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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 11/11/2025 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Status of Claims
Applicant’s submission filed 11/05/2025 has been entered. The status of claims is as follows:
Claims 1-20 remain pending in the application.
Claims 1, 9 and 17 are amended.
Response to Arguments
In reference to the Claim Objections:
Applicant asserts in Remarks pg. 2 that the typographical error in claims 4 and 19 “appears to have been addressed in a previous response”.
Examiner respectfully disagrees and notes that review of the record indicates that claims 4 and 19 were not amended in the previous response, and no correction to the identified typographical error is present. Accordingly, the objection to claims 4 and 19 is maintained.
In reference to the claim rejections under 35 U.S.C 101:
Regarding Step 2A, Prong 1:
Applicant asserts in Remarks pg. 3 that rather than the abstract ideas of "mathematical calculation", "mathematical concept," or "mental process," however, the claim 1 is directed to the technological improvement of "sending only the segment of the transformed data over a network to a user device" to reduce traffic over the network and as such, the claims are patent eligible. Therefore, Applicant contends that claim 1 (and similarly claims 9 and 17) is not directed to a judicial exception of an abstract idea and is therefore patent-eligible.
Examiner respectfully disagrees and notes that although claim 1 reciting “sending only the segment of the transformed data over a network to a user device” is not directed to an abstract idea or mental process under Step 2A, Prong One, this limitation is directed to an insignificant extra solution activities under Step 2A, Prong Two and is directed to selective data transmission under Step 2B. The claim does not recite any improvement to network technology itself, such as specific transmission protocol, network architecture, or mechanism that alters how data is transmitted. Rather, the limitation is a result-oriented statement of reducing transmitted data volume, which constitutes data selection and transmission, which is an abstract idea, implemented using generic networking components. Accordingly, claim 1 remains directed to an abstract idea and is not patent eligible.
Applicant’s arguments filed on 11/05/2025 have been fully considered but they are not persuasive.
Regarding Step 2A, Prong 2:
Applicant asserts in Remarks pg. 3-4 that the recited features are not limited to a generic computer. Rather, claim 1 recites that "a dataset of user personal data" has a "dimensionality reduction action" performed such that "only the segment of the transformed data" is sent "over a network." Here, the claims are limited to rules with specific characteristics. The claims themselves set out meaningful requirements for the "sending only the segment of the transformed data." Whether at step one or step two of the Alice test, in determining the patentability of a method, a court must look to the claims as an ordered combination, without ignoring the requirements of the individual steps. The specific, claimed features of these rules allow for the improvement realized by the invention. Therefore, the approach recites an improvement to the functioning of a technology beyond generally linking the use of the purported judicial exception to a particular technological environment; and the approach thus qualifies as patent eligible subject matter under Step 2A prong 2.
Examiner respectfully disagrees and notes that although claim 1 recites performing a dimensionality reduction action and sending only a segment of transformed data over a network, the claim does not recite any specific improvement to network technology or computer functionality. The claimed limitations merely describe selecting, transforming, and transmitting data based on information content, which constitutes abstract data processing and management. The claim lacks any technical details regarding how the dimensionality reduction or data transmission is performed in a manner that improves network operation itself. Accordingly, the recited judicial exception is not integrated into a practical application and claim 1 remains directed to an abstract idea.
Applicant’s arguments filed on 11/05/2025 have been fully considered but they are not persuasive.
Regarding Step 2B:
Applicant asserts in Remarks pg. 4-5 that, under Step 2B, claim 1 recites additional elements that amount to significantly more than the recited judicial exception. Applicant asserts that although dimensionality reduction is a mathematical operation, claim 1 includes additional limitations such as sending only a segment of transformed data over a network to a user device, which allegedly adds meaningful and unconventional steps. Applicant contends these features confine the claim to a particular useful application, improve technological functioning, and go beyond mere generic computer implementation. Applicant further relies on Cosmokey Solutions GMBH & Co. v. Duo Security LLC No. 20-2043 (Fed. Cir. 2021) to argue that the claimed combination yields advantages over prior art and therefore constitutes patent-eligible subject matter. Applicant requests withdrawal of the 101 rejection for claim 1 and its dependent claims.
Applicant’s argument under Step 2B is not persuasive. Although claim 1 recites additional steps beyond the dimensionality reduction operation, such as sending only a segment of transformed data over a network, these limitations merely reflect conventional data selection and transmission using generic computer and networking components. The claim does not recite any unconventional computer, network or data processing techniques, nor does it specify a particular manner in which the additional steps are performed that would amount to significantly more than the judicial exception. Accordingly, the claim as a whole, does not include an inventive concept and the rejection under 35 U.S.C 101 is maintained.
Applicant’s arguments filed on 11/05/2025 have been fully considered but they are not persuasive.
In reference to the claim rejections under 35 U.S.C 103:
Applicant’s arguments, see Remarks filed on 11/05/2025, with respect to the rejection(s) of claim(s) under 35 U.S.C 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Thomas et al. (US 11,182,502 B2)
Claim Objections
Claims 4, 19 are objected to because of the following informalities:
In Claim 4, Line 8: “transforming said at least one of said data segments” should
read “transforming at least one of said data segments”
In Claim 19, Line 8: “transforming said at least one of said data segments” should
read “transforming at least one of said data segments”
Appropriate correction is required.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claims contain an abstract idea without significantly more.
Regarding Claim 1:
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is a process.
Step 2A – Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes, the claim recites an abstract idea.
computing a benefit-to-resource score for a dataset of user personal data by combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
selecting an autoencoder architecture based on said benefit-to-resource score wherein said autoencoder architecture balances minimizing reconstruction loss and minimizing required storage space for a dimensionality reduction action based on said benefit-to-resource score;
- This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.). Additionally, the limitation “based on said autoencoder architecture” is interpreted as merely using the autoencoder to perform the transformation process, which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permission - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
Step 2A – Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, there are no additional elements that integrate the judicial exception into a practical application. The additional elements:
a memory; – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
a processor in communication with said memory, – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
said processor being configured to perform operations comprising: – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
storing said transformed data in a user space. – This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
sending only the segment of the transformed data over a network to a user device. – This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, there are no additional elements that amount to significantly more than the judicial exception. The additional elements are:
a memory; – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
a processor in communication with said memory, – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
said processor being configured to perform operations comprising: – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
storing said transformed data in a user space. – This limitation is directed to storing and retrieving information in memory, which the court have recognized as well-understood, routine, and conventional activity (see MPEP 2106.05 (d) II. iv)
sending only the segment of the transformed data over a network to a user device. This limitation is directed to receiving or transmitting data over a network. The courts have recognized receiving or transmitting data over a network as well understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity (see MPEP 2106.05(d) II.).
Regarding claim 2,
Claim 2 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 1 which includes an abstract idea (see rejection for claim 1). The additional limitations:
segmenting said dataset into data segments according to type of personal data; - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
computing a segment benefit-to-resource score for a first data segment; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
transforming said first data segment into a transformed data segment with a first transformer function based on a first autoencoder architecture, – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.).
said first autoencoder architecture selected based on said segment benefit-to-resource score; and - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
streaming one or more of said transformed data segments to a content personalizer. – This limitation amounts to insignificant extra-solution activity (see MPEP 2106.05(g)) under Step 2A Prong 2. Furthermore, under step 2B, it is also directed to receiving or transmitting data over a network. The courts have recognized receiving or transmitting data over a network as well understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity (see MPEP 2106.05(d) II.).
Regarding claim 3,
Claim 3 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 1 which includes an abstract idea (see rejection for claim 1). The additional limitations:
determining a weightage for each of said data segments wherein said weightage is based on a semantic purpose of an analytics service - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
enabling reduce-transformation intensities in said autoencoder architecture according to said weightage. – The examiner interprets limitation “reduce-transformation intensities” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.). Additionally, “enabling” here is interpreted as just doing the transformation process by using the autoencoder, which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
Regarding claim 4,
Claim 4 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 1 which includes an abstract idea (see rejection for claim 1). The additional limitations:
segmenting said dataset into data segments; - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
computing a segment benefit-to-resource score for each of said data segments; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
selecting a segment autoencoder architecture for each of said data segments based on said segment benefit-to-resource scores; - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
leveraging at least one of said segment autoencoder architectures to – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
at least one of said data segments into at least one transformed data segments; – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.).
transforming said at least one of said data segments. – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.).
Regarding claim 5,
Claim 5 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 4 which includes an abstract idea (see rejection for claim 4). The additional limitations:
at least one transformed data segment is streamed to a machine learning service. – This limitation amounts to insignificant extra-solution activity (see MPEP 2106.05(g)) under Step 2A Prong 2. Furthermore, under step 2B, it is also directed to receiving or transmitting data over a network. The courts have recognized receiving or transmitting data over a network as well understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity (see MPEP 2106.05(d) II.).
Regarding claim 6,
Claim 6 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 1 which includes an abstract idea (see rejection for claim 1). The additional limitations:
enabling dimensionality reduction based on explainable machine learning. – The examiner interprets limitation “reduce-transformation intensities” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.). Additionally, “enabling” here is interpreted as just doing the transformation process by using the explainable machine learning, which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
Regarding claim 7,
Claim 7 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 1 which includes an abstract idea (see rejection for claim 1). The additional limitations:
said autoencoder architecture is additionally based on dominant features extracted from Shapley additive explanations analysis. – The examiner interprets Shapley additive explanation analysis as a mathematical concept (see MPEP 2106.04(a)(2) l. C.) as it has the same concept with performing an algorithm to extract the dominant features. Additionally, “said autoencoder architecture” is interpreted as merely using the autoencoder to perform the mathematical concept which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
Regarding claim 8,
Claim 8 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 1 which includes an abstract idea (see rejection for claim 1). The additional limitations:
said benefit-to-resource score exceeds a threshold and This claim merely recites a further limitation on the computing a benefit-to-resource score for a dataset from Claim 1 which was directed to mathematical calculation (see MPEP 2106.04(a)(2) l. C.)
said autoencoder architecture is selected to minimize reconstruction loss. - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
Regarding Claim 9,
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is a process.
Step 2A – Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes, the claim recites an abstract idea.
computing, [by a processor], a benefit-to-resource score for a dataset of user personal data by combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
selecting an autoencoder architecture based on said benefit-to-resource score wherein said autoencoder architecture balances minimizing reconstruction loss and minimizing required storage space for a dimensionality reduction action based on said benefit-to-resource score; - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.). Additionally, the limitation “based on said autoencoder architecture” is interpreted as merely using the autoencoder to perform the transformation process, which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permission - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
Step 2A – Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, there are no additional elements that integrate the judicial exception into a practical application. The additional elements:
by a processor – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
storing said transformed data in a user space – This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
sending only the segment of the transformed data over a network to a user device. – This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, there are no additional elements that amount to significantly more than the judicial exception. The additional elements are:
by a processor – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
storing said transformed data in a user space. – This limitation is directed to storing and retrieving information in memory, which the court have recognized as well-understood, routine, and conventional activity (see MPEP 2106.05 (d) II. iv)
sending only the segment of the transformed data over a network to a user device. This limitation is directed to receiving or transmitting data over a network. The courts have recognized receiving or transmitting data over a network as well understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity (see MPEP 2106.05(d) II.).
Regarding claim 10,
Claim 10 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 9 which includes an abstract idea (see rejection for claim 9). The additional limitations:
segmenting said dataset into data segments according to type of personal data; - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
computing a segment benefit-to-resource score for a first data segment; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
transforming said first data segment into a transformed data segment with a first transformer function based on a first autoencoder architecture, – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.).
said first autoencoder architecture selected based on said segment benefit-to-resource score; and - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
streaming one or more of said transformed data segments to a content personalizer. – This limitation amounts to insignificant extra-solution activity (see MPEP 2106.05(g)) under Step 2A Prong 2. Furthermore, under step 2B, it is also directed to receiving or transmitting data over a network. The courts have recognized receiving or transmitting data over a network as well understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity (see MPEP 2106.05(d) II.).
Regarding claim 11,
Claim 11 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 9 which includes an abstract idea (see rejection for claim 9). The additional limitations:
determining a weightage for each of said data segments wherein said weightage is based on a semantic purpose of an analytics service - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
enabling reduce-transformation intensities in said autoencoder architecture according to said weightage. – The examiner interprets limitation “reduce-transformation intensities” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.). Additionally, “enabling” here is interpreted as just doing the transformation process by using the autoencoder, which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
Regarding claim 12,
Claim 12 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 9 which includes an abstract idea (see rejection for claim 9). The additional limitations:
segmenting said dataset into data segments; - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
computing a segment benefit-to-resource score for each of said data segments; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
selecting a segment autoencoder architecture for each of said data segments based on said segment benefit-to-resource scores; and - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
transforming at least one of said data segments into at least one transformed data segments by leveraging at least one of said segment autoencoder architectures; – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.).
Regarding claim 13,
Claim 13 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 12 which includes an abstract idea (see rejection for claim 12). The additional limitations:
at least one transformed data segment is streamed to a machine learning service. – This limitation amounts to insignificant extra-solution activity (see MPEP 2106.05(g)) under Step 2A Prong 2. Furthermore, under step 2B, it is also directed to receiving or transmitting data over a network. The courts have recognized receiving or transmitting data over a network as well understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity (see MPEP 2106.05(d) II.).
Regarding claim 14,
Claim 14 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 9 which includes an abstract idea (see rejection for claim 9). The additional limitations:
enabling dimensionality reduction based on explainable machine learning. – The examiner interprets limitation “reduce-transformation intensities” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.). Additionally, “enabling” here is interpreted as just doing the transformation process by using the explainable machine learning, which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
Regarding claim 15,
Claim 15 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 9 which includes an abstract idea (see rejection for claim 9). The additional limitations:
said autoencoder architecture is additionally based on dominant features extracted from Shapley additive explanations analysis. – The examiner interprets Shapley additive explanation analysis as a mathematical concept (see MPEP 2106.04(a)(2) l. C.) as it has the same concept with performing an algorithm to extract the dominant features. Additionally, “said autoencoder architecture” is interpreted as merely using the autoencoder to perform the mathematical concept which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
Regarding claim 16,
Claim 16 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 9 which includes an abstract idea (see rejection for claim 9). The additional limitations:
said benefit-to-resource score exceeds a threshold and This claim merely recites a further limitation on the computing a benefit-to-resource score for a dataset from Claim 9 which was directed to mathematical calculation (see MPEP 2106.04(a)(2) l. C.)
said autoencoder architecture is selected to minimize reconstruction loss. - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
Regarding Claim 17,
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
Yes, the claim is a process.
Step 2A – Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
Yes, the claim recites an abstract idea.
computing, [by said processor], a benefit-to-resource score for a dataset of user personal data by combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
selecting an autoencoder architecture based on said benefit-to-resource score wherein said autoencoder architecture balances minimizing reconstruction loss and minimizing required storage space for a dimensionality reduction action based on said benefit-to-resource score; - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture; – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.). Additionally, the limitation “based on said autoencoder architecture” is interpreted as merely using the autoencoder to perform the transformation process, which amounts to adding the words "apply it" (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea [see MPEP 2106.05(f)] and therefore fails to integrate the exception into a practical application.
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permission - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
Step 2A – Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, there are no additional elements that integrate the judicial exception into a practical application. The additional elements:
said computer program product comprising a computer readable storage medium having program instructions embodied therewith, – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
said program instructions executable by a processor to cause said processor to perform a function, – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
said function comprising: – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
by said processor – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
storing said transformed data in a user space – This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
sending only the segment of the transformed data over a network to a user device. – This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, there are no additional elements that amount to significantly more than the judicial exception. The additional elements are:
said computer program product comprising a computer readable storage medium having program instructions embodied therewith, – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
said program instructions executable by a processor to cause said processor to perform a function, – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
said function comprising: – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
by said processor – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
storing said transformed data in a user space. – This limitation is directed to storing and retrieving information in memory, which the court have recognized as well-understood, routine, and conventional activity (see MPEP 2106.05 (d) II. iv)
sending only the segment of the transformed data over a network to a user device. This limitation is directed to receiving or transmitting data over a network. The courts have recognized receiving or transmitting data over a network as well understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity (see MPEP 2106.05(d) II.).
Regarding claim 18,
Claim 18 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 17 which includes an abstract idea (see rejection for claim 17). The additional limitations:
said benefit-to-resource score exceeds a threshold and This claim merely recites a further limitation on the computing a benefit-to-resource score for a dataset from Claim 17 which was directed to mathematical calculation (see MPEP 2106.04(a)(2) l. C.)
said autoencoder architecture is selected to minimize reconstruction loss. - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
Regarding claim 19,
Claim 19 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 17 which includes an abstract idea (see rejection for claim 17). The additional limitations:
computing a segment benefit-to-resource score for each of said data segments; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
selecting a segment autoencoder architecture for each of said data segments based on said segment benefit-to-resource scores; and - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
transforming at least one of said data segments into at least one transformed data segments by leveraging at least one of said segment autoencoder architectures; – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.).
Regarding claim 20,
Claim 20 is rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more. The claim is dependent on claim 17 which includes an abstract idea (see rejection for claim 17). The additional limitations:
segmenting said dataset into data segments according to type of personal data; - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
computing a segment benefit-to-resource score for a first data segment; This limitation is directed to mathematical calculation as it is computing a benefit-to-resource score by calculating a user benefit and a resource cost, then dividing the user benefit by the resource cost (see MPEP 2106.04(a)(2) l. C.)
transforming said first data segment into a transformed data segment with a first transformer function based on a first autoencoder architecture, – The examiner interprets limitation “transforming said dataset into transformed data” here as the dimensionality reduction using a transformation function which amounts mathematical concept (see MPEP 2106.04(a)(2) l. C.).
said first autoencoder architecture selected based on said segment benefit-to-resource score; and - This limitation is directed to the abstract idea of a mental process (including an observation, evaluation, judgement, opinion) which can be performed in the human mind, or by a human using pen and paper (see MPEP 2106.04(a)(2) Ill. C.)
streaming one or more of said transformed data segments to a content personalizer. – This limitation amounts to insignificant extra-solution activity (see MPEP 2106.05(g)) under Step 2A Prong 2. Furthermore, under step 2B, it is also directed to receiving or transmitting data over a network. The courts have recognized receiving or transmitting data over a network as well understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity (see MPEP 2106.05(d) II.).
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1-2, 4-6, 8-10, 12-14, and 16-20 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang et al. (US 12,008,462 B2) (hereafter referred to as “Zhang”), in view of Nguyen et al. (“AutoGAN-based dimension reduction for privacy preservation”) (hereafter referred to as “Nguyen”), Burke & Kayem (“K-Anonymity for Privacy Preserving Crime Data Publishing in Resource Constrained Environments”) (hereafter referred to as “Burke”) and further in view of Thomas et al. (US 11,182,502 B2)
Regarding Claim 1, Zhang explicitly discloses:
A system for privacy-driven data sharing, said system comprising: a memory; and (Zhang, Col. 3, Lines 40-41: “In another aspect, the present disclosure provides a persistent memory storage”)
a processor in communication with said memory, said processor being configured to perform operations comprising: (Zhang, Col. 3, Lines 40-46: “In another aspect, the present disclosure provides a persistent memory storage, having stored thereon a set of instructions that, when executed by a processor, cause the processor to initiate a resource-aware runtime scheduler that encodes an inference accuracy and processing latency of each of a number of descendent models of a number of concurrently running applications into a cost function”)
selecting an autoencoder architecture based on said benefit-to-resource score (Zhang, Col. 6, Lines 20-27: “Given the cost functions of all the concurrently running vision applications, a device can employ a runtime scheduler that selects the most suitable sub-model of a multi-capacity model for each application and determines the optimal amount of runtime resources to allocate to each selected sub-model to jointly maximize the accuracy and minimize the inference latency of concurrent vision applications.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the same framework can be applied to other types of neural network including an autoencoder. So, “selecting an autoencoder architecture” is being interpreted as “selects the most suitable sub-model”]
wherein said autoencoder architecture balances minimizing reconstruction loss and minimizing required storage space for a dimensionality reduction action based on said benefit-to- resource score; (Zhang, Abstract: “The multi-capacity framework enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the framework of balancing between performance and resources consumed can also be used for other type of neural networks such as autoencoder]
Zhang fails to disclose:
computing a benefit-to-resource score for a dataset of user personal data by combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data;
a dimensionality reduction action
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture; and
storing said transformed data in a user space,
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permissions.
sending only the segment of the transformed data over a network to a user
device.
However, Burke explicitly discloses:
computing a benefit-to-resource score for a dataset of user personal data; (Burke, Pg. 833, Col. 2, Section A.: “Providing the service provider with a useful but privacy protected dataset is vital to ensuring the validity of the results returned and the anonymity of the reporters (users).”, Pg. 835, Figure 1:
PNG
media_image1.png
448
571
media_image1.png
Greyscale
,Pg. 837, Col. 1, ¶[2]: “The utility of the anonymized dataset indicates how useful it is, in terms of capacity for returning useful information in response to queries, in comparison to the original (non-anonymized) data set. Data utility is computed as follows:
PNG
media_image2.png
98
490
media_image2.png
Greyscale
”) [Examiner’s note: the user personal data is being interpreted as the crime report data]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Burke. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Burke teaches an anonymization model used for privacy preserving crime data publishing in resource constrained environments. One of ordinary skill would have motivation to combine Zhang and Burke to improve performance, and minimize information loss while maximizing classification accuracy.
However, Thomas explicitly discloses:
combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data (Thomas, Col. 4, Lines 5-9: “data privacy-utility tradeoff calculator configured to: compute a utility index based on midpoint of the balanced buckets and the privacy data; and compute attribute variations based on the number of variations between the buckets and the balanced buckets”, Col. 7, Lines 7-10: “FIG. 3A and FIG. 3B represent an exemplary flow diagram illustrating a computer implemented method 200 for computing the data privacy-utility tradeoff, in accordance with an embodiment of the present disclosure”, Col. 7, Lines 13-37: “In an embodiment, at step 202, the system 100 is configured to receive via a data connector 108A, data generated or captured from one or more data sources (Data Source 1, Data Source 2, ... Data Source n) to make it consumable by one or more data buyers. In an embodiment, the data connector 108A can also be configured to receive a query and other parameters along with the data. For instance, the query may be in the form of a request for data pertaining to people in the age group 40-50 years and suffering from diabetes or people in a particular region having heart problems, and the like, wherein parameters of the query may include age, gender, place of residence, and the like. In accordance with the present disclosure, the received data is analyzed or processed to make it suitable for dissemination to one or more data buyers without losing its utility in the process of retaining privacy associated with the data. In an embodiment, the data connector 108A connects to the one or more data sources to receive generated or captured data in bulk format or streaming content format. Data in bulk format may be uploaded at pre-determined intervals or randomly by data sellers. Data in streaming content format may be data provided in real time by connecting with the one or more data sources such as Fitbit™ devices, accelerometer devices, temperature devices and electrical consumption devices.”, Col. 7, Lines 63-67: “generate an adversary model by partitioning the search space into sets of buckets; each set corresponding to a sensitive attribute having a privacy data associated thereof. For instance a bucket set may pertain to age and include buckets [0-9], [10-19], and the like. Another bucket set may pertain to diseases and include [diabetes], [heart disease], [lung disease], and the like.”) [Examiner’s note: Under BRI, the “benefit value” is being interpreted as privacy value of user data and this value is received by a manual query from buyers or automatically streamed from more data sources such as Fitbit devices, “cost value” is being interpreted as the “attribute variations” based on the information variation between buckets wherein the bucket contains set of sensitive attributes (i.e., function of the bucket if to collect, maintain and transmit data)]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Thomas. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Thomas teaches systems and methods for computing data privacy-utility tradeoff. One of ordinary skill would have motivation to combine Zhang and Thomas because MPEP 2143 sets forth the Supreme Court rationales for obviousness including: (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results; (E): “Obvious to try” choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success; (F) Known work in one field of endeavor may prompt variations of it for use in either the same field or a different one based on design incentives or other market forces if the variations are predictable to one of the ordinary skill in the art.
However, Nguyen explicitly discloses:
a dimensionality reduction action (Nguyen, Abstract: “We first introduce theoretical approach to evaluate dimension reduction-based privacy preserving mechanisms, then propose a non-linear dimension reduction framework motivated by state-of-the-art neural network structures for privacy preservation”)
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
storing said transformed data in a user space, (Nguyen, Page 95, Col. 1, ¶[2]: “Our dimension reduction (DR) framework can be applied to different types of data and used in several practical applications without heavy computation of encryption and impact of query number. The proposed framework can be applied directly to the access control system mentioned above. More elaboratively, face images are locally collected, nonlinearly compressed to achieve DR, and sent to the authentication center.”) [Examiner’s note: a user space i.e., the authentication center]
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permissions. (Nguyen, Page 96, Col. 2, Section 3.1: “For example, if member n requests to access web server 2, the local device first takes a facial photo of the member by an attached camera, locally transforms it into lower dimension data, and sends to an authentication center. The authentication server then obtains the low dimensional data and determines member access eligibility by using a classifier without clear face images of the requesting member.”, Page 99, Col. 1, ¶[1]: “We randomly divide the dataset into two groups of subjects and labels their images to (1) or (0) depending on their access permission.”) [Examiner’s note: user space i.e., the authentication server, a segment of transformed data i.e., a facial photo of the member]
sending only the segment of the transformed data over a network to a user
device. (Nguyen, Pg. 96, Col. 2, Section 3.1: “the local device first takes a facial photo of the member by an attached camera, locally transforms it into lower dimension data, and sends to an authentication center. The authentication server then obtains the low dimensional
data and determines member access eligibility by using a classifier without clear face images of the requesting member”) [Examiner’s note: Under BRI, this corresponds to transforming original user data on the user device, resulting in transformed data that is distinct from the original raw data (i.e., the full face image)]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Nguyen. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Nguyen teaches a new dimension reduction-based method for privacy preservation that generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. One of ordinary skill would have motivation to combine Zhang and Nguyen to achieve privacy-preserving via dimension reduction by transforming the data through an encoder. This allows sensitive or identifiable information to be obscured and the transformed data represents an encoded form of the original data, which might not be directly interpretable, thus enhancing privacy (Nguyen, Page 97, Col. 1, Section 3.3, ¶1])
Regarding Claim 2, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 1 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
segmenting said dataset into data segments according to type of personal data; (Burke, Pg. 838, Col. 2, Section A: “Two relevant non sensitive attributes were specified (age and gender) as well as one sensitive attribute. In one experiment, the “Reports” dataset was divided into subsets of different sizes, starting at 10 tuples and stopping at 10000 tuples. The intuition behind this decision being that a crime reporting service will start out with a small number of reports, but this can and likely will grow exponentially as the system is implemented and used.”, Pg. 838, Col. 2, Section B: “For consistency with Datafly, Incognito, and Mondrian, in measuring processing time, the attribute “age” was selected as quasi-identifier and the “crime” category as the sensitive value.”)
computing a segment benefit-to-resource score for a first data segment; (Burke, Pg. 833, Col. 2, Section A.: “Providing the service provider with a useful but privacy protected dataset is vital to ensuring the validity of the results returned and the anonymity of the reporters (users).”, Pg. 835, Figure 1:
PNG
media_image1.png
448
571
media_image1.png
Greyscale
,Pg. 837, Col. 1, ¶[2]: “The utility of the anonymized dataset indicates how useful it is, in terms of capacity for returning useful information in response to queries, in comparison to the original (non-anonymized) data set. Data utility is computed as follows:
PNG
media_image2.png
98
490
media_image2.png
Greyscale
”) [Examiner’s note: the user personal data is being interpreted as the crime report data]
transforming said data segments into transformed data segments; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
streaming one or more of said transformed data segments to a content personalizer. (Nguyen, Page 95, Col. 1, ¶[2]: “Our dimension reduction (DR) framework can be applied to different types of data and used in several practical applications without heavy computation of encryption and impact of query number. The proposed framework can be applied directly to the access control system mentioned above. More elaboratively, face images are locally collected, nonlinearly compressed to achieve DR, and sent to the authentication center. The server then performs classification tasks on the dimension-reduced data.”, Page 95, Fig. 1:
PNG
media_image3.png
509
655
media_image3.png
Greyscale
) [Examiner’s note: a content personalizer i.e., the web servers]
Regarding Claim 4, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 1 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
wherein said operations further comprise: segmenting said dataset into data segments; (Burke, Pg. 838, Col. 2, Section A: “Two relevant non sensitive attributes were specified (age and gender) as well as one sensitive attribute. In one experiment, the “Reports” dataset was divided into subsets of different sizes, starting at 10 tuples and stopping at 10000 tuples. The intuition behind this decision being that a crime reporting service will start out with a small number of reports, but this can and likely will grow exponentially as the system is implemented and used.”, Pg. 838, Col. 2, Section B: “For consistency with Datafly, Incognito, and Mondrian, in measuring processing time, the attribute “age” was selected as quasi-identifier and the “crime” category as the sensitive value.”)
computing a segment benefit-to-resource score for each of said data segments; (Burke, Pg. 833, Col. 2, Section A.: “Providing the service provider with a useful but privacy protected dataset is vital to ensuring the validity of the results returned and the anonymity of the reporters (users).”, Pg. 835, Figure 1:
PNG
media_image1.png
448
571
media_image1.png
Greyscale
,Pg. 837, Col. 1, ¶[2]: “The utility of the anonymized dataset indicates how useful it is, in terms of capacity for returning useful information in response to queries, in comparison to the original (non-anonymized) data set. Data utility is computed as follows:
PNG
media_image2.png
98
490
media_image2.png
Greyscale
”) [Examiner’s note: the user personal data is being interpreted as the crime report data]
selecting a segment autoencoder architecture for each of said data segments based on said segment benefit-to-resource scores; (Zhang, Col. 6, Lines 20-27: “Given the cost functions of all the concurrently running vision applications, a device can employ a runtime scheduler that selects the most suitable sub-model of a multi-capacity model for each application and determines the optimal amount of runtime resources to allocate to each selected sub-model to jointly maximize the accuracy and minimize the inference latency of concurrent vision applications.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the same framework can be applied to other types of neural network including an autoencoder. So, “selecting an autoencoder architecture” is being interpreted as “selects the most suitable sub-model”]
leveraging at least one of said segment autoencoder architectures to transform at least one of said data segments into at least one transformed data segments; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
transforming said at least one of said data segments. (Nguyen, Page
97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
Regarding Claim 5, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 4 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
at least one transformed data segment is streamed to a machine learning service. (Nguyen, Page 95, Col. 1, ¶[2]: “Our dimension reduction (DR) framework can be applied to different types of data and used in several practical applications without heavy computation of encryption and impact of query number. The proposed framework can be applied directly to the access control system mentioned above. More elaboratively, face images are locally collected, nonlinearly compressed to achieve DR, and sent to the authentication center. The server then performs classification tasks on the dimension-reduced data.”, Page 95, Fig. 1:
PNG
media_image3.png
509
655
media_image3.png
Greyscale
) [Examiner’s note: a machine learning service i.e., the web servers]
Regarding Claim 6, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 1 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
enabling dimensionality reduction based on explainable machine learning. (Nguyen, Abstract: “To address these problems, we propose a new dimension reduction-based method for privacy preservation. Our method generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. We first introduce a theoretical approach to evaluate dimension reduction-based privacy preserving mechanisms, then propose a non-linear dimension reduction framework motivated by state-of-the-art neural network structures for privacy preservation.”)
Regarding Claim 8, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 1 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
said benefit-to-resource score exceeds a threshold and said autoencoder architecture is selected to minimize reconstruction loss. (Nguyen, Page 98, Col. 1, Section 3.5: “In order to meet a certain level of reconstruction distance, we consider the constrained problem:
PNG
media_image4.png
68
585
media_image4.png
Greyscale
. The optimization problem above can be approximated as an unconstrained problem:
PNG
media_image5.png
49
575
media_image5.png
Greyscale
, where γ is a penalty parameter and C is a penalty function
PNG
media_image6.png
39
578
media_image6.png
Greyscale
Note that C is nonnegative, and C(θ ) = 0 iff the constraint in (8) is satisfied.”) [Examiner’s note: Nguyen discloses architectures (or models) are optimized to meet a certain reconstruction accuracy (within the threshold ϵ) by minimizing the reconstruction loss with this constraint]
Regarding Claim 9, Zhang explicitly discloses:
selecting an autoencoder architecture based on said benefit-to-resource score (Zhang, Col. 6, Lines 20-27: “Given the cost functions of all the concurrently running vision applications, a device can employ a runtime scheduler that selects the most suitable sub-model of a multi-capacity model for each application and determines the optimal amount of runtime resources to allocate to each selected sub-model to jointly maximize the accuracy and minimize the inference latency of concurrent vision applications.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the same framework can be applied to other types of neural network including an autoencoder. So, “selecting an autoencoder architecture” is being interpreted as “selects the most suitable sub-model”]
wherein said autoencoder architecture balances minimizing reconstruction loss and minimizing required storage space for a dimensionality reduction action based on said benefit-to- resource score; (Zhang, Abstract: “The multi-capacity framework enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the framework of balancing between performance and resources consumed can also be used for other type of neural networks such as autoencoder]
Zhang fails to disclose:
computing a benefit-to-resource score for a dataset of user personal data by combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data;
a dimensionality reduction action
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture; and
storing said transformed data in a user space,
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permissions.
sending only the segment of the transformed data over a network to a user
device.
However, Burke explicitly discloses:
computing a benefit-to-resource score for a dataset of user personal data; (Burke, Pg. 833, Col. 2, Section A.: “Providing the service provider with a useful but privacy protected dataset is vital to ensuring the validity of the results returned and the anonymity of the reporters (users).”, Pg. 835, Figure 1:
PNG
media_image1.png
448
571
media_image1.png
Greyscale
,Pg. 837, Col. 1, ¶[2]: “The utility of the anonymized dataset indicates how useful it is, in terms of capacity for returning useful information in response to queries, in comparison to the original (non-anonymized) data set. Data utility is computed as follows:
PNG
media_image2.png
98
490
media_image2.png
Greyscale
”) [Examiner’s note: the user personal data is being interpreted as the crime report data]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Burke. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Burke teaches an anonymization model used for privacy preserving crime data publishing in resource constrained environments. One of ordinary skill would have motivation to combine Zhang and Burke to improve performance, and minimize information loss while maximizing classification accuracy.
However, Thomas explicitly discloses:
combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data (Thomas, Col. 4, Lines 5-9: “data privacy-utility tradeoff calculator configured to: compute a utility index based on midpoint of the balanced buckets and the privacy data; and compute attribute variations based on the number of variations between the buckets and the balanced buckets”, Col. 7, Lines 7-10: “FIG. 3A and FIG. 3B represent an exemplary flow diagram illustrating a computer implemented method 200 for computing the data privacy-utility tradeoff, in accordance with an embodiment of the present disclosure”, Col. 7, Lines 13-37: “In an embodiment, at step 202, the system 100 is configured to receive via a data connector 108A, data generated or captured from one or more data sources (Data Source 1, Data Source 2, ... Data Source n) to make it consumable by one or more data buyers. In an embodiment, the data connector 108A can also be configured to receive a query and other parameters along with the data. For instance, the query may be in the form of a request for data pertaining to people in the age group 40-50 years and suffering from diabetes or people in a particular region having heart problems, and the like, wherein parameters of the query may include age, gender, place of residence, and the like. In accordance with the present disclosure, the received data is analyzed or processed to make it suitable for dissemination to one or more data buyers without losing its utility in the process of retaining privacy associated with the data. In an embodiment, the data connector 108A connects to the one or more data sources to receive generated or captured data in bulk format or streaming content format. Data in bulk format may be uploaded at pre-determined intervals or randomly by data sellers. Data in streaming content format may be data provided in real time by connecting with the one or more data sources such as Fitbit™ devices, accelerometer devices, temperature devices and electrical consumption devices.”, Col. 7, Lines 63-67: “generate an adversary model by partitioning the search space into sets of buckets; each set corresponding to a sensitive attribute having a privacy data associated thereof. For instance a bucket set may pertain to age and include buckets [0-9], [10-19], and the like. Another bucket set may pertain to diseases and include [diabetes], [heart disease], [lung disease], and the like.”) [Examiner’s note: Under BRI, the “benefit value” is being interpreted as privacy value of user data and this value is received by a manual query from buyers or automatically streamed from more data sources such as Fitbit devices, “cost value” is being interpreted as the “attribute variations” based on the information variation between buckets wherein the bucket contains set of sensitive attributes (i.e., function of the bucket if to collect, maintain and transmit data)]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Thomas. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Thomas teaches systems and methods for computing data privacy-utility tradeoff. One of ordinary skill would have motivation to combine Zhang and Thomas because MPEP 2143 sets forth the Supreme Court rationales for obviousness including: (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results; (E): “Obvious to try” choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success; (F) Known work in one field of endeavor may prompt variations of it for use in either the same field or a different one based on design incentives or other market forces if the variations are predictable to one of the ordinary skill in the art.
However, Nguyen explicitly discloses:
a dimensionality reduction action (Nguyen, Abstract: “We first introduce theoretical approach to evaluate dimension reduction-based privacy preserving mechanisms, then propose a non-linear dimension reduction framework motivated by state-of-the-art neural network structures for privacy preservation”)
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
storing said transformed data in a user space, (Nguyen, Page 95, Col. 1, ¶[2]: “Our dimension reduction (DR) framework can be applied to different types of data and used in several practical applications without heavy computation of encryption and impact of query number. The proposed framework can be applied directly to the access control system mentioned above. More elaboratively, face images are locally collected, nonlinearly compressed to achieve DR, and sent to the authentication center.”) [Examiner’s note: a user space i.e., the authentication center]
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permissions. (Nguyen, Page 96, Col. 2, Section 3.1: “For example, if member n requests to access web server 2, the local device first takes a facial photo of the member by an attached camera, locally transforms it into lower dimension data, and sends to an authentication center. The authentication server then obtains the low dimensional data and determines member access eligibility by using a classifier without clear face images of the requesting member.”, Page 99, Col. 1, ¶[1]: “We randomly divide the dataset into two groups of subjects and labels their images to (1) or (0) depending on their access permission.”) [Examiner’s note: user space i.e., the authentication server, a segment of transformed data i.e., a facial photo of the member]
sending only the segment of the transformed data over a network to a user
device. (Nguyen, Pg. 96, Col. 2, Section 3.1: “the local device first takes a facial photo of the member by an attached camera, locally transforms it into lower dimension data, and sends to an authentication center. The authentication server then obtains the low dimensional
data and determines member access eligibility by using a classifier without clear face images of the requesting member”) [Examiner’s note: Under BRI, this corresponds to transforming original user data on the user device, resulting in transformed data that is distinct from the original raw data (i.e., the full face image)]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Nguyen. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Nguyen teaches a new dimension reduction-based method for privacy preservation that generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. One of ordinary skill would have motivation to combine Zhang and Nguyen to achieve privacy-preserving via dimension reduction by transforming the data through an encoder. This allows sensitive or identifiable information to be obscured and the transformed data represents an encoded form of the original data, which might not be directly interpretable, thus enhancing privacy (Nguyen, Page 97, Col. 1, Section 3.3, ¶1])
Regarding Claim 10, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 9 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
segmenting said dataset into data segments according to type of personal data; (Burke, Pg. 838, Col. 2, Section A: “Two relevant non sensitive attributes were specified (age and gender) as well as one sensitive attribute. In one experiment, the “Reports” dataset was divided into subsets of different sizes, starting at 10 tuples and stopping at 10000 tuples. The intuition behind this decision being that a crime reporting service will start out with a small number of reports, but this can and likely will grow exponentially as the system is implemented and used.”, Pg. 838, Col. 2, Section B: “For consistency with Datafly, Incognito, and Mondrian, in measuring processing time, the attribute “age” was selected as quasi-identifier and the “crime” category as the sensitive value.”)
computing a segment benefit-to-resource score for a first data segment; (Burke, Pg. 833, Col. 2, Section A.: “Providing the service provider with a useful but privacy protected dataset is vital to ensuring the validity of the results returned and the anonymity of the reporters (users).”, Pg. 835, Figure 1:
PNG
media_image1.png
448
571
media_image1.png
Greyscale
,Pg. 837, Col. 1, ¶[2]: “The utility of the anonymized dataset indicates how useful it is, in terms of capacity for returning useful information in response to queries, in comparison to the original (non-anonymized) data set. Data utility is computed as follows:
PNG
media_image2.png
98
490
media_image2.png
Greyscale
”) [Examiner’s note: the user personal data is being interpreted as the crime report data]
transforming said data segments into transformed data segments; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
streaming one or more of said transformed data segments to a content personalizer. (Nguyen, Page 95, Col. 1, ¶[2]: “Our dimension reduction (DR) framework can be applied to different types of data and used in several practical applications without heavy computation of encryption and impact of query number. The proposed framework can be applied directly to the access control system mentioned above. More elaboratively, face images are locally collected, nonlinearly compressed to achieve DR, and sent to the authentication center. The server then performs classification tasks on the dimension-reduced data.”, Page 95, Fig. 1:
PNG
media_image3.png
509
655
media_image3.png
Greyscale
) [Examiner’s note: a content personalizer i.e., the web servers]
Regarding Claim 12, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 9 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
segmenting said dataset into data segments; (Burke, Pg. 838, Col. 2, Section A: “Two relevant non sensitive attributes were specified (age and gender) as well as one sensitive attribute. In one experiment, the “Reports” dataset was divided into subsets of different sizes, starting at 10 tuples and stopping at 10000 tuples. The intuition behind this decision being that a crime reporting service will start out with a small number of reports, but this can and likely will grow exponentially as the system is implemented and used.”, Pg. 838, Col. 2, Section B: “For consistency with Datafly, Incognito, and Mondrian, in measuring processing time, the attribute “age” was selected as quasi-identifier and the “crime” category as the sensitive value.”)
computing a segment benefit-to-resource score for each of said data segments; (Burke, Pg. 833, Col. 2, Section A.: “Providing the service provider with a useful but privacy protected dataset is vital to ensuring the validity of the results returned and the anonymity of the reporters (users).”, Pg. 835, Figure 1:
PNG
media_image1.png
448
571
media_image1.png
Greyscale
,Pg. 837, Col. 1, ¶[2]: “The utility of the anonymized dataset indicates how useful it is, in terms of capacity for returning useful information in response to queries, in comparison to the original (non-anonymized) data set. Data utility is computed as follows:
PNG
media_image2.png
98
490
media_image2.png
Greyscale
”) [Examiner’s note: the user personal data is being interpreted as the crime report data]
selecting a segment autoencoder architecture for each of said data segments based on said segment benefit-to-resource scores; and (Zhang, Col. 6, Lines 20-27: “Given the cost functions of all the concurrently running vision applications, a device can employ a runtime scheduler that selects the most suitable sub-model of a multi-capacity model for each application and determines the optimal amount of runtime resources to allocate to each selected sub-model to jointly maximize the accuracy and minimize the inference latency of concurrent vision applications.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the same framework can be applied to other types of neural network including an autoencoder. So, “selecting an autoencoder architecture” is being interpreted as “selects the most suitable sub-model”]
leveraging at least one of said segment autoencoder architectures to transform at least one of said data segments into at least one transformed data segments; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
transforming at least one of said data segments. (Nguyen, Page
97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
Regarding Claim 13, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 12 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
at least one transformed data segment is streamed to a machine learning service. (Nguyen, Page 95, Col. 1, ¶[2]: “Our dimension reduction (DR) framework can be applied to different types of data and used in several practical applications without heavy computation of encryption and impact of query number. The proposed framework can be applied directly to the access control system mentioned above. More elaboratively, face images are locally collected, nonlinearly compressed to achieve DR, and sent to the authentication center. The server then performs classification tasks on the dimension-reduced data.”, Page 95, Fig. 1:
PNG
media_image3.png
509
655
media_image3.png
Greyscale
) [Examiner’s note: a machine learning service i.e., the web servers]
Regarding Claim 14, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 9 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
enabling dimensionality reduction based on explainable machine learning. (Nguyen, Abstract: “To address these problems, we propose a new dimension reduction-based method for privacy preservation. Our method generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. We first introduce a theoretical approach to evaluate dimension reduction-based privacy preserving mechanisms, then propose a non-linear dimension reduction framework motivated by state-of-the-art neural network structures for privacy preservation.”)
Regarding Claim 16, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 9 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
said benefit-to-resource score exceeds a threshold and said autoencoder architecture is selected to minimize reconstruction loss. (Nguyen, Page 98, Col. 1, Section 3.5: “In order to meet a certain level of reconstruction distance, we consider the constrained problem:
PNG
media_image4.png
68
585
media_image4.png
Greyscale
. The optimization problem above can be approximated as an unconstrained problem:
PNG
media_image5.png
49
575
media_image5.png
Greyscale
, where γ is a penalty parameter and C is a penalty function
PNG
media_image6.png
39
578
media_image6.png
Greyscale
Note that C is nonnegative, and C(θ ) = 0 iff the constraint in (8) is satisfied.”) [Examiner’s note: Nguyen discloses architectures (or models) are optimized to meet a certain reconstruction accuracy (within the threshold ϵ) by minimizing the reconstruction loss with this constraint]
Regarding Claim 17, Zhang explicitly discloses:
A computer program product for privacy-driven data sharing, said computer program product comprising a computer readable storage medium having program instructions embodied therewith, (Zhang, Col. 3, Lines 40-43: “In another aspect, the present disclosure provides a persistent memory storage, having stored thereon a set of instructions that, when executed by a processor, cause the processor to initiate a resource-aware runtime scheduler”)
said program instructions executable by a processor to cause said processor to perform a function, said function comprising: (Zhang, Col. 3, Lines 40-43: “In another aspect, the present disclosure provides a persistent memory storage, having stored thereon a set of instructions that, when executed by a processor, cause the processor to initiate a resource-aware runtime scheduler”)
selecting an autoencoder architecture based on said benefit-to-resource score (Zhang, Col. 6, Lines 20-27: “Given the cost functions of all the concurrently running vision applications, a device can employ a runtime scheduler that selects the most suitable sub-model of a multi-capacity model for each application and determines the optimal amount of runtime resources to allocate to each selected sub-model to jointly maximize the accuracy and minimize the inference latency of concurrent vision applications.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the same framework can be applied to other types of neural network including an autoencoder. So, “selecting an autoencoder architecture” is being interpreted as “selects the most suitable sub-model”]
wherein said autoencoder architecture balances minimizing reconstruction loss and minimizing required storage space for a dimensionality reduction action based on said benefit-to- resource score; (Zhang, Abstract: “The multi-capacity framework enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the framework of balancing between performance and resources consumed can also be used for other type of neural networks such as autoencoder]
Zhang fails to disclose:
computing a benefit-to-resource score for a dataset of user personal data by combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data;
a dimensionality reduction action
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture; and
storing said transformed data in a user space,
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permissions.
sending only the segment of the transformed data over a network to a user
device.
However, Burke explicitly discloses:
computing a benefit-to-resource score for a dataset of user personal data; (Burke, Pg. 833, Col. 2, Section A.: “Providing the service provider with a useful but privacy protected dataset is vital to ensuring the validity of the results returned and the anonymity of the reporters (users).”, Pg. 835, Figure 1:
PNG
media_image1.png
448
571
media_image1.png
Greyscale
,Pg. 837, Col. 1, ¶[2]: “The utility of the anonymized dataset indicates how useful it is, in terms of capacity for returning useful information in response to queries, in comparison to the original (non-anonymized) data set. Data utility is computed as follows:
PNG
media_image2.png
98
490
media_image2.png
Greyscale
”) [Examiner’s note: the user personal data is being interpreted as the crime report data]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Burke. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Burke teaches an anonymization model used for privacy preserving crime data publishing in resource constrained environments. One of ordinary skill would have motivation to combine Zhang and Burke to improve performance, and minimize information loss while maximizing classification accuracy.
However, Thomas explicitly discloses:
combining i) a benefit value comprising a manual input and an automatically collected input with ii) a cost value comprising resources to collect, maintain and transmit data (Thomas, Col. 4, Lines 5-9: “data privacy-utility tradeoff calculator configured to: compute a utility index based on midpoint of the balanced buckets and the privacy data; and compute attribute variations based on the number of variations between the buckets and the balanced buckets”, Col. 7, Lines 7-10: “FIG. 3A and FIG. 3B represent an exemplary flow diagram illustrating a computer implemented method 200 for computing the data privacy-utility tradeoff, in accordance with an embodiment of the present disclosure”, Col. 7, Lines 13-37: “In an embodiment, at step 202, the system 100 is configured to receive via a data connector 108A, data generated or captured from one or more data sources (Data Source 1, Data Source 2, ... Data Source n) to make it consumable by one or more data buyers. In an embodiment, the data connector 108A can also be configured to receive a query and other parameters along with the data. For instance, the query may be in the form of a request for data pertaining to people in the age group 40-50 years and suffering from diabetes or people in a particular region having heart problems, and the like, wherein parameters of the query may include age, gender, place of residence, and the like. In accordance with the present disclosure, the received data is analyzed or processed to make it suitable for dissemination to one or more data buyers without losing its utility in the process of retaining privacy associated with the data. In an embodiment, the data connector 108A connects to the one or more data sources to receive generated or captured data in bulk format or streaming content format. Data in bulk format may be uploaded at pre-determined intervals or randomly by data sellers. Data in streaming content format may be data provided in real time by connecting with the one or more data sources such as Fitbit™ devices, accelerometer devices, temperature devices and electrical consumption devices.”, Col. 7, Lines 63-67: “generate an adversary model by partitioning the search space into sets of buckets; each set corresponding to a sensitive attribute having a privacy data associated thereof. For instance a bucket set may pertain to age and include buckets [0-9], [10-19], and the like. Another bucket set may pertain to diseases and include [diabetes], [heart disease], [lung disease], and the like.”) [Examiner’s note: Under BRI, the “benefit value” is being interpreted as privacy value of user data and this value is received by a manual query from buyers or automatically streamed from more data sources such as Fitbit devices, “cost value” is being interpreted as the “attribute variations” based on the information variation between buckets wherein the bucket contains set of sensitive attributes (i.e., function of the bucket if to collect, maintain and transmit data)]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Thomas. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Thomas teaches systems and methods for computing data privacy-utility tradeoff. One of ordinary skill would have motivation to combine Zhang and Thomas because MPEP 2143 sets forth the Supreme Court rationales for obviousness including: (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results; (E): “Obvious to try” choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success; (F) Known work in one field of endeavor may prompt variations of it for use in either the same field or a different one based on design incentives or other market forces if the variations are predictable to one of the ordinary skill in the art.
However, Nguyen explicitly discloses:
a dimensionality reduction action (Nguyen, Abstract: “We first introduce theoretical approach to evaluate dimension reduction-based privacy preserving mechanisms, then propose a non-linear dimension reduction framework motivated by state-of-the-art neural network structures for privacy preservation”)
performing the dimensionality reduction action by transforming said dataset into transformed data with a transformation function based on said autoencoder architecture; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
storing said transformed data in a user space, (Nguyen, Page 95, Col. 1, ¶[2]: “Our dimension reduction (DR) framework can be applied to different types of data and used in several practical applications without heavy computation of encryption and impact of query number. The proposed framework can be applied directly to the access control system mentioned above. More elaboratively, face images are locally collected, nonlinearly compressed to achieve DR, and sent to the authentication center.”) [Examiner’s note: a user space i.e., the authentication center]
wherein said user space determines whether or not to submit a segment of said transformed data to a requester based at least in part on user permissions. (Nguyen, Page 96, Col. 2, Section 3.1: “For example, if member n requests to access web server 2, the local device first takes a facial photo of the member by an attached camera, locally transforms it into lower dimension data, and sends to an authentication center. The authentication server then obtains the low dimensional data and determines member access eligibility by using a classifier without clear face images of the requesting member.”, Page 99, Col. 1, ¶[1]: “We randomly divide the dataset into two groups of subjects and labels their images to (1) or (0) depending on their access permission.”) [Examiner’s note: user space i.e., the authentication server, a segment of transformed data i.e., a facial photo of the member]
sending only the segment of the transformed data over a network to a user
device. (Nguyen, Pg. 96, Col. 2, Section 3.1: “the local device first takes a facial photo of the member by an attached camera, locally transforms it into lower dimension data, and sends to an authentication center. The authentication server then obtains the low dimensional
data and determines member access eligibility by using a classifier without clear face images of the requesting member”) [Examiner’s note: Under BRI, this corresponds to transforming original user data on the user device, resulting in transformed data that is distinct from the original raw data (i.e., the full face image)]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang and Nguyen. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Nguyen teaches a new dimension reduction-based method for privacy preservation that generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. One of ordinary skill would have motivation to combine Zhang and Nguyen to achieve privacy-preserving via dimension reduction by transforming the data through an encoder. This allows sensitive or identifiable information to be obscured and the transformed data represents an encoded form of the original data, which might not be directly interpretable, thus enhancing privacy (Nguyen, Page 97, Col. 1, Section 3.3, ¶1])
Regarding Claim 18, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 17 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
said benefit-to-resource score exceeds a threshold and said autoencoder architecture is selected to minimize reconstruction loss. (Nguyen, Page 98, Col. 1, Section 3.5: “In order to meet a certain level of reconstruction distance, we consider the constrained problem:
PNG
media_image4.png
68
585
media_image4.png
Greyscale
. The optimization problem above can be approximated as an unconstrained problem:
PNG
media_image5.png
49
575
media_image5.png
Greyscale
, where γ is a penalty parameter and C is a penalty function
PNG
media_image6.png
39
578
media_image6.png
Greyscale
Note that C is nonnegative, and C(θ ) = 0 iff the constraint in (8) is satisfied.”) [Examiner’s note: Nguyen discloses architectures (or models) are optimized to meet a certain reconstruction accuracy (within the threshold ϵ) by minimizing the reconstruction loss with this constraint]
Regarding Claim 19, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 17 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
wherein said function further comprises: segmenting said dataset into data segments; (Zhang, Col. 11, Lines 3-11: “To demonstrate how this can be done in one embodiment, let {anc, pos, neg} denote a triplet that consists of an anchor input dataset (anc) such as an anchor image, a positive input dataset (pos) such as a positive image, and a negative input dataset (neg) such as a negative image where the anchor dataset and the positive dataset are from the same class (e.g., reflect the same sensed condition), while the negative image is from a different class (e.g., reflects a different sensed condition).”)
computing a segment benefit-to-resource score for each of said data segments; (Zhang, Col. 16, Lines 7-22: “Additionally, let Mv denote the multi capacity model of the application v, and let mv denote a descendant model mv ϵ Mv. The cost function of the descendant model mv from v is defined as follows:
PNG
media_image7.png
68
488
media_image7.png
Greyscale
where A(mv) is the inference accuracy of mv, uv ϵ [0, 1] is the computing resource percentage allocated to v, is the inference accuracy of mv, and L(mv) is the inference latency of mv when 100% of the computing resources the descendant model might consume are allocated to v, and αv ϵ [0, 1] is the weight that determines the accuracy-latency tradeoff for v.”) [Examiner’s note: “a dataset” is being interpreted as “application v”, “computing a benefit-to-resource score” is being interpreted as “determines the accuracy-latency tradeoff for v”]
selecting a segment autoencoder architecture for each of said data segments based on said segment benefit-to-resource scores; and (Zhang, Col. 6, Lines 20-27: “Given the cost functions of all the concurrently running vision applications, a device can employ a runtime scheduler that selects the most suitable sub-model of a multi-capacity model for each application and determines the optimal amount of runtime resources to allocate to each selected sub-model to jointly maximize the accuracy and minimize the inference latency of concurrent vision applications.”, Zhang, Col. 6, Lines 53-60: “However, based on the discussion of this method and the content of this disclosure, it will be understood that the concepts shown here for CNN architectures could also be used for other types of neural networks, such as a recurrent neural network ( e.g., LSTM), other types of feedforward neural networks besides CNNs (e.g., probabilistic, autoencoder, and time delay), RBF, GRNN, and modular.”) [Examiner’s note: Zhang discloses the same framework can be applied to other types of neural network including an autoencoder. So, “selecting an autoencoder architecture” is being interpreted as “selects the most suitable sub-model”]
leveraging at least one of said segment autoencoder architectures to transform at least one of said data segments into at least one transformed data segments; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
transforming said at least one of said data segments into at least one transformed data segments. (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
Regarding Claim 20, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 17 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
segmenting said dataset into data segments according to type of personal data; (Burke, Pg. 838, Col. 2, Section A: “Two relevant non sensitive attributes were specified (age and gender) as well as one sensitive attribute. In one experiment, the “Reports” dataset was divided into subsets of different sizes, starting at 10 tuples and stopping at 10000 tuples. The intuition behind this decision being that a crime reporting service will start out with a small number of reports, but this can and likely will grow exponentially as the system is implemented and used.”, Pg. 838, Col. 2, Section B: “For consistency with Datafly, Incognito, and Mondrian, in measuring processing time, the attribute “age” was selected as quasi-identifier and the “crime” category as the sensitive value.”)
computing a segment benefit-to-resource score for a first data segment; (Burke, Pg. 833, Col. 2, Section A.: “Providing the service provider with a useful but privacy protected dataset is vital to ensuring the validity of the results returned and the anonymity of the reporters (users).”, Pg. 835, Figure 1:
PNG
media_image1.png
448
571
media_image1.png
Greyscale
,Pg. 837, Col. 1, ¶[2]: “The utility of the anonymized dataset indicates how useful it is, in terms of capacity for returning useful information in response to queries, in comparison to the original (non-anonymized) data set. Data utility is computed as follows:
PNG
media_image2.png
98
490
media_image2.png
Greyscale
”) [Examiner’s note: the user personal data is being interpreted as the crime report data]
transforming said data segments into transformed data segments; and (Nguyen, Page 97, Col. 1, Section 3.4: “We leverage the structure of an auto-encoder [22] which contains encoder and decoder (in this work, we called them generator and re-constructor) in order to reduce data dimension.”, Page 97, Col. 2, ¶[2]: “Let X be the public training dataset. (xi , yi) is the ith sample in the dataset in which each sample xi has d features and a ground truth label yi. The system is aimed at learning a dimension reduction transformation F ( · ) which transforms the data from d dimensions to d dimensions in which d’ << d.”) [Examiner’s note: a transformation function i.e., function F ( · )]
streaming one or more of said transformed data segments to a content personalizer. (Nguyen, Page 95, Col. 1, ¶[2]: “Our dimension reduction (DR) framework can be applied to different types of data and used in several practical applications without heavy computation of encryption and impact of query number. The proposed framework can be applied directly to the access control system mentioned above. More elaboratively, face images are locally collected, nonlinearly compressed to achieve DR, and sent to the authentication center. The server then performs classification tasks on the dimension-reduced data.”, Page 95, Fig. 1:
PNG
media_image3.png
509
655
media_image3.png
Greyscale
) [Examiner’s note: a content personalizer i.e., the web servers]
Claim(s) 3, 11 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Nguyen, Burke, and in further view of Sheikhalishahi & Martinelli (“Privacy-Utility Feature Selection as a Privacy Mechanism in Collaborative Data Classification”) (hereafter referred to as “Sheikhalishahi”)
Regarding Claim 3, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 1 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
segmenting said dataset into data segments; (Burke, Pg. 838, Col. 2, Section A: “Two relevant non sensitive attributes were specified (age and gender) as well as one sensitive attribute. In one experiment, the “Reports” dataset was divided into subsets of different sizes, starting at 10 tuples and stopping at 10000 tuples. The intuition behind this decision being that a crime reporting service will start out with a small number of reports, but this can and likely will grow exponentially as the system is implemented and used.”, Pg. 838, Col. 2, Section B: “For consistency with Datafly, Incognito, and Mondrian, in measuring processing time, the attribute “age” was selected as quasi-identifier and the “crime” category as the sensitive value.”)
Zhang in view of Nguyen and Burke fails to disclose:
determining a weightage for each of said data segments wherein said weightage is based on a semantic purpose of an analytics service; and
enabling reduce-transformation intensities in said autoencoder architecture according to said weightage.
However, Sheikhalishahi explicitly discloses:
determining a weightage for each of said data segments wherein said weightage is based on a semantic purpose of an analytics service; and (Sheikhalishahi, Page 245, Col. 2, Section IV., ¶[a]: “The aim of feature selection in data mining algorithms is to obtain an optimal set of features such that it contains all relevant features which are not redundant in terms of identifying the class labels of a dataset… The filter techniques in feature selection, with the use of general characteristics of the data, rank the features based on their relevance. More precisely, feature ranking creates a scoring function, say υ(j), computed from the impact of a feature Aj in discriminating class labels. Generally, a high score is indicative of a valuable feature.”) [Examiner’s note: determining a weightage of each data segments i.e., raking the features based on their relevance, a semantic purpose of an analytics service i.e., identifying class labels of a dataset]
enabling reduce-transformation intensities in said autoencoder architecture according to said weightage. (Sheikhalishahi, Page 248, TABLE I:
PNG
media_image8.png
496
650
media_image8.png
Greyscale
, Page 248, Col. 2, Section VI, ¶[6]: “As it can be inferred from Table I, the feature “Bone Marrow” has the minimum trade-off score, and so it is the first candidate to be removed. The result of its removal does not provide the required privacy of any party (Table II, first column). Hence, the next feature having the minimum privacy-utility trade-off score, i.e. “Skin”, is selected to be removed. Still the minimum privacy threshold is not respected for any party (Table II, second column). Hence, the next feature, i.e. “Brain” is removed. After eliminating this feature, the privacy requirement of all parties, i.e. θ ≥ 0.1, is satisfied (Table II, third column)”, Page 249, TABLE II:
PNG
media_image9.png
188
642
media_image9.png
Greyscale
, Page 249, Col. 1, ¶[3]: “From the results of Tables II and III, it can be inferred that by removing a set of irrelevant features in terms of privacy-utility trade-off, general privacy gain improves, while at the same time the dataset remain practically useful.”) [Examiner’s note: weightage i.e., the trade-off score which shows the relevance of each features, “enabling reduce-transformation intensities” is being interpreted as keeping the most relevant features that has the highest trade-off scores]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang, Nguyen, Burke and Sheikhalishahi. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Nguyen teaches a new dimension reduction-based method for privacy preservation that generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. Burke teaches an anonymization model used for privacy preserving crime data publishing in resource constrained environments. Sheikhalishahi teaches a privacy mechanism based on privacy-utility feature selection, which by eliminating the most irrelevant set of features in terms of accuracy and privacy, guarantees the privacy requirements of data providers, whilst the data remain practically useful for classification. One of ordinary skill would have motivation to combine Zhang, Burke, Nguyen and Sheikhalishahi because MPEP 2143 sets forth the Supreme Court rationales for obviousness including: (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results; (E) “Obvious to try” choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success; (F) Known work in one field of endeavor may prompt variations of it for use in either the same field or a different one based on design incentives or other market forces if the variations are predictable to one of the ordinary skilled in the art.
Regarding Claim 11, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 9 (as shown in the rejection above).
Zhang in view of Nguyen and Burke further discloses:
segmenting said dataset into data segments; (Burke, Pg. 838, Col. 2, Section A: “Two relevant non sensitive attributes were specified (age and gender) as well as one sensitive attribute. In one experiment, the “Reports” dataset was divided into subsets of different sizes, starting at 10 tuples and stopping at 10000 tuples. The intuition behind this decision being that a crime reporting service will start out with a small number of reports, but this can and likely will grow exponentially as the system is implemented and used.”, Pg. 838, Col. 2, Section B: “For consistency with Datafly, Incognito, and Mondrian, in measuring processing time, the attribute “age” was selected as quasi-identifier and the “crime” category as the sensitive value.”)
Zhang in view of Nguyen and Burke fails to disclose:
determining a weightage for each of said data segments wherein said weightage is based on a semantic purpose of an analytics service; and
enabling reduce-transformation intensities in said autoencoder architecture according to said weightage.
However, Sheikhalishahi explicitly discloses:
determining a weightage for each of said data segments wherein said weightage is based on a semantic purpose of an analytics service; and (Sheikhalishahi, Page 245, Col. 2, Section IV., ¶[a]: “The aim of feature selection in data mining algorithms is to obtain an optimal set of features such that it contains all relevant features which are not redundant in terms of identifying the class labels of a dataset… The filter techniques in feature selection, with the use of general characteristics of the data, rank the features based on their relevance. More precisely, feature ranking creates a scoring function, say υ(j), computed from the impact of a feature Aj in discriminating class labels. Generally, a high score is indicative of a valuable feature.”) [Examiner’s note: determining a weightage of each data segments i.e., raking the features based on their relevance, a semantic purpose of an analytics service i.e., identifying class labels of a dataset]
enabling reduce-transformation intensities in said autoencoder architecture according to said weightage. (Sheikhalishahi, Page 248, TABLE I:
PNG
media_image8.png
496
650
media_image8.png
Greyscale
, Page 248, Col. 2, Section VI, ¶[6]: “As it can be inferred from Table I, the feature “Bone Marrow” has the minimum trade-off score, and so it is the first candidate to be removed. The result of its removal does not provide the required privacy of any party (Table II, first column). Hence, the next feature having the minimum privacy-utility trade-off score, i.e. “Skin”, is selected to be removed. Still the minimum privacy threshold is not respected for any party (Table II, second column). Hence, the next feature, i.e. “Brain” is removed. After eliminating this feature, the privacy requirement of all parties, i.e. θ ≥ 0.1, is satisfied (Table II, third column)”, Page 249, TABLE II:
PNG
media_image9.png
188
642
media_image9.png
Greyscale
, Page 249, Col. 1, ¶[3]: “From the results of Tables II and III, it can be inferred that by removing a set of irrelevant features in terms of privacy-utility trade-off, general privacy gain improves, while at the same time the dataset remain practically useful.”) [Examiner’s note: weightage i.e., the trade-off score which shows the relevance of each features, “enabling reduce-transformation intensities” is being interpreted as keeping the most relevant features that has the highest trade-off scores]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang, Nguyen, Burke and Sheikhalishahi. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Burke teaches an anonymization model used for privacy preserving crime data publishing in resource constrained environments. Nguyen teaches a new dimension reduction-based method for privacy preservation that generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. Sheikhalishahi teaches a privacy mechanism based on privacy-utility feature selection, which by eliminating the most irrelevant set of features in terms of accuracy and privacy, guarantees the privacy requirements of data providers, whilst the data remain practically useful for classification. One of ordinary skill would have motivation to combine Zhang, Nguyen, Burke and Sheikhalishahi because MPEP 2143 sets forth the Supreme Court rationales for obviousness including: (D) Applying a known technique to a known device (method, or product) ready for improvement to yield predictable results; (E) “Obvious to try” choosing from a finite number of identified, predictable solutions, with a reasonable expectation of success; (F) Known work in one field of endeavor may prompt variations of it for use in either the same field or a different one based on design incentives or other market forces if the variations are predictable to one of the ordinary skilled in the art.
Claim(s) 7, 15 are rejected under 35 U.S.C. 103 as being unpatentable over Zhang in view of Nguyen, Burke and in further view of Maree et al (“Towards Responsible AI for Financial Transactions”) (hereafter referred to as “Maree”)
Regarding Claim 7, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 1 (as shown in the rejections above).
Zhang in view of Nguyen and Burke fails to disclose:
said autoencoder architecture is additionally based on dominant features extracted from Shapley additive explanations analysis.
However, Maree explicitly discloses:
said autoencoder architecture is additionally based on dominant features extracted from Shapley additive explanations analysis. (Maree, Page 17, Section IV.A, Col. 2: “The features in the dataset include categorical, numerical and text attributes. The target model is a series of two opaque models: a word2vec encoder followed by a deep neural network (DNN). In the first model, the transaction text is encoded into a vector representation
PNG
media_image10.png
55
424
media_image10.png
Greyscale
”, and Section IV.B, Col. 2: “Shapley additive explanations (SHAP) [10] offers an alternative, mathematically sound and parsimonious approach to salient feature extraction. SHAP is based on the collaborative game theory method, Shapley values [15]. It clarifies the prediction of an instance x ϵ X, where X is the set of all instances, by computing the contribution of each input feature xi ϵ x, i ϵ {1, …, N} where N is the number of features in the dataset.”) [The examiner interprets “dominant features extracted” here is the “salient feature extraction”]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang, Nguyen, Burke and Maree. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Nguyen teaches a new dimension reduction-based method for privacy preservation that generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. Burke teaches an anonymization model used for privacy preserving crime data publishing in resource constrained environments. Maree teaches identifying salient features of the transaction classification model and extracting an explanation for the function of the model. One of ordinary skill would have motivation to combine Zhang, Nguyen, Burke and Maree to improve the prediction accuracy by assessing the usefulness of the additional extracted feature by using SHAP values (Maree, Page 17, Section III, Col. 1, ¶[3])
Regarding Claim 15, the combination of Zhang, Burke and Nguyen discloses all the limitations of Claim 9 (as shown in the rejections above).
Zhang in view of Nguyen and Burke fails to disclose:
said autoencoder architecture is additionally based on dominant features extracted from Shapley additive explanations analysis.
However, Maree explicitly discloses:
said autoencoder architecture is additionally based on dominant features extracted from Shapley additive explanations analysis. (Maree, Page 17, Section IV.A, Col. 2: “The features in the dataset include categorical, numerical and text attributes. The target model is a series of two opaque models: a word2vec encoder followed by a deep neural network (DNN). In the first model, the transaction text is encoded into a vector representation
PNG
media_image10.png
55
424
media_image10.png
Greyscale
”, and Section IV.B, Col. 2: “Shapley additive explanations (SHAP) [10] offers an alternative, mathematically sound and parsimonious approach to salient feature extraction. SHAP is based on the collaborative game theory method, Shapley values [15]. It clarifies the prediction of an instance x ϵ X, where X is the set of all instances, by computing the contribution of each input feature xi ϵ x, i ϵ {1, …, N} where N is the number of features in the dataset.”) [The examiner interprets “dominant features extracted” here is the “salient feature extraction”]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Zhang, Nguyen, Burke and Maree. Zhang teaches a multi-capacity framework which enables deep learning models to offer flexible resource-accuracy trade-offs and other similar balancing of performance and resources consumed. Nguyen teaches a new dimension reduction-based method for privacy preservation that generates dimension-reduced data for performing machine learning tasks and prevents a strong adversary from reconstructing the original data. Burke teaches an anonymization model used for privacy preserving crime data publishing in resource constrained environments. Maree teaches identifying salient features of the transaction classification model and extracting an explanation for the function of the model. One of ordinary skill would have motivation to combine Zhang, Nguyen, Burke and Maree to improve the prediction accuracy by assessing the usefulness of the additional extracted feature by using SHAP values (Maree, Page 17, Section III, Col. 1, ¶[3])
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 AMY TRAN whose telephone number is (571)270-0693. The examiner can normally be reached Monday - Friday 7:30 am - 5:00 pm EST.
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, David Yi can be reached at (571) 270-7519. 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.
/AMY TRAN/Examiner, Art Unit 2126
/VAN C MANG/
Primary Examiner, Art Unit 2126
Prosecution Insights