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
In the previous Office Action issued September 17, 2025 (hereinafter “the previous Office Action”), claims 1-20 were pending.
This action is in response to the amendment and remarks filed December 3, 2025. In the amendment, claims 1, 8, 11, 17, and 20 were amended, no claims were canceled, and no claims were added. Thus, claims 1-20 are pending.
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 1-20 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claims contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Regarding Claims 1, 11, and 20:
(1) Claims 1, 11, and 20 each recite “based on original data”.
(2) Furthermore, claims 1, 11, and 20 each recite: “to generate, without using a representation of the original data, a token set…”
Regarding both (1) and (2), Applicant’s Remarks, filed December 3, 2025, do not indicate support for the above limitation in the specification filed October 29, 2021. Specifically regarding (1), Examiner was unable to find any instances in the specification that explicitly disclose “original data”. On the other hand, regarding (2), Examiner found various paras. of the specification that describe generating a token set using a syntax generator. In particular, at least paras. [0035], [0050], [0060], [0065], and [0081] were found to disclose generating a token set using a syntax generator. Although the disclosure describes generating a token set, none of the found paras. disclose generating a token set without using a representation of the original data.
Therefore, claims 1, 11, and 20 are rejected for new matter because the specification does not provide adequate support for the claimed limitations involving original data and generating a token set without using representation of the original data.
Regarding Claims 2-10 and 12-19:
Claims 2-10 are rejected for inheriting the deficiencies of claim 1.
Claims 12-19 are rejected for inheriting the deficiencies of claim 11.
Claim Rejections - 35 USC § 103
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
Claims 1, 2, 4, 6, 11-13, and 17-19 are rejected under 35 U.S.C. 103 as being unpatentable over Goodsitt et al. (US 20200012892), hereinafter Goodsitt, in view of Philipsz (US 9369464).
Regarding Claim 1:
Goodsitt discloses:
A system for modular generation of synthetic data, comprising: at least one memory storing instructions; and at least one processor configured to execute the instructions to perform operations comprising:
Goodsitt, [0006], “The system may include one or more memory units for storing instructions; and one or more processors configured to execute the instructions to perform operations.”
receiving, from a device, a first syntax generator machine learning model of a plurality of syntax generator machine learning models modularly interchangeable with a plurality of content generator machine learning models
Goodsitt, [0056], “In step 401, dataset generator 103 can be configured to retrieve actual data (e.g. query input and/or returned data). As a non-limiting example, the actual data may have been gathered from user input executing a search query during the course of ordinary business operations, marketing operations, research operations, or the like.”
In para. 56, Goodsitt discloses receiving from a device, the query.
[0057], “Consistent with disclosed embodiments, dataset generator 103 can be configured with a classifier for distinguishing different classes of sensitive information. In some embodiments, dataset generator 103 can be configured with a recurrent neural network for distinguishing different classes of sensitive information”
In para. 57, Goodsitt then goes to Step 403, which discloses using a trained recurrent neural network with the dataset generator, which is understood to be the syntax generator that is selected based on the classes of the actual data. The syntax generator is understood to be also stored in a storage device, as Goodsitt teaches para. 6 as given above, which stores the entirety of the system. Furthermore, as the model is used, it is considered as being retrieved by the system under BRI. It is also considered a machine learning model as it uses a classifier. Back to para. 57, Goodsitt also teaches another syntax generator. As taught by Goodsitt, neural network can be used instead of the classifier and work with the plurality of content generators. Goodsitt teaches the syntax generator(s) being modularly interchangeable with a plurality of content generator machine learning models.
[0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
In para. 58, its understood that the first class-specific model is used to generate synthetic data. The models are class-specific therefore they generate different classes (types) of content and as taught by Goodsitt, there exists multiple classes, therefore a plurality of these models which work interchangeably with the one classifier of Goodsitt.
the first syntax generator machine learning model trained based on original data Goodsitt, [0054], “In step 309, computing resources 101 can generate a trained data model using the data model received from model optimizer 107 and the synthetic dataset received from dataset generator 103. For example, computing resources 101 can be configured to train the data model received from model optimizer 107 until some training criterion is satisfied.”
In para. 54, Goodsitt discloses training a machine learning model based on the query (includes sensitive data) that is trained on a synthetic dataset that replicates the syntax of the sensitive data.
executing the first syntax generator machine learning model to generate, without using a representation of the original data, syntax that includes a token set, corresponding to a synthetic data format, comprising a first token and a second token that are variables that represent types of information
Goodsitt, [0057], “Dataset generator 103 can be configured to apply the classifier to the actual data to determine that a sensitive portion of the training dataset belongs to the data class. For example, when the data stream includes the text string ‘Lorem ipsum 012-34-5678 dolor sit amet’ the classifier may be configured to indicate that positions 13-23 of the text string are the same format as a social security number, and thus may potentially constitute a social security number…dataset generator 103 can be configured with a recurrent neural network for distinguishing different classes of sensitive information”
[0041], “The synthetic data may be similar to the actual data in terms of query type, query input values, value distributions (e.g., univariate and multivariate statistics of the synthetic data may be similar to that of the actual data), returned data, structure and ordering, or the like. In this manner, the data model for the machine learning application can be generated without directly using the actual data.”
[0117], “In various embodiments, dataset generator 1207 can be configured to generate the synthetic data stream without reference to the reference data steam data.”
In para. 57, Goodsitt teaches in Step 403, using the dataset and a classifier to determine sensitive parts of the data: This is understood to be the token set, as there is multiple of these sensitive portions of the data and they are content (variables) that represent a type of data. Goodsitt also teaches there are multiple, there is at least a first and second element that describes the token set.
Paras. 41 and 117 discloses that the synthetic data generated by the data model can be generated without using the actual data [without using a representation of the original data].
retrieving, based on a first token type corresponding to the first token and a second token type corresponding to the second token, a first content generator machine learning model associated with the first token type and a second content generator machine learning model associated with the second token type from a storage device storing the plurality of content generator machine learning models
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Goodsitt teaches in step 405, therefore as the models are “class-specific”, they correspond to the types of the tokens. The content generator is represented as the dataset generator which uses the class-specific model as the model is used to generate data or “content”. The models are also part of the system as taught by Goodsitt para. 6 which teaches them existing in memory units.
wherein the first content generator machine learning model and the second content generator machine learning model are usable with any of the plurality of syntax generator machine learning models, wherein the first content generator machine learning model and the second content generator machine learning model are configured to generate different types of content
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
It is understood that the first class-specific model are used to generate synthetic data. The models are class-specific therefore they generate different classes (types) of content and as taught by Goodsitt, there exists multiple classes, therefore a plurality of these models exist and they interchangeably directly with the syntax generator(s) of the earlier step (step 403).
wherein the first content generator machine learning model or the second content generator machine learning model is configured to create synthetic data relating to the original data
As taught by Goodsitt in Step 405, the content generator machine model(s) are configured to generate synthetic data that based on the class of the sensitive data (therefore relating to the original data.), see para. 58 above.
executing the first content generator machine learning model, using the first content generator machine learning model modularly with the first syntax generator machine learning model, to generate first content data
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Therefore its understood that the first class-specific model are used to generate synthetic data, which represents the first content data.
executing the second content generator machine learning model, using the second content generator machine learning model modularly with the first syntax generator machine learning model, to generate second content data
Similarly, Goodsitt teaches this limitation in para. 58. As taught by Goodsitt in the earlier limitation, there are multiple syntax strings for the class-specific models in which the syntax strings are different from each other.
generating a first synthetic data set, that is a synthetic version of the original data…the first token with the first content data and the second token with the second content data
Goodsitt, [0059], ““In step 407, dataset generator 103 can be configured to replace the sensitive portion of the actual data with the synthetic portion. To continue the previous example, dataset generator 103 could be configured to replace the characters at positions 13-23 of the text string with the values ‘013-74-3285,’ creating the synthetic text string ‘Lorem ipsum 013-74-3285 dolor sit amet’.”
Goodsitt discloses generating a synthetic data set by replacing the generated content with the position of the first token, in which the data that includes the sensitive data [original data] isn’t being accessed or used but rather a text string and position data are used instead.”
training a machine learning model based on the first syntenic data set
Goodsitt, [0059], “But this text string can still be used to train models that make valid inferences regarding the actual data, because synthetic social security numbers generated by the synthetic data model share the statistical characteristic of social security numbers in the actual data.”
Goodsitt discloses that the synthetic dataset is used for training models, as outlined in Process 300 Paragraphs 0051 – 0053, in which the dataset has been stored and provided from the database.
Goodsitt does not explicitly disclose:
…by replacing, in the token set…
However, in the same field, analogous art Philipsz teaches:
…by replacing, in the token set…
Philipsz, [41], “To perform the key update, the CAS 210 generates a new key Key 3′ and establishes a secure communication channel with the token 10 using Key 2 to replace the stored Key 3 with Key 3′”
Philipsz teaches replacing a generated content in place of an original token in a set.
Goodsitt, Philipsz, and the instant application are analogous art because they are all directed to data generation.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Philipsz system of updating the stored token values of Goodsitt with the updated tokens as taught by Philipsz for the improvement of allowing the system to remove private keys and information. As Philipsz teaches: “Removal of the need for a hierarchy of trust and the need for a revocation procedure for compromised private keys can therefore be very useful.” (Column 1, Lines 29 – 32) in which the revocation procedure is the procedure of replacing keys.
Regarding Claim 2:
As discussed above Goodsitt in view of Philipsz teach [the] system of claim 1, and Goodsitt further discloses:
wherein retrieving the first content generator machine learning model is based on a criterion
Goodsitt, [0058]], “To continue the previous example, dataset generator 103 can generate a synthetic social security number using a synthetic data model trained to generate social security numbers”
Goodsitt teaches the requirement of the generator being targeted towards social security number is understood to be a criterion.
Regarding Claim 4:
As discussed above Goodsitt in view of Philipsz teach [the] system of claim 1, and Goodsitt further discloses:
wherein retrieving the first content generator machine learning model is based on a language criterion
Goodsitt, [0058], “To continue the previous example, dataset generator 103 can generate a synthetic social security number using a synthetic data model trained to generate social security numbers”
The requirement of the generator being targeted towards social security number is understanded to be a criterion. Furthermore, social security numbers are an element of language (as they are represented as a collection of numbers), therefore it follows a language criterion.
Regarding Claim 6:
As discussed above Goodsitt in view of Philipsz teach [the] system of claim 1, and Goodsitt further discloses:
wherein the synthetic data format is a text-based format
Goodsitt, [0058], “To continue the previous example, the social security-specific synthetic data model could generate the synthetic portion ‘03-74-3285’…”
Goodsitt teaches that the synthetic data format is that of a text format (as the social security number as expressed above and outputted by the generator is a text formatted string).
Regarding Claim 11:
Goodsitt discloses:
A method for modular generation of synthetic data, comprising:
Goodsitt, [0002], “The disclosed embodiments concern a platform for management of artificial intelligence systems. In particular, the disclosed embodiments concern using a disclosed platform to create models of data. These data models can be used to generate synthetic data for testing or training artificial intelligence systems.”
receiving, from a device, a first syntax generator model of a plurality of syntax generator models modularly interchangeable with a plurality of content generator models
Goodsitt, [0056], “In step 401, dataset generator 103 can be configured to retrieve actual data (e.g. query input and/or returned data). As a non-limiting example, the actual data may have been gathered from user input executing a search query during the course of ordinary business operations, marketing operations, research operations, or the like.”
In para. 56, Goodsitt discloses receiving from a device, the query.
[0057], “Consistent with disclosed embodiments, dataset generator 103 can be configured with a classifier for distinguishing different classes of sensitive information. In some embodiments, dataset generator 103 can be configured with a recurrent neural network for distinguishing different classes of sensitive information”
In para. 57, Goodsitt then goes to Step 403, which discloses using a trained recurrent neural network with the dataset generator, which is understood to be the syntax generator that is selected based on the classes of the actual data. The syntax generator is understood to be also stored in a storage device, as Goodsitt teaches para. 6 as given above, which stores the entirety of the system. Furthermore, as the model is used, it is considered as being retrieved by the system under BRI. It is also considered a machine learning model as it uses a classifier. Back to para. 57, Goodsitt also teaches another syntax generator. As taught by Goodsitt, neural network can be used instead of the classifier and work with the plurality of content generators. Goodsitt teaches the syntax generator(s) being modularly interchangeable with a plurality of content generator machine learning models.
[0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
In para. 58, its understood that the first class-specific model is used to generate synthetic data. The models are class-specific therefore they generate different classes (types) of content and as taught by Goodsitt, there exists multiple classes, therefore a plurality of these models which work interchangeably with the one classifier of Goodsitt.
the first syntax generator model trained based on original data
Goodsitt, [0054], “In step 309, computing resources 101 can generate a trained data model using the data model received from model optimizer 107 and the synthetic dataset received from dataset generator 103. For example, computing resources 101 can be configured to train the data model received from model optimizer 107 until some training criterion is satisfied.”
In para. 54, Goodsitt discloses training a machine learning model based on the query (includes sensitive data) that is trained on a synthetic dataset that replicates the syntax of the sensitive data.
executing the first syntax generator model to generate, without using representation of the original data, a token set, corresponding to a synthetic data format, comprising a first token and a second token that are variables that represent types of information
Goodsitt, [0057], “Dataset generator 103 can be configured to apply the classifier to the actual data to determine that a sensitive portion of the training dataset belongs to the data class. For example, when the data stream includes the text string ‘Lorem ipsum 012-34-5678 dolor sit amet’ the classifier may be configured to indicate that positions 13-23 of the text string are the same format as a social security number, and thus may potentially constitute a social security number…dataset generator 103 can be configured with a recurrent neural network for distinguishing different classes of sensitive information”
[0041], “The synthetic data may be similar to the actual data in terms of query type, query input values, value distributions (e.g., univariate and multivariate statistics of the synthetic data may be similar to that of the actual data), returned data, structure and ordering, or the like. In this manner, the data model for the machine learning application can be generated without directly using the actual data.”
[0117], “In various embodiments, dataset generator 1207 can be configured to generate the synthetic data stream without reference to the reference data steam data.”
In para. 57, Goodsitt teaches in Step 403, using the dataset and a classifier to determine sensitive parts of the data: This is understood to be the token set, as there is multiple of these sensitive portions of the data and they are content (variables) that represent a type of data. Goodsitt also teaches there are multiple, there is at least a first and second element that describes the token set.
Paras. 41 and 117 discloses that the synthetic data generated by the data model can be generated without using the actual data [without using a representation of the original data].
retrieving a first content generator model corresponding to a first token type corresponding to the first token and a second content generator model corresponding to second token type corresponding to the second token, from a storage device storing the plurality of content generator models
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Goodsitt teaches in step 405, therefore as the models are “class-specific”, they correspond to the types of the tokens. The content generator is represented as the dataset generator which uses the class-specific model as the model is used to generate data or “content”. The models are also part of the system as taught by Goodsitt para. 6 which teaches them existing in memory units.
executing the first content generator model, using the first content generator model modularly with the first syntax generator model, to generate first content data corresponding to the first token
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
It is understood that the first class-specific model are used to generate synthetic data, which represents the first content data.
executing the second content generator model, using the second content generator model modularly with the first syntax generator model, to generate second content data corresponding to the second token
Similarly, Goodsitt teaches this limitation in para. 58. As taught by Goodsitt in the earlier limitation, there are multiple syntax strings for the class-specific models in which the syntax strings are different from each other.
generating synthetic data set, that is a synthetic version of the original data…the first token with the first content data and the second token with the second content data
Goodsitt, [0059], ““In step 407, dataset generator 103 can be configured to replace the sensitive portion of the actual data with the synthetic portion. To continue the previous example, dataset generator 103 could be configured to replace the characters at positions 13-23 of the text string with the values ‘013-74-3285,’ creating the synthetic text string ‘Lorem ipsum 013-74-3285 dolor sit amet’.”
Goodsitt discloses generating a synthetic data set by replacing the generated content with the position of the first token, in which the data that includes the sensitive data [original data] isn’t being accessed or used but rather a text string and position data are used instead.”
training a machine learning model based on the first syntenic data set
Goodsitt, [0059], “But this text string can still be used to train models that make valid inferences regarding the actual data, because synthetic social security numbers generated by the synthetic data model share the statistical characteristic of social security numbers in the actual data.”
Goodsitt discloses that the synthetic dataset is used for training models, as outlined in Process 300 Paragraphs 0051 – 0053, in which the dataset has been stored and provided from the database.
Goodsitt does not explicitly disclose:
…by replacing, in the token set…
However, in the same field, analogous art Philipsz teaches:
…by replacing, in the token set…
Philipsz, [41], “To perform the key update, the CAS 210 generates a new key Key 3′ and establishes a secure communication channel with the token 10 using Key 2 to replace the stored Key 3 with Key 3′”
Philipsz teaches replacing a generated content in place of an original token in a set.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Philipsz system of updating the stored token values of Goodsitt with the updated tokens as taught by Philipsz for the improvement of allowing the system to remove private keys and information. As Philipsz teaches: “Removal of the need for a hierarchy of trust and the need for a revocation procedure for compromised private keys can therefore be very useful.” (Column 1, Lines 29 – 32) in which the revocation procedure is the procedure of replacing keys.
Regarding Claim 12:
As discussed above Goodsitt in view of Philipsz teach [the] method of claim 11, and Goodsitt further discloses:
wherein the first token type is different from the second token type
Goodsitt, [0057], “In some embodiments, dataset generator 103 can be configured with a recurrent neural network for distinguishing different classes of sensitive information.”
As mentioned above, the token types have been taken to represent the classes of the tokens from Goodsitt. The classifier specifically distinguishes multiple classes of the tokens, therefore the token types can be different.
Regarding Claim 13:
As discussed above Goodsitt in view of Philipsz teach [the] method of claim 11, and Goodsitt further discloses:
wherein the synthetic data format is a document type
Goodsitt, [0057], “As a non-limiting example, the actual data can include unstructured data (e.g., character strings, tokens, and the like) and structured data (e.g., key-value pairs, relational database files, spreadsheets, and the like).”
Goodsitt teaches the types of data that is being used for the tokens, in which it mentions “relational database files” and “spreadsheets” which constitute a document type.
Regarding Claim 17:
As discussed above Goodsitt in view of Philipsz teach [the] method of claim 11, and Goodsitt further discloses:
wherein the synthetic data format is a first synthetic data format and the token set is a first token set, and the method further comprises: receiving a second syntax generator model, the second syntax generator model being different from the first syntax generator
Goodsitt, [0058], “In some embodiments, dataset generator 103 can be configured with a recurrent neural network for distinguishing different classes of sensitive information”
[0057], “As an additional non-limiting example, when the actual data is personnel records, classes could include employee identification numbers, employee names, employee addresses, contact information, marital or beneficiary information, title and salary information, and employment actions.”
In para. 58, Goodsitt teaches a second syntax generator model that is different than the classifier model of Claim 11. This one is the combined neural network and dataset generator which generates Syntax as Goodsitt teaches the classes of data that can be observed in para. 57.
generating, using the second syntax generator model, a second token set different from the first token set and comprising the first token and the second token that are the same token types as the first token set
Goodsitt, [0057], “Dataset generator 103 can be configured to apply the classifier to the actual data to determine that a sensitive portion of the training dataset belongs to the data class.”
Goodsitt teaches using the dataset and the neural network to determine sensitive parts of the data. Goodsitt teaches directly that the neural network can be used instead of the classifier, as described above. This is understood to be the token set, as there is multiple of these sensitive portions of the data. As there are multiple, there is at least a new first and second element that describes the token set and as these are created by a new syntax generator, the token sets differ from the first token set but can have the same token types as the first token set generated.
executing the first content generator model, using the first content generator model modularly with the second syntax generator model, to generate first additional content data
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Therefore its understood that the class-specific models are used to generate synthetic data, which represents the first additional content data.
executing the second content generator model, using the second content generator model modularly with the second syntax generator model, to generate second additional content data
Similarly, Goodsitt teaches this limitation in para. 58. As taught by Goodsitt in the earlier limitation, there are multiple syntax strings for the class-specific models in which the syntax strings are different from each other.
generating second synthetic data by replacing, in the second token set, the first token with the first additional content data and the second token with the second additional content data
Goodsitt, [0059], “In step 407, dataset generator 103 can be configured to replace the sensitive portion of the actual data with the synthetic portion.” As taught in Claim 11, Philipsz teaches replacing tokens in the token set: “To perform the key update, the CAS 210 generates a new key Key 3′ and establishes a secure communication channel with the token 10 using Key 2 to replace the stored Key 3 with Key 3′” (Column 10, line 32) which teaches replacing a generated content in place of an original token in a set.
wherein training the machine learning model is based on the synthetic data and the second synthetic data
Goodsitt, [0059], “But this text string can still be used to train models that make valid inferences regarding the actual data, because synthetic social security numbers generated by the synthetic data model share the statistical characteristic of social security numbers in the actual data.”
Goodsitt discloses that the synthetic dataset is used for training models, as outlined in Process 300 Paragraphs 0051 – 0053, in which the dataset has been stored and provided from the database.
Regarding Claim 18:
As discussed above Goodsitt in view of Philipsz teach [the] method of claim 17, and Philipsz further discloses:
wherein the second token set further comprises a third token having a third token type
Philipsz, col. 15, “the token stores ID 1, ID 2, Key 1 a, Key 2, and Key 3…”
Philipsz teaches a token with multiple sets of data within the token. The token in this example is the token set, while the multiples of data stored within it are the tokens.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine Philipsz system of storing multiple tokens with the syntax token system of Goodsitt for the known result of allowing the system to keep track of more tokens.
Regarding Claim 19:
As discussed above Goodsitt in view of Philipsz teach [the] method of claim 18, and Goodsitt further discloses:
selecting, from the plurality of content generator models, a third content generator associated with the third token type
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Therefore as the models are “class-specific”, they are associated with the types of the tokens as described before. Furthermore, they are used as models to generate synthetic portions of data, therefore they fulfill the requirement of being a content generator. Furthermore, para. 60 teaches process 410, which directly describes “generating synthetic data using class and subclass-specific models”, therefore its understood that there is a plurality of these content generators.
wherein generating the second synthetic data further comprises replacing the third token with the third content data
Goodsitt, [0059], “In step 407, dataset generator 103 can be configured to replace the sensitive portion of the actual data with the synthetic portion.”
As taught in Claim 11, Philipsz teaches replacing tokens in the token set: “To perform the key update, the CAS 210 generates a new key Key 3′ and establishes a secure communication channel with the token 10 using Key 2 to replace the stored Key 3 with Key 3′” (Column 10, line 32) which teaches replacing a generated content in place of an original token in a set.
The motivation for this combination can be found in Claim 11.
Claims 3 and 5 are rejected under 35 U.S.C. 103 as being unpatentable over Goodsitt in view of Philipsz as applied to claim 1 above, and further in view of Cheng (“A Regular Expression-based DGL for Meaningful Synthetic Data Generation”).
Regarding Claim 3:
As discussed above Goodsitt in view of Philipsz teach [the] system of claim 1, and Goodsitt further discloses:
wherein retrieving the first content generator machine learning model…
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Goodsitt teaches selecting a first content generator machine model: in which the class-specific model being used with the dataset generator is a machine learning model used to generate content.
Goodsitt in view of Philipsz do not explicitly disclose:
…based on the synthetic data format
However, in the same field, analogous art Cheng teaches:
…based on the synthetic data format
Cheng teaches figure 1, which describes a type/format inference of the random sample generator being an aspect for the generator.
Goodsitt, Philipsz, Cheng, and the instant application are analogous art because they are all directed to data generation.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the selection step of Goodsitt to also include the requirement of selecting based on a desired synthetic data format as taught by Chen for the improvement of allowing the system to select carefully formatted types which are supported by the underlying system. Cheng teaches: “data should be carefully formatted to data types supported by the underlying database system.” (Section 2.A), which helps maintain that the system only deals with data that it is meant to handle.
Regarding Claim 5:
As discussed above Goodsitt in view of Philipsz teach [the] system of claim 1, but do not explicitly disclose:
wherein receiving the first syntax generator machine learning model is further based on a language criterion
However, in the same field, analogous art Cheng teaches:
wherein receiving the first syntax generator machine learning model is further based on a language criterion
Cheng teaches figure 1, which describes a RegExp parser of the random sample generator being an aspect for the generator. As mentioned in Section IV, “RegExp parser takes a regular expression as input, analyzes its syntax structure and builds a sequence of elements in the form <character set, repetitions>.”, which describes that the generator is one that is selected based on its ability to handle a language criterion (the specific sequence of elements).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the step of using a syntax generator machine learning model of Goodsitt with the addition limitation of selecting the generator based on the ability to handle a language criterion as taught by Cheng as it leads to the improvement of limiting the use of the generator to only parts that are applicable to the system. This is taught by Cheng in Section 1: “This allows to apply a quantifier to the entire group or to restrict alternation to part of the whole regular expression.”
Claims 7 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over Goodsitt in view of Philipsz as applied to claims 1 and 14 above, respectively, and further in view of Kim et al. (“Long-Term Video Generation with Evolving Residual Video Frames”), hereinafter Kim.
Regarding Claim 7:
As discussed above Goodsitt in view of Philipsz teach [the] system of claim 1, but do not explicitly disclose:
wherein the synthetic data format is an image format
However, in the same field, analogous art Kim teaches:
wherein the synthetic data format is an image format
Kim, p. 1, Introduction section: “The adversary loss allows the learning to generate realistic and synthetic images.”
Kim teaches a generation of synthetic images for a deep learning architecture.
Goodsitt, Philipsz, Kim, and the instant application are analogous art because they are all directed to data generation.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the possible synthetic data format to be that of an image format for the improvement of allowing the system to create data for many different purposes. Kim teaches: “The technique is widely used for predicting human behaviors, tracking objects, and classifying activities” (Introduction).
Regarding Claim 14:
As discussed above Goodsitt in view of Philipsz teach [the] method of claim 11, but do not explicitly disclose:
wherein the synthetic data format is a video format
However, in the same field, analogous art Kim teaches:
wherein the synthetic data format is a video format
Kim, p. 4, Conclusion section: “We proposed a long-term video generation algorithm to learn internal video representations and reproduce realistic pixels in the future video frames”
Kim teaches generation of synthetic videos.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the synthetic data format to also include that of a video format for the improvement of allowing the system to create data for many different purposes. Kim teaches: “The technique is widely used for predicting human behaviors, tracking objects, and classifying activities” (Introduction).
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Goodsitt in view of Philipsz as applied to claim 1 above, and further in view of Kale et al. (US 20200334557), hereinafter Kale.
Regarding Claim 8:
As discussed above Goodsitt in view of Philipsz teach [the] system of claim 1, and Goodsitt further discloses:
calculating a similarity metric reflecting a measure of similarity between the first synthetic data set and reference data
Goodsitt, [0073], “configured to determine a similarity metric value between the synthetic dataset and the normalized reference dataset”
Goodsitt in view of Philipsz do not explicitly disclose:
retraining, based on the similarity metric and a threshold, at least one of the first content generator machine learning model or the second content generator machine learning model
However, in the same field, analogous art Kale teaches:
retraining, based on the similarity metric and a threshold, at least one of the first content generator machine learning model or the second content generator machine learning model
Kale, [0174], “configured to determine whether the quality metric is above a threshold and, if not, issue instructions to retrain the model(s).”
Kale teaches retraining models based on a threshold and specified value. The quality metric is defined in para. 29 of Kale and it serves as a similarity metric as the process to create it involves calculating the similarity between two datasets and then ranking them.
Goodsitt, Philipsz, Kale, and the instant application are analogous art because they are all directed to data generation.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to apply Kale’s process for retraining the data model of Goodsitt to allow the system to analyze the quality of the model and preform remedial action (Kate teaches in paragraph 0168 “the embodiments described herein allow users to easily determine the quality of the predictions so that remedial action may be taken if the quality is low (for instance, retraining the system).”
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Goodsitt in view of Philipsz, further in view of Kale as applied to claim 8 above, and further in view of Soni et al. (US 20200134446), hereinafter Soni.
Regarding Claim 9:
As discussed above Goodsitt in view of Philipsz, further in view of Kale teach [the] system of claim 8, but do not explicitly disclose:
wherein the operations further comprise generating, based on the retrained content generator machine learning model, a second set of synthetic data
However, in the same field, analogous art Soni teaches:
wherein the operations further comprise generating, based on the retrained content generator machine learning model, a second set of synthetic data
Soni, [0005], “based on a comparison of a first output of the first DNN model and a second output of the second DNN model, the synthetic data generator to generate third synthetic data”
Soni teaches using a synthetic data generator machine learning model used to create a new set of data after the first original synthetic data set.
Goodsitt, Philipsz, Kale, Soni, and the instant application are analogous art because they are all directed to data generation.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the retrained content generator of Goodsitt as modified by Philipsz and further modified by Kale to create a new synthetic dataset for the improvement of having a machine learning model that tested on the synthetic data set and confirmed to create content that align to desired results. This improvement is taught by Kale: “deploy an artificial intelligence model trained and tested using the synthetic data set.” [0004] as well “to generate third synthetic data from a first site when the comparison indicates that performance of the first DNN model and performance of the second DNN model are aligned” [0005] when regarding the third generated model.
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Goodsitt in view of Philipsz as applied to claim 1 above, and further in view of Rossenbach et al. (“Generating Synthetic Audio Data for Attention-Based Speech Recognition Systems”), hereinafter Rossenbach.
Regarding Claim 10:
As discussed above Goodsitt in view of Philipsz teach [the] system of claim 1, but do not explicitly disclose:
wherein the synthetic data format is an audio format
However, in the same field, analogous art Rossenbach teaches:
wherein the synthetic data format is an audio format
Rossenbach teaches this limitation in Section 4.1, “In a next step, we used the text of the LibriSpeech-360h corpus to generate synthetic audio with the GST-TTS model” which teaches using a model to generate audio (therefore the synthetic data is an audio format).
Goodsitt, Philipsz, Rossenbach, and the instant application are analogous art because they are all directed to data generation.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the synthetic data format to also include that of an audio format, as it allows for the known result of being able to produce synthetic audio content.
Claims 15-16 are rejected under 35 U.S.C. 103 as being unpatentable over Goodsitt in view of Philipsz as applied to claim 11 above, and further in view of Fogarty et al. (US 20190362268), hereinafter Fogarty.
Regarding Claim 15:
As discussed above Goodsitt in view of Philipsz teach [the] method of claim 11, and Goodsitt and Philipsz further disclose:
generating, using the first content generator model, first content data corresponding to the first token
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Therefore its understood that the first class-specific model are used to generate synthetic data, which represents the first content data.
generating, using the second content generator model, second content data corresponding to the second token
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Therefore its understood that the second class-specific model are used to generate synthetic data, which represents the second content data.
As taught in Claim 11, Philipsz teaches replacing tokens in the token set: “To perform the key update, the CAS 210 generates a new key Key 3′ and establishes a secure communication channel with the token 10 using Key 2 to replace the stored Key 3 with Key 3′” (Column 10, line 32) which teaches replacing a generated content in place of an original token in a set.
The motivation for this combination can be found in Claim 11.
Goodsitt in view of Philipsz do not explicitly disclose:
generating a plurality of synthetic data records by repeating the steps of:
However, in the same field, analogous art Fogarty teaches:
generating a plurality of synthetic data records by repeating the steps of:
Fogarty teaches this limitation in paragraph 0059: “In a repeated replacing process, the propensity scores for each target group are developed”
Goodsitt, Philipsz, Fogarty, and the instant application are analogous art because they are all directed to data generation.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to perform a repetition of the process of synthetic data generation that Goodsitt teaches which Fogarty teaches for the purposes of creating an optimal synthetic data set which involves a proper compilation of the target group information (Fogarty teaches this in Paragraph 0058: “The procedure to develop the optimal synthetic data set starts with the development of a generic response model whose dependent variable set comprises a compilation of the target group (determined using the nearest neighbor searching described earlier) along with all of the independent variables available to model.”)
Regarding Claim 16:
As discussed above Goodsitt in view of Philipsz, further in view of Fogarty teach [the] method of claim 15, and Goodsitt further discloses:
wherein training the machine learning model comprises: training the machine learning model by using the plurality of generated synthetic data records as training data
Goodsitt, [0053], “In step 309, computing resources 101 can generate a trained data model using the data model received from model optimizer 107 and the synthetic dataset received from dataset generator 103.”
Where the synthetic dataset is the training data and the process is regarding training a data model. As mentioned in para. 42 of Goodsitt, a data model types are a “neural network, recurrent neural network, …”, which are considered machine learning models.
Claim 20 is rejected under 35 U.S.C. 103 as being unpatentable over Goodsitt in view of Philipsz, and further in view of Kale.
Regarding Claim 20:
Goodsitt discloses:
A method for modular generation of synthetic data, comprising: training, using training data having a synthetic data format that replicates the syntax of original data, a syntax generator model to generate a plurality of tokens corresponding to the synthetic data format
Goodsitt, [0049], “Process 300 begins with step 301. In step 301, dataset generator 103 can retrieve a query input and/or output. In some aspects, the query input and/or output can include actual training data (e.g. query input and returned data).”
[0051], “In step 305, in some embodiments, dataset generator 103 can generate synthetic data. Dataset generator 103 can also be configured to use the data model retrieved from model storage 109 to generate a synthetic dataset by replacing the sensitive data items with synthetic data items.”
[0054], “In step 309, computing resources 101 can generate a trained data model using the data model received from model optimizer 107 and the synthetic dataset received from dataset generator 103. For example, computing resources 101 can be configured to train the data model received from model optimizer 107 until some training criterion is satisfied.”
In para. 49, Goodsitt teaches Process 300, which includes Step 305 from para. 51 an further includes Step 309 in para. 54, which teaches training a machine learning model based on the query (includes sensitive data) that is trained on a synthetic dataset that replicates the syntax of the sensitive data and as seen by Process 400, generates a plurality of tokens corresponding to a synthetic data format.
executing the syntax generator to generate, without using a representation of the original data, a token set comprising a first token and a second token that are variables that represent types of information
Goodsitt, [0057], “Dataset generator 103 can be configured to apply the classifier to the actual data to determine that a sensitive portion of the training dataset belongs to the data class. For example, when the data stream includes the text string ‘Lorem ipsum 012-34-5678 dolor sit amet’ the classifier may be configured to indicate that positions 13-23 of the text string are the same format as a social security number, and thus may potentially constitute a social security number…dataset generator 103 can be configured with a recurrent neural network for distinguishing different classes of sensitive information”
[0041], “The synthetic data may be similar to the actual data in terms of query type, query input values, value distributions (e.g., univariate and multivariate statistics of the synthetic data may be similar to that of the actual data), returned data, structure and ordering, or the like. In this manner, the data model for the machine learning application can be generated without directly using the actual data.”
[0117], “In various embodiments, dataset generator 1207 can be configured to generate the synthetic data stream without reference to the reference data steam data.”
In para. 57, Goodsitt teaches in Step 403, using the dataset and a classifier to determine sensitive parts of the data: This is understood to be the token set, as there is multiple of these sensitive portions of the data and they are content (variables) that represent a type of data. Goodsitt also teaches there are multiple, there is at least a first and second element that describes the token set.
Paras. 41 and 117 discloses that the synthetic data generated by the data model can be generated without using the actual data [without using a representation of the original data].
retrieving a first content generator model, of a plurality of content generator models corresponding to a first token type corresponding to the first token and a second content generator model corresponding to the second token type corresponding to the second token
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Therefore as the models are “class-specific”, they correspond to the types of the tokens. The content generator is represented as the dataset generator which uses the class-specific model as the model is used to generate data or “content”.
executing the first content generator model, using the first content generator model modularity with the syntax generator model, to generate first content data corresponding to the first token
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Therefore its understood that the first class-specific model are used to generate synthetic data, which represents the first content data.
executing the second content generator model, using the second content generator, second content data corresponding to the second token
Goodsitt, [0058], “In step 405, dataset generator 103 can be configured to generate a synthetic portion using a class-specific model”
Therefore its understood that the second class-specific model are used to generate synthetic data, which represents the second content data.
generating synthetic data…the first token with the first content data and the second token with the second content data
Similarly, Goodsitt teaches this limitation in para. 58. As taught by Goodsitt in the earlier limitation, there are multiple syntax strings for the class-specific models in which the syntax strings are different from each other.
calculating a similarity metric of the synthetic data by comparing the synthetic data to reference data
Goodsitt, [0073], “configured to determine a similarity metric value between the synthetic dataset and the normalized reference dataset”
Goodsitt does not explicitly disclose:
…by replacing, in the token set…
determining that the similarity metric does not exceed a similarity threshold; and
retraining, based on the determination, the syntax generator model
However, in the same field, analogous art Philipsz teaches:
…by replacing, in the token set…
Philipsz, [41], “To perform the key update, the CAS 210 generates a new key Key 3′ and establishes a secure communication channel with the token 10 using Key 2 to replace the stored Key 3 with Key 3′”
Philipsz teaches replacing a generated content in place of an original token in a set.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Philipsz system of updating the stored token values of Goodsitt with the updated tokens as taught by Philipsz for the improvement of allowing the system to remove private keys and information. As Philipsz teaches: “Removal of the need for a hierarchy of trust and the need for a revocation procedure for compromised private keys can therefore be very useful.” (Column 1, Lines 29-32) in which the revocation procedure is the procedure of replacing keys.
Goodsitt in view of Philipsz do not explicitly disclose:
determining that the similarity metric does not exceed a similarity threshold; and
retraining, based on the determination, the syntax generator model
However, in the same field, analogous art Kale teaches:
determining that the similarity metric does not exceed a similarity threshold; and
retraining, based on the determination, the syntax generator model
Kale, [0174], “configured to determine whether the quality metric is above a threshold and, if not, issue instructions to retrain the model(s).”
Kale teaches retraining models based on a threshold and specified value. The quality metric is defined in para. 29 of Kale and it serves as a similarity metric as the process to create it involves calculating the similarity between two datasets and then ranking them.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to apply Kale’s process for retraining the data model of Goodsitt to allow the system to analyze the quality of the model and preform remedial action (Kate teaches in para. 168 “the embodiments described herein allow users to easily determine the quality of the predictions so that remedial action may be taken if the quality is low (for instance, retraining the system).”
Response to Arguments
Applicant's arguments filed December 3, 2025 (“Remarks”) have been fully considered but they are not persuasive.
35 U.S.C. § 103:
Remarks, pp. 11-12. Applicant argues the applied references do not teach “executing the first syntax generator machine learning model to generate, without using a representation of the original data, syntax that includes a token set”. Examiner respectfully disagrees. As discussed in detail under the section 103 rejections above, Goodsitt teaches other embodiments of generating synthetic data without using actual data/reference data. Therefore, Goodsitt teaches the claimed limitation and remains rejected under 35 U.S.C. § 103.
Remarks, pp. 12-13. Applicant’s arguments in view of the Lesh reference used in the previous Office Action have been considered but are moot because the new ground of rejection does not rely on the Lesh reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
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.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to STEVEN PHUNG whose telephone number is (703) 756-1499. The examiner can normally be reached Monday-Thursday: 9:00AM-4:00PM ET.
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, KAMRAN AFSHAR can be reached at (571) 272-7796. 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.
/S.H.P./Examiner, Art Unit 2125
/KAMRAN AFSHAR/Supervisory Patent Examiner, Art Unit 2125
Prosecution Insights