autoencoder taught by Zhang so that the latent representation is a semantic representation as taught by Yuan, with a reasonable expectation of success. The motivation for doing so would have been to reject background noise and small talk of another person (i.e., provide semantic understanding) in order to improve the user experience [Yuan, par. 3].
Referring to claim 2, Zhang discloses The system of claim 1, wherein the processor-executable instructions, when executed, configure the processor to: identify one or more data values of the input data set as being out-of-distribution by a threshold amount [par. 42; note the threshold level]; generate an updated training data set including the identified one or more out-of-distribution data values; and providing the training data set for training the auto-encoder based on the updated training data set [pars. 42, 46, 50, and 54; one or more portions of the production data may be replaced, changed, or treated differently].
Referring to claim 3, Zhang discloses The system of claim 1, the transformation includes a set of transformations: [SYMBOLIC REPRESENTATION OF A SET OF TRANFORMATIONS] configured to transform a training data set into an alternate data set representation while preserving the semantic meaning for encoding [pars. 30, 36, and 40; note the encoding of the input into the latent representation that is sufficient to prevent loss of meaningful features].
Referring to claim 6, Zhang discloses The system of claim 1, wherein the auto-encoder includes a decoder network having removed fully-connected layers for minimizing expressive properties of the decoder network to provide a regularized asymmetric auto-encoder [pars. 30-33; the auto-encoder contains a decoder that is provided with the encoding (i.e., the latent representation) from the encoder; the latent representation is a lower dimension (i.e., not fully-connected) relative to the input, which means that the auto-encoder is asymmetric; the latent representation also results in the discovery of latent variables that capture the complex interaction of the input, which means that the auto-encoder is regularized].
Referring to claim 10, Zhang discloses The system of claim 1, wherein the auto-encoder includes scoring operations based on a linear classifier trained by learned latent representation of in-distribution and out-of-distribution data sets [pars. 30-35; the auto-encoder computes the reconstruction error based on linear patterns/nodes included in the latent representation].
Referring to claim 13, see the rejection for claim 1, which incorporates the claimed method.
Referring to claim 14, see the rejection for claim 2.
Referring to claim 17, see the rejection for claim 6.
Referring to claim 19, see the rejection for claim 10.
Referring to claim 20, see at least the rejection for claim 1. Zhang further discloses A non-transitory computer-readable medium having stored thereon machine interpretable instructions or data representing the auto-encoder [pars. 55 and 56; features are implemented in one or more computer programs executable by at least one programmable processor coupled to a machine-readable medium].
Claims 4, 5, 15, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang and Yuan in view of Goodwin et al. (US Pub. 20240095983).
Referring to claim 4, Zhang and Yuan do not appear to explicitly disclose The system of claim 3, wherein the set of transformations includes at least one of rotation transformation, segmentation transformation, image data warping operations, or chromatic aberration transformations of a spatial data set.
However, Goodwin discloses The system of claim 3, wherein the set of transformations includes at least one of rotation transformation, segmentation transformation, image data warping operations, or chromatic aberration transformations of a spatial data set [pars. 70, 74, 88, 89, 94; an encoder is trained using image transformations such as dividing (i.e., segmenting) into corresponding areas, rotating, resizing, scaling, stretching, or skewing].
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 autoencoder taught by the combination of Zhang and Yuan so that the autoencoder is trained using image transformations as taught by Goodwin, with a reasonable expectation of success. The motivation for doing so would have been to assist in generating training data for deep learning models [Goodwin, par. 5].
Referring to claim 5, Zhang and Yuan do not appear to explicitly disclose The system of claim 1, wherein the auto-encoder is trained based on a Gram matrix of the training data set, the Gram matrix associated with a base pooling operation for identifying structural correlations among data values of the training data set, the base pooling operation to reduce data set dimensions.
However, Goodwin discloses The system of claim 1, wherein the auto-encoder is trained based on a Gram matrix of the training data set, the Gram matrix associated with a base pooling operation for identifying structural correlations among data values of the training data set, the base pooling operation to reduce data set dimensions [pars. 120-127; an encoder is trained based on computing a loss by taking the difference in Gram matrix values of three maximum pooling layers for original and reconstructed images; a maximum pooling layer, by definition, downsamples a feature map by selecting the maximum value within each region, reducing spatial dimensions].
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 autoencoder taught by the combination of Zhang and Yuan so that the autoencoder is trained using a Gram matrix as taught by Goodwin, with a reasonable expectation of success. The motivation for doing so would have been to assist in generating training data for deep learning models [Goodwin, pars. 5 and 120].
Referring to claim 15, see the rejection for claim 4, which incorporates the claimed method.
Referring to claim 16, see the rejection for claim 5, which incorporates the claimed method.
Claims 7 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang and Yuan in view of Honkala (US Pub. 20220027709).
Referring to claim 7, Zhang and Yuan do not appear to explicitly disclose The system of claim 6, wherein the auto-encoder includes an encoder network based on a DenseNet architecture.
However, Honkala discloses The system of claim 6, wherein the auto-encoder includes an encoder network based on a DenseNet architecture [par. 39 and 54; an encoder and/or decoder is implemented based on DenseNet].
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 autoencoder taught by the combination of Zhang and Yuan so that the encoder and/or decoder is based on DenseNet as taught by Honkala, with a reasonable expectation of success. The motivation for doing so would have been to minimize loss by allowing direct connection of features between layers [Honkala, pars. 31 and 39].
Referring to claim 12, Zhang and Yuan do not appear to explicitly disclose The system of claim 1, wherein the auto-encoder is based on a Wasserstein Auto-encoder for out-of-distribution detection.
However, Honkala discloses The system of claim 1, wherein the auto-encoder is based on a Wasserstein Auto-encoder for out-of-distribution detection [par. 49; the auto-encoder uses a Wasserstein GAN to determine a probability that an input is a real noisy data sample vs. a fake (i.e., OOD) noisy data sample].
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 autoencoder taught by the combination of Zhang and Yuan so that the auto-encoder uses a Wasserstein GAN as taught by Honkala, with a reasonable expectation of success. The motivation for doing so would have been to optimize the output of the auto-encoder by adjusting the model based on OOD predictions [Honkala, par. 40].
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
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.
Contact Information
Any inquiry concerning this communication or earlier communications from the examiner should be directed to GRACE PARK whose telephone number is (571)270-7727. The examiner can normally be reached M-F 8AM-5PM.
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, TAMARA KYLE can be reached at (571)272-4241. 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.
/Grace Park/Primary Examiner, Art Unit 2144