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
The amendment filed on 08-SEPTEMBER-2026 in response to the non-final office action mailed 10-JULY-2026 has been entered. Claims 1-20 remain pending in the application.
With regards to the 103 rejections, the applicant’s amendments to the claims have been considered. Examiner has introduced new prior art WONG in response to the amendments.
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, 3, 7-8, 10, 13-15, 17, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over LAI et al (U.S. Pub. No. U.S. 20230059367 A1) in view of Zhang (U.S. Pub No. US 20220147818 A1) in view of WONG (J.P. Pub No. JP 6737487 B1)
Regarding claim 1, LAI substantially teaches the claimed invention, including:
A computer implemented method performed by a computer system having a memory and at least one hardware processor, the computer implemented method comprising: Obtaining a privacy loss parameter of differential privacy; ([0085] As further shown in the algorithm, the user-entity differential privacy system 106 utilizes the moments accountant to determine the T training steps privacy budget consumption (lines 14-15). In some embodiments, use of the moments accountant facilitates user-entity differential privacy by bounding the total privacy loss of T steps of the Gaussian mechanism) Training a neural network to perform data obfuscation operation, ([0085] As further shown in the algorithm, the user-entity differential privacy system 106 utilizes the moments accountant [AltContent: rect] to determine the T training steps privacy budget consumption (lines 14-15). In some embodiments, use of the moments accountant facilitates user-entity differential privacy by bounding the total privacy loss of T steps of the Gaussian mechanism with the noise [AltContent: rect](0, Iσ.sup.2). In other words, given the bounded sensitivity of the estimator f.sub.ε, the user-entity differential privacy system 106 uses the moments accountant [AltContent: rect] to obtain a tight bound on the total privacy consumption of T steps of the Gaussian mechanism. Thus, the user-entity differential privacy system 106 provides a user-entity differential privacy (UeDP) guarantee. In some cases, the user-entity differential privacy system 106 determines that, for the estimator f.sub.ε, the moments accountant [AltContent: rect] of the sampled Gaussian mechanism correctly computes the UeDP privacy loss with the scale z=σ/)
While LAI does teach generating a model to perform data obfuscation, it does not explicitly teach:
the training of the neural network comprising training the neural network using training data to automatically and dynamically learn a variance parameter that obeys a privacy loss parameters, with an objective of achieving a value of a privacy loss parameter;
However, in analogous art that similarly teaches data obfuscation, WONG teaches:
the training of the neural network comprising training the neural network using training data to automatically and dynamically learn a variance parameter that obeys a privacy loss parameters, with an objective of achieving a value of a privacy loss parameter;0 (((3) The auto encoder learns an image pattern of a medical image having no abnormality by repeating the process while changing the value of the parameter used for the calculation until the value of the predefined loss function becomes equal to or less than the predetermined threshold value. I chose As a result, it is possible to improve the learning accuracy of the auto encoder and output a medical image in a normal state as a restored image even when various images are input.
(4) The variance estimator learns the variance of a medical image having no abnormality by repeating the process while changing the value of the parameter used for the calculation until the value of the predefined loss function becomes equal to or less than the predetermined threshold value. I chose As a result, it is possible to improve the learning accuracy of the variance estimator and accurately estimate the variance of a medical image having no abnormality even when various images are input.)
It would have been obvious to a person skilled in the art before the effective filing date of the
invention to have combined with WONG‘s parameter learning method and, with LAI‘s training of a model that obfuscates, with a reasonable expectation of success, a method that learns a parameter, as in WONG, for a model trained to obfuscate data, as found in LAI. A person of ordinary skill would have been motivated to improve accuracy. (WONG page 3, paragraph 2).
LAI further teaches:
Using the trained neural network to perform data obfuscation by: (In other words, given the bounded sensitivity of the estimator f.sub.ε, the user-entity differential privacy system 106 uses the moments accountant [AltContent: rect] to obtain a tight bound on the total privacy consumption of T steps of the Gaussian mechanism. Thus, the user-entity differential privacy system 106 provides a user-entity differential privacy (UeDP) guarantee. In some cases, the user-entity differential privacy system 106 determines that, for the estimator f.sub.ε, the moments accountant [AltContent: rect] of the sampled Gaussian mechanism correctly computes the UeDP privacy loss with the scale z=σ/[AltContent: rect](f.sub.ε) for f.sub.ε for T training steps. In some instances, the user-entity differential privacy system 106 employs a moments accountant as described in Abadi, M. et al., Deep Learning with Differential Privacy, in ACM SIGSAC Conference on Computer and Communications Security, pp. 308-18, 2016, which is incorporated herein by reference in its entirety.
[0086] Thus, the user-entity differential privacy system 106 generates a natural language model that provides user-entity differential privacy to simultaneously protect the users and the sensitive entities represented by the underlying data)
While LAI does teach generating a model to perform data obfuscation, it does not explicitly teach:
Encoding input data different from the training data into a latent space representation of the input data, the encoding of the input data comprising inferring latent space parameters of a latent space probability distribution based on the input data and sampling data from the latent space distribution, the latent probability space distribution being based on the variance parameter; and Decoding the sampled data of the latent space representation into output data.
However, in analogous art that similarly obfuscates data, ZHANG teaches:
Encoding input data different from the training data into a latent space representation of the input data, (ZHANG P[0144] ” The encoder 208q is arranged to receive the observed feature vector Xo as an input and encode it into a latent vector Z (a representation in a latent space).”), the encoding of the input data comprising inferring latent space parameters of a latent space probability distribution based on the input data and sampling data from the latent space distribution, (ZHANG P[0170-0171] “In some embodiments each of the latent representations Z, is one-dimensional, i.e. consists of only a single latent variable ( element). Note however this does not imply the latent variable Z, is a modelled only as simple, fixed scalar value. Rather, as the auto-encoder is a variational auto-encoder, then for each latent variable Z, the encoder learns a statistical or probabilistic distribution, and the value input to the decoder is a random sample from the distribution. This means that for each individual element of latent space, the encoder learns one or more parameters of the respective distribution, e.g. a measure of centre point and spread of the distribution. For instance each latent variable Z, (a single dimension) may be modelled in the encoder by a respective mean value and standard deviation or variance. However preferably each of the latent space representations Z, is multi-dimensional, in which case each dimension is modelled by one or more parameters of a respective distribution.”), the latent probability space distribution being based on the variance parameter; and (ZHANG P[0170] “This means that for each individual element of latent space, the encoder learns one or more parameters of the respective distribution, e.g. a measure of centre point and spread of the distribution. For instance each latent variable Z, (a single dimension) may be modelled in the encoder by a respective mean value and standard deviation or variance.”);
Decoding the sampled data of the latent space representation into output data. (ZHANG “The encoder 208q is arranged to receive the observed feature vector Xo as an input and encode it into a latent vector Z (a representation in a latent space). The decoder 208p is arranged to receive the latent vector Z and decode back to the original feature space of the feature vector. The version of the feature vector output by the decoder 208p may be labelled herein X.”).
It would have been obvious to a person skilled in the art before the effective filing date of the
invention to have combined with ZHANG‘s obfuscation method and, with LAI‘s, as modified by WONG, training of a model that obfuscates, with a reasonable expectation of success, a method that obfuscates data, as in ZHANG, using a model trained to obfuscate data, as found in LAI, as modified by WONG. A person of ordinary skill would have been motivated to reduce processing time. (ZHANG: P[0126]).
Regarding claim 3, ZHANG further teaches:
The neural network comprises a variational autoencoder. (ZHANG P[0170] “In some embodiments each of the latent representations Z, is one-dimensional, i.e. consists of only a single latent variable ( element). Note however this does not imply the latent variable Z, is a modelled only as simple, fixed scalar value. Rather, as the auto-encoder is a variational auto-encoder, then for each latent variable Z, the encoder learns a statistical or probabilistic distribution, and the value input to the decoder is a random sample from the distribution.” ).
Regarding claim 7, ZHANG further teaches:
The latent space probability distribution is further based on a mean that is bound within a finite space. (ZHANG P[0170] “This means that for each individual element of latent space, the encoder learns one or more parameters of the respective distribution, e.g. a measure of centre point and spread of the distribution. For instance each latent variable Z, (a single dimension) may be modelled in the encoder by a respective mean value and standard deviation or variance.” P[0113] “k-Nearest Neighbour Head Parameters: Generate the new head parameters en as the mean of the head parameters of the k- nearest neighbour features in terms of Euclidean distance [finite space], where colunm-wise mean imputing is used to fill in unobserved values.”).
Regarding claim 13, ZHANG further teaches:
The variance parameter is configured to comprise a global value that is independent of the input data. (ZHANG P[0010] “In this case, for each element of Z the encoder learns one or more parameters of the distribution, e.g. a measure of centre point and spread of the distribution. For instance the centre point could be the mean and the spread [global value] could be the variance or standard deviation”. P[0170] “In some embodiments each of the latent representations Z, is one- dimensional, i.e. consists of only a single latent variable (element). Note however this does not imply the latent variable Z, is a modelled only as simple, fixed scalar value. Rather, as the auto-encoder is a variational auto-encoder, then for each latent variable Z, the encoder learns a statistical or probabilistic distribution [independent of the input data], and the value input to the decoder is a random sample from the distribution. This means that for each individual element of latent space, the encoder learns one or more parameters of the respective distribution, e.g. a measure of centre point and spread of the distribution. For instance each latent variable Z, (a single dimension) may be modelled in the encoder by a respective mean value and standard deviation or variance.” ).
Regarding claims 8, 10, and 14, they comprise of limitations similar to those of claims 1, 3, and 7 and are therefore rejected for similar rationale. Regarding claims 15 and 17, they comprise of limitations similar to those of claims 1 and 3 and are therefore rejected for similar rationale. Regarding claim 20, it comprises of limitations similar to those of claim 13 and is therefore rejected for similar rationale.
Claims 2, 9 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over LAI et al (U.S. Pub. No. U.S. 20230059367 A1), Zhang (U.S. Pub No. US 20220147818 A1), WONG (J.P. Pub No. JP 6737487 B1) in further view of CAO (U.S. Pub. No. US 11847538 B2).
Regarding claim 2, while LAI, as modified by ZHANG, does teach claim 1, which claim 2 is dependent upon, it does not explicitly teach:
Obtaining the input data from a client machine; Computing the output data by feeding the obtained input data into the trained neural network; Transmitting the computed output data to a server machine via a network.
However, in analogous art that similarly maintains privacy of users, CAO teaches:
Obtaining the input data from a client machine; (Column 4 Lines 42-45, “In various embodiments, the set of data curators 110A, 110B, and 110C include distinct entities that collect and/or maintain data including private data that is maintained in the private datasets 102A, 102B, or 102C.”)
Computing the output data by feeding the obtained input data into the trained neural network; (Column 5 Lines 35-39, “However, in other embodiments, the set of differentially private generative models 104A, 104B, and 104C is maintained by the corresponding data curators 110A, 110B, and 110C and the training data is provided to the data consumer 116.” discloses feeding input data into the generative models)
Transmitting the computed output data to a server machine via a network. (Column 5 Lines 55-59, “The set of differentially private generative models 104A, 104B, and 104C trained with privacy constraints on private data (e.g., the private datasets 102A, 102B, or 102C), in various embodiments, provide the data consumer 116 indirect access to the private data.” discloses providing data to a consumer)
It would have been obvious to a person skilled in the art before the effective filing date of the
invention to have combined with CAO‘s data computation and transmission methods and, with LAI‘s, as modified by ZHANG, method of data privacy, with a reasonable expectation of success, a method for handling data and computing data, as in CAO, for the sake of preserving user privacy, as found in LAI, as modified by ZHANG. A person of ordinary skill would have been motivated to increase efficiency. (CAO Column 3, lines 34-48).
Regarding claims 9 and 16, they comprise of limitations similar to those of claim 2 and are therefore rejected for similar rationale
Claims 4, 11 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over LAI et al (U.S. Pub. No. US 20230059367 A1), Zhang (U.S. Pub No. US 20220147818 A1), WONG (J.P. Pub No. JP 6737487 B1) in further view of Amroabadi et al (U.S. Pub No. US 20210133590 A1).
Regarding claim 4, while LAI, as modified by ZHANG, does teach claim 1, which claim 4 is dependent upon, it does not explicitly teach:
The input data comprises sequential data.
However, in analogous art that similarly handles private data, AMROABADI teaches:
The input data comprises sequential data. (AMROABADI P[0197] “Input data 110 can be time- series signals such as those illustrated in FIG. 4, may be sequential data such as credit card transactions, debit card transactions, or other financial transactions.”).
It would have been obvious to a person skilled in the art before the effective filing date of the
invention to have combined with AMROABADI‘s sequential input data and, with LAI‘s, as modified by ZHANG, data privatization, with a reasonable expectation of success, sequential input data, as in AMROABADI, which is then encoded, as found in LAI, as modified by ZHANG. A person of ordinary skill would have been motivated to improve privacy. (AMROABADI: P[0046]).
Regarding claims 11 and 18, they comprise of limitations similar to those of claim 4 and are therefore rejected for similar rationale.
Claims 5-6 are rejected under 35 U.S.C. 103 as being unpatentable over LAI et al (U.S. Pub. No. U.S. 20230059367 A1), Zhang (U.S. Pub No. US 20220147818 A1), WONG (J.P. Pub No. JP 6737487 B1)in further view of WANG (U.S. Pub No. US 20190318268 A1)
While LAI, as modified by ZHANG, does teach claim 1, which claim 5 is dependent upon, it does not explicitly teach:
The variance parameter comprises a global value that is independent of the input data.
However, in analogous art that similarly teaches training models, WANG teaches:
The variance parameter comprises a global value that is independent of the input data. (WANG [0068] “Thus, edge nodes 204A, 204B, . . . , 204N receive synchronized values of the global model parameters 208 but provide the synchronization node 202 with locally generated values of the new model parameters 210A, 210B, . . . , 210N, which can vary between edge nodes 210A-201N. The global model parameters 208 can include a global value of a model parameter (w(t))”).
It would have been obvious to a person skilled in the art before the effective filing date of the
invention to have combined with WANG‘s global value and, with LAI‘s, as modified by ZHANG, variance parameter, with a reasonable expectation of success, a global value, as in WANG, that is in a variance parameter, as found in LAI, as modified by ZHANG. A person of ordinary skill would have been motivated to improve user privacy. (WANG: P[0053]).
Regarding claim 6, it comprises of limitations similar to claim 5 and is therefore rejected for similar rationale.
Response to Arguments
Applicant’s arguments filed 08-SEPTEMBER-2026 have been fully considered, but they are found to be non-persuasive.
With regards to applicant’s remarks regarding the 103 rejection, the applicant argues that the prior art does not teach the newly amended claim 1. The examiner acknowledges this argument and has replaced part of the mapping of the LAI reference of claim 1 with new prior art WONG.
With regards to the applicant’s remarks regarding motivation for combination, the applicant argues that the final rejection does not provide sufficient basis for the motivation of combining ZHANG, LAI, and MOHAMMADY. Specifically, the applicant argues:
These references are directed to disparate problems in disparate architectures. Lai concerns simultaneously protecting users and sensitive entities in natural-language datasets. Mohammady concerns outsourcing anomaly detection to a third-party managed security service provider while enforcing a sensitivity value. Zhang concerns a partial variational autoencoder for predicting values of newly added features. The Final Office Action does not explain what would have led a skilled artisan to select and assemble these particular teachings, other than the roadmap provided by Applicants' own claims. Such reliance on the claims as a template is impermissible hindsight.
With regards to this argument, it should be noted that the architectures and problems are not exactly the same, they are not entirely disparate. LAI specifically uses a differential privacy methodology to protect users, similarly ZHANG utilizes privacy protecting methodologies such as encoding to achieve its benefit. Further, both LAI and ZHANG share a similar utilization of variance on datasets, which also increases the privacy and security of a user, which is the primary utilization of a differential privacy system. Just because ZHANG does not specifically cite its security or privacy benefits does not remove the benefits of its methodology nor should it be ignored that the methodology of ZHANG would provide further improvement to other areas of the present invention, which is a factor one of ordinary skill in the art would acknowledge, find useful, and utilize. Finally, regarding MOHAMMADY, MOHAMMADY has been removed from the present rejection and, as such, does not require further elaboration.
With regards to applicant’s arguments regarding the 103 rejection of claims 5-6, the applicant argues that WANG does not teach the type of global value claimed. Specifically, applicant argues:
Applicants respectfully submit that the Office has equated two unrelated uses of the word "global." Wang is directed to distributed machine learning at edge nodes of a distributed computer system. In Wang, "global model parameters" are the shared model weights that are synchronized across a plurality of distributed edge nodes by a synchronization node. Wang's use of "global" means shared across distributed nodes, that is, common to the edge nodes rather than locally generated at an individual node.
That is not the meaning of "global" in claims 5 and 6. In the claims, the variance parameter is "a global value that is independent of the input data," that is, the variance parameter does not vary as a function of the particular input data point being processed. This is confirmed by the specification, which explains that instead of computing the standard deviation as a function of the data point, the system may use a global standard deviation that is independent from the input. Wang's synchronized, node-shared model weight vector has nothing to do with a variance parameter that is independent of the input data point in the differential privacy sense recited in the claims.
With regards to this argument, while a ‘global standard deviation’ would be different in interpretation to the ‘global value’ that is present in the claim, the claim itself only cites a ‘global value’. There is no backing or paragraph in the specification that is provided that specifically equates the term ‘global value’ to mean ‘global standard deviation’, and as such the rejections to claims 5-6 will be maintained.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SKIELER A KOWALIK whose telephone number is (571)272-1850. The examiner can normally be reached 8-5.
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, Mariela D Reyes can be reached at (571)270-1006. 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.
/SKIELER ALEXANDER KOWALIK/ Examiner, Art Unit 2142
/Mariela Reyes/ Supervisory Patent Examiner, Art Unit 2142
