DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under
the first inventor to file provisions of the AIA .
Priority
Acknowledgment is made of applicant’s claim for foreign priority under 35 U.S.C. 119 (a)-(d). The certified copy has been filed in the current application, filed on 06/14/2021.
Status of Claims
Applicant’s submission filed 07/03/2025 has been entered. The status of the claims is as follows:
Claims 1-13 remain pending in the application.
Claims 1, 6, and 13 are amended.
Response to Arguments
In reference to the 35 U.S.C 101 Claim Rejection:
Regarding the arguments of independent claims 1 and 13 and Prong One of Step 2A:
Applicant asserts (see Remarks pages 11-13) that the amended independent claims recite specific technical features directed to securely performing prediction processing using an encrypted prediction model and user information maintained in a secret state. According to the Applicant, the invention ensures privacy by distributing and encrypting the prediction model through a secret sharing method and processing user input without revealing private data. The Applicant further contends that the claimed conversion of negative numerical values to positive ones for homogenization processing reduces computational complexity, avoids integration processing, and improves prediction accuracy. Overall, the Applicant asserts that these features collectively provide technical advantages beyond abstract data processing
Examiner respectfully disagrees and notes that the amendments to the independent claim do not integrate the abstract idea into a practical application. The recited “secret sharing” and “secret state” merely describe conventional cryptographic and data privacy techniques implemented with generic computing components. The claimed steps do not improve the functioning of the computer or any other technology or technical field, but instead use a computer as a tool to execute abstract idea of securely processing and converting data. Applicant’s amendment further adds that “integerization processing is avoided for reducing an amount of computation of the prediction processing, improving accuracy of the prediction processing, and reducing a drop in prediction accuracy ”. However, this additional limitation does not change the eligibility determination. The recited feature merely describes a mathematical operation and its intended benefit, namely avoiding a particular type of numerical conversion (integerization processing) to achieve improved computational efficiency or prediction accuracy. Such language represents an abstract optimization of mathematical processing rather than a specific improvement to computer technology or functioning. The claim does not recite any particular technical means by which the computer’s hardware or software is improved, nor does it alter how the computer performs its basic operations. Instead, it describes the expected result of applying known mathematical principles. Therefore, this limitation does not integrate the abstract idea into a practical application under Step 2A.
Applicant’s arguments filed 07/03/2025 have been fully considered but they are not persuasive.
Regarding the arguments of Prong 2 of Revised Step 2A (and/ or Step 2B):
Applicant asserts that (see Remarks pages 13-14) the amended independent claims now integrate any alleged abstract idea into a practical application under Step 2A, Prong 2. Specifically, the Applicant argues that the claimed features, such as encrypting through secret sharing, processing user information in a secret state, and converting negative numerical values to positive values, allow a service provider to perform prediction processing while keeping the user’s information private, thereby improving computer functioning. The Applicant further asserts that avoiding integerization processing and adding a divisor to convert numerical values enhance the operation of the underlying device by reducing computation and improving accuracy. Based on these asserted technical effects, the Applicant maintains that the claims are not directed to an abstract idea and requests withdrawal of the 101 rejection.
Examiner respectfully disagrees and notes that the amended claim features, such as encrypting through a secret sharing method, processing user information in a secret state, converting negative to positive numerical values, and avoiding integerization processing, remain directed to mathematical concepts and data manipulation. These steps describe abstract data processing operations performed by a generic computer and do not improve the functioning of the computer or any other technology. The assertions that such features reduce computation or improve accuracy are result-oriented and do not specify any particular technical means by which such effects are achieved. Under Step 2A, the amendments do not integrate the abstract idea into a practical application and under Step 2B, they fail to provide an inventive concept beyond the abstract idea itself.
Applicant’s arguments filed 07/03/2025 have been fully considered but they are not persuasive.
In reference to the claim rejection under 35 U.S.C 103:
Applicant’s arguments filed 07/03/2025, 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 Fan et al (US 11,245,522 B2).
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 1-13 are rejected under U.S.C 101 for containing 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.
converting, [by the data providing device], a prediction model by converting at least one parameter which is included in the prediction model and is performing homogenization processing into at least one parameter for performing processing including nonlinear processing - 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.) because according to the Instant Specification Page 8, Lines 12-22, the step of converting the parameter and performing homogenization process involves performing an equation yi = si xi + ti, where xi is an input and yi is an output, si and ti may be the plurality of parameters for performing the homogenization process, which includes the nonlinear process performed by equation (1):
PNG
media_image1.png
99
401
media_image1.png
Greyscale
generating, [by the data providing device], an encrypted prediction model that performs prediction processing with input in a secret state remaining secret by encrypting the prediction model that has been converted. - 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.)
in the converting, at least one negative numerical value is converted to a positive numerical value by converting the at least one parameter for performing the homogenization processing into the at least one parameter for performing the processing including the nonlinear processing - 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.)
and integerization processing is avoided for reducing an amount of computation of the prediction processing, improving accuracy of the prediction processing, and reducing a drop in prediction accuracy, 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.)
a divisor used in the secret sharing method is added to the negative numerical value to convert the negative numerical value to the positive numerical value.
This limitation is directed to mathematical calculation (see MPEP 2106.04(a)(2) l. 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 the data providing device – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
the prediction model being a neural network – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
receiving, as the input, user information, the user information being in the secret state; This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
performing the prediction processing using the encrypted prediction model and the user information in the secret state; and - 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.
presenting, as at least one of an image or audio, prediction results to a user based on the prediction processing, the user corresponding to the user information, This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
wherein, in the generating: the prediction model is encrypted by distributing, [through a secret sharing method], the prediction model that has been converted, to generate at least two distributed encrypted prediction models for keeping the prediction model secret and safely performing the prediction processing, This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
through a secret sharing method – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
in the prediction processing, the at least two distributed encrypted prediction models are applied to at least two distributed data by the secret sharing method, and 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.
at least two secret prediction results are output in the secret state, and – This limitation is directed to mere data outputting (see MPEP 2106.05(g))
the at least two secret prediction results, in the secret state, are necessary to obtain a decrypted prediction result. – This limitation is directed to mere data outputting (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 the data providing device – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
the prediction model being a neural network – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
receiving, as the input, user information, the user information being in the secret state; This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)). This limitation is directed to receiving or transmitting data over a network, which the courts (as per Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362) 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.).
performing the prediction processing using the encrypted prediction model and the user information in the secret state; and - 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.
presenting, as at least one of an image or audio, prediction results to a user based on the prediction processing, the user corresponding to the user information, This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)). This limitation is directed to receiving or transmitting data over a network, which the courts (as per Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362) 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.).
wherein, in the generating: the prediction model is encrypted by distributing, [through a secret sharing method], the prediction model that has been converted, to generate at least two distributed encrypted prediction models for keeping the prediction model secret and safely performing the prediction processing, This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)). This limitation is directed to receiving or transmitting data over a network, which the courts (as per Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362) 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.).
through a secret sharing method – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
in the prediction processing, the at least two distributed encrypted prediction models are applied to at least two distributed data by the secret sharing method, and 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.
at least two secret prediction results are output in the secret state, and – This limitation is directed to mere data outputting (see MPEP 2106.05(g))
the at least two secret prediction results, in the secret state, are necessary to obtain a decrypted prediction result. – This limitation is directed to mere data outputting (see MPEP 2106.05(g))
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:
wherein the at least one parameter for performing the homogenization processing comprises a plurality of parameters, – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
the at least one parameter for performing the processing including the nonlinear processing is one parameter, and – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
in the converting, the plurality of parameters for performing the homogenization processing is converted into the one parameter for performing the processing including the nonlinear processing - 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 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:
wherein the homogenization processing is processing performed by an equation yi = si xi + ti, where xi is an input and yi is an output, si and ti are the plurality of parameters for performing the homogenization processing - This limitation is directed to mathematical calculation as the homogenization processing is performed by using an equation (see MPEP 2106.04(a)(2) l. C.)
the processing including the nonlinear processing is processing performed by Equation
(1), and [Math 1]
PNG
media_image2.png
51
236
media_image2.png
Greyscale
, ki is the at least one parameter for performing the processing including the nonlinear processing, and is determined using si and ti - This limitation is directed to mathematical calculation as the nonlinear processing is performed by using an equation (see MPEP 2106.04(a)(2) l. C.)
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 3 which includes an abstract idea (see rejection for claim 3). The additional limitations:
wherein ki is expressed by Equation (2), [Math 2]
PNG
media_image3.png
94
286
media_image3.png
Greyscale
, where u is a theoretical maximum value during computation of the prediction processing, and p is a divisor used in the encrypting. - This limitation is directed to mathematical calculation as value ki is calculated by using an equation (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 1 which includes an abstract idea (see rejection for claim 1). The additional limitations:
the prediction model is encrypted by distributing, through a secret sharing method, the prediction model that has been converted, and – 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.).
in the distributing of the prediction model, the at least one parameter for performing the processing including the nonlinear processing is distributed. – 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 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 5 which includes an abstract idea (see rejection for claim 5). The additional limitations:
determining [[a]] the divisor used in the secret sharing method in a range greater than an element of the prediction model - 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 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:
wherein the prediction model is a binarized neural network including a plurality of parameters each comprising a binary value of -1 or 1. – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
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:
training the prediction model using training data collected in advance, - 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 a parameter obtained through the training as the at least one parameter for performing the homogenization processing is converted in the converting. - 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,
Claim 9 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 6 which includes an abstract idea (see rejection for claim 6). The additional limitations:
wherein in the converting, the divisor used in the secret sharing method is added to [[a]] the at least one negative numerical value in the prediction model to - 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.)
convert the at least one negative numerical value to [[a]] the positive
numerical value. - 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 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 1 which includes an abstract idea (see rejection for claim 1). The additional limitations:
wherein in the converting, [[a]] the at least one negative numerical value is
converted to [[a]] the positive numerical value by converting a numerical value
in a plurality of parameters included in the prediction model to a set including a sign part indicating a sign of the numerical value as 0 or 1 and a numerical value part indicating an absolute value of the numerical value. - 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 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 5 which includes an abstract idea (see rejection for claim 5). The additional limitations:
wherein, [in the prediction processing], a feature amount is calculated from data obtained by sensing; and - This limitation is directed to mathematical calculation (see MPEP 2106.04(a)(2) l. C.)
in the prediction processing – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
the feature amount that has been calculated is distributed, [through the secret sharing method]. – This limitation is directed to insignificant extra solution activity in Step 2A Prong 2 (see MPEP 2106.05(g)) and further amounts to receiving or transmitting data over a network, which the courts (as per Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362) 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.). under Step 2B
through the secret sharing method – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
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 11 which includes an abstract idea (see rejection for claim 11). The additional limitations:
wherein, [in the prediction processing], the prediction processing is executed by the prediction model that has been distributed, by inputting, to the prediction model that has been distributed, the feature amount that has been distributed, – 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.).
in the prediction processing – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
wherein the prediction processing includes the nonlinear processing, – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
and the nonlinear processing is processing of converting an input to the nonlinear processing into 1 when the input is 0 or a numerical value corresponding to a positive, and into a positive numerical value corresponding to -1 when the input is a numerical value corresponding to a negative. - 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 13:
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.
converting a prediction model by converting at least one parameter which is included in the prediction model and is for performing homogenization processing into at least one parameter for performing processing including nonlinear processing- - 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.) because according to the Instant Specification Page 8, Lines 12-22, the step of converting the parameter and performing homogenization process involves performing an equation yi = si xi + ti, where xi is an input and yi is an output, si and ti may be the plurality of parameters for performing the homogenization process, which includes the nonlinear process performed by equation (1):
PNG
media_image1.png
99
401
media_image1.png
Greyscale
generating an encrypted prediction model that performs prediction processing with input in a secret state remaining secret by encrypting the prediction model that has been converted. - 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.)
in the converting, at least one negative numerical value is converted to a positive numerical value by converting the at least one parameter for performing the homogenization processing into the at least one parameter for performing the processing including the nonlinear processing - 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.)
and integerization processing is avoided for reducing an amount of computation of the prediction processing, improving accuracy of the prediction processing, and reducing a drop in prediction accuracy, 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.)
a divisor used in the secret sharing method is added to the negative numerical value to convert the negative numerical value to the positive numerical value.
This limitation is directed to mathematical calculation (see MPEP 2106.04(a)(2) l. 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 processor; This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
a memory including a computer program, the computer program, when executed by the processor, causing the processor to perform functions, the functions including This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
the prediction model being a neural network – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
receiving, as the input, user information, the user information being in the secret state; This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
performing the prediction processing using the encrypted prediction model and the user information in the secret state; and - 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.
presenting, as at least one of an image or audio, prediction results to a user based on the prediction processing, the user corresponding to the user information, This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
wherein, in the generating: the prediction model is encrypted by distributing, [through a secret sharing method], the prediction model that has been converted, to generate at least two distributed encrypted prediction models for keeping the prediction model secret and safely performing the prediction processing, This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)).
through a secret sharing method – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
in the prediction processing, the at least two distributed encrypted prediction models are applied to at least two distributed data by the secret sharing method, and 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.
at least two secret prediction results are output in the secret state, and – This limitation is directed to mere data outputting (see MPEP 2106.05(g))
the at least two secret prediction results, in the secret state, are necessary to obtain a decrypted prediction result. – This limitation is directed to mere data outputting (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 processor; This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
a memory including a computer program, the computer program, when executed by the processor, causing the processor to perform functions, the functions including This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
the prediction model being a neural network – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
receiving, as the input, user information, the user information being in the secret state; This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)). This limitation is directed to receiving or transmitting data over a network, which the courts (as per Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362) 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.).
performing the prediction processing using the encrypted prediction model and the user information in the secret state; and - 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.
presenting, as at least one of an image or audio, prediction results to a user based on the prediction processing, the user corresponding to the user information, This limitation is directed to insignificant extra-solution activity (see MPEP 2106.05(g)). This limitation is directed to receiving or transmitting data over a network, which the courts (as per Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362) 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.).
wherein, in the generating: the prediction model is encrypted by distributing, [through a secret sharing method], the prediction model that has been converted, to generate at least two distributed encrypted prediction models for keeping the prediction model secret and safely performing the prediction processing, This limitation is directed to receiving or transmitting data over a network, which the courts (as per Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362) 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.).
through a secret sharing method – This limitation is directed to a computer merely used as a tool to perform an existing process (see MPEP 2106.05(f) (2)).
in the prediction processing, the at least two distributed encrypted prediction models are applied to at least two distributed data by the secret sharing method, and 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.
at least two secret prediction results are output in the secret state, and – This limitation is directed to mere data outputting (see MPEP 2106.05(g))
the at least two secret prediction results, in the secret state, are necessary to obtain a decrypted prediction result. – This limitation is directed to mere data outputting (see MPEP 2106.05(g)).
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, 5, 7, 8, 10 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Ghasemzadeh et al (“ReBNet: Residual Binarized Neural Network”) (hereafter referred to as “Ghasemzadeh”) in view of GOTO (US 2018/0089574 A1) (hereafter referred to as “Goto”) , Mohassel et al. (US 2020/0242466 A1) (hereafter referred to as “Mohassel”) and further in view of Fan et al. (“US 11,245,522 B2”) (hereafter referred to as “Fan”)
Regarding Claim 1, Ghasemzadeh explicitly discloses:
A prediction model conversion method for use in a data providing device, the prediction
model conversion method comprising: converting, by the data providing device, a prediction model by converting at least one parameter which is included in the prediction model and is performing homogenization processing into at least one parameter for performing processing including nonlinear processing, (Ghasemzadeh, Page 2, Col. 1, Section II. A, In the Binary dot product section, real-valued vectors with values like
PNG
media_image4.png
26
37
media_image4.png
Greyscale
and
PNG
media_image5.png
24
38
media_image5.png
Greyscale
are replaced by sign vectors (+- 1), and then encoded into binary vectors (e.g., -1
→
0 and +1
→
1). This is a direct conversion of negative values to positive binary representations, where a negative sign (-1) becomes 0. Figure 1 (top) in page 2 shows the original dot product which represents a homogenization processing:
PNG
media_image6.png
80
363
media_image6.png
Greyscale
because homogenization typically refers to linear or smoothing operations – like standard dot products, or typical batch normalization. Page 2, Col. 2, Section Binary batch-normalization: “It is often useful to normalize the result of the dot product
PNG
media_image7.png
25
106
media_image7.png
Greyscale
before feeding it to the binary activation function described above. A batch normalization layer converts each input y into
α
x y –
β
, where
α
a
n
d
β
are the parameters of the layer. Authors of [17] suggest combining batch-normalization and binary activation layers into a single thresholding layer. The cascade of the two layers computes the following:
PNG
media_image8.png
55
407
media_image8.png
Greyscale
”, Figure 1 (bottom):
PNG
media_image9.png
52
331
media_image9.png
Greyscale
) [Examiner’s note: The text and Figure 1 (bottom) show how the linear dot product (homogenization processing) is replaced by XNOR + Popcount, which is a nonlinear bitwise operation that maps to dot product results using a different computational mechanism. Additionally, the binary activation function replaces analog thresholding with a sign-based threshold – a nonlinear comparator-based transformation]
the prediction model being a neural network; (Ghasemzadeh, pg. 2, col. 1, section II.A: “Neural networks are composed of multiple convolution, fully-connected, activation, batch-normalization, and max-pooling layers. Binarization enables the use of a simpler equivalent for each layer as explained in this section.”)
Ghasemzadeh fails to disclose:
receiving, as the input, user information, the user information being in the secret state;
performing the prediction processing using the encrypted prediction model and the user
information in the secret state; and
presenting, as at least one of an image or audio, prediction results to a user based on the prediction processing, the user corresponding to the user information,
generating, by the data providing device, an encrypted prediction model that performs prediction processing with input in a secret state remaining secret by encrypting the prediction model that has been converted.
wherein, in the generating:the prediction model is encrypted by distributing, through a secret sharing method, the prediction model that has been converted, to generate at least two distributed encrypted prediction models for keeping the prediction model secret and safely performing the prediction processing,
in the prediction processing, the at least two distributed encrypted prediction models are applied to at least two distributed data by the secret sharing method, and at least two secret prediction results are output in the secret state, and
the at least two secret prediction results, in the secret state, are necessary to obtain a decrypted prediction result.
and integerization processing is avoided for reducing an amount of computation of the
prediction processing, improving accuracy of the prediction processing, and reducing a drop in prediction accuracy, and
a divisor used in the secret sharing method is added to the negative numerical value to
convert the negative numerical value to the positive numerical value.
However, Goto explicitly discloses:
generating, by the data providing device, an encrypted prediction model that performs prediction processing with input in a secret state remaining secret by encrypting the prediction model that has been converted. (Goto, ¶[0048]: “A data processing device 100 according to the present exemplary embodiment shown in FIG. 1 is intended to provide learning data to a cloud system 200 that generates a prediction model by performing machine learning.”, [0050]: “The encryption unit 20 encrypts the learning data so that a prediction model generated from the learning data in an unencrypted state and a prediction model generated from the learning data in an encrypted state have a corresponding relationship with each other in terms of parameters, numeric values, and operators. The data output unit 30 outputs the encrypted learning data to the cloud system 200.”, and [0051]: “Therefore, even when the learning data is encrypted, the cloud system 200 according to the present exemplary embodiment generates a prediction model that is similar to a prediction model generated when the learning data is not encrypted. Thus, the cloud system 200 according to the present exemplary embodiment can perform machine learning without executing decryption processing, even when data used in machine learning is encrypted.”) [The examiner interprets the process of “performing prediction without executing decryption processing” as “performs prediction processing with input in a secret state remaining secret”]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh and Goto. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Goto teaches generating prediction model in an encrypted state. One of ordinary skill would have motivation to combine Ghasemzadeh and Goto to protect user privacy and secure inference without trust. In many use cases (e.g., healthcare, finance, or personal data), the input data is sensitive; sending raw input data to a server for prediction exposes it to the model provider, so the input data should be encrypted to ensure that the server never sees the original data
However, Mohassel explicitly discloses:
receiving, as the input, user information, the user information being in the secret state; (Mohassel, ¶[0015]: “According to an embodiment, the private input data can be represented as integers (e.g., by shifting bits of floating-point numbers), and the training can involve multiplying these integers (and other intermediate values) and integer-represented weights”)
performing the prediction processing using the encrypted prediction model and the user
information in the secret state; and (Mohassel, ¶[0054]: “This competition phase can include multiplication of input the data by weights to obtain a predicted output.”, ¶[0056]: “Thus, if the machine learning used profile data of a user to predict actions by the user, each data item in the profile can be split among the two servers.”, ¶[0057]: “The sharing can be done in a secure manner.”)
presenting, as at least one of an image or audio, prediction results to a user based on the prediction processing, the user corresponding to the user information, (Mohassel, ¶[0003]: “For an example data sample, the input data can be the pixel values of an image, and the output data can be a classification of what is in the image (e.g., that the image is of a dog).”)
wherein, in the generating:the prediction model is encrypted by distributing, through a secret sharing method, the prediction model that has been converted, to generate at least two distributed encrypted prediction models for keeping the prediction model secret and safely performing the prediction processing, (Mohassel, ¶[0140]: “In some embodiments, the training data is secret shared between two servers S0 and S1 . We denote the shares of two values X and Y at the two servers by <X>0, <Y>0 and <X>1 , <Y>1 . In practice, the clients can distribute the shares between the two servers, or encrypt the first share using the public key of S0 and upload both the first encrypted share and the second plaintext share to S1 . S1 can then pass the encrypted shares to S0 to decrypt. Herein, both implementations are encompassed by secret sharing”, ¶0011]: “Using fully homomorphic encryption, the neural network model can make predictions on encrypted data. In this case, it is assumed that the neural network is trained on plaintext data and the model is known to one party who evaluates it on private data of another.”) [Examiner’s note: Mohassel discloses the neural network model is trained to make predictions using encrypted data, while the training data is secret shared to 2 different servers. This aligns with the concept of the prediction model is encrypted using secret sharing method then generate 2 separate encrypted prediction models]
in the prediction processing, the at least two distributed encrypted prediction models are applied to at least two distributed data by the secret sharing method, and at least two secret prediction results are output in the secret state, and (Mohassel, ¶[0039]: “As examples, the secret sharing can involve splitting a data item up into shares that require a sufficient number ( e.g., all) of training computers to reconstruct and/or encryption mechanisms where decryption requires collusion among the training computers.”, ¶[0140]: “In some embodiments, the training data is secret shared between two servers S0 and S1 . We denote the shares of two values X and Y at the two servers by <X>0, <Y>0 and <X>1 , <Y>1 . In practice, the clients can distribute the shares between the two servers, or encrypt the first share using the public key of S0 and upload both the first encrypted share and the second plaintext share to S1 . S1 can then pass the encrypted shares to S0 to decrypt. Herein, both implementations are encompassed by secret sharing”, ¶0011]: “Using fully homomorphic encryption, the neural network model can make predictions on encrypted data.) [Examiner’s note: Mohassel discloses training the neural network model to make predictions on encrypted data, wherein the training data is distributed for 2 servers using the secret sharing method which will end up with 2 encrypted distributed shares]
the at least two secret prediction results, in the secret state, are necessary to obtain a decrypted prediction result. (Mohassel, ¶0108]: “In some implementations, the sender prepares the garbled circuit by determining a truth table for each gate using the random numbers that replaced the two bits on the input wires. The output values are then encrypted ( e.g., using double-key symmetric encryption) with the random numbers from the truth table. Thus, one can only decrypt the gate only if one knows the two correct random numbers for a given output value.”) [Examiner’s note: “two secret prediction results” is being interpreted as the two random numbers needed to decrypt the gate]
and integerization processing is avoided for reducing an amount of computation of the
prediction processing, improving accuracy of the prediction processing, and reducing a drop in prediction accuracy, and (Mohassel, ¶[0015]: “A secret-shared result ( e.g., the delta value for updating the weights) can be truncated by truncating the secret-shared parts at the training computers, thereby allowing efficient computation and limiting the amount of memory for storing the integer values.”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh and Mohassel. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Mohassel teaches techniques for efficient implementations that allow multiple client computers to use their provided data in creating a machine learning model, without having to expose the private data. One of ordinary skill would have motivation to combine Ghasemzadeh and Mohassel to ensure that no single party can fully reconstruct the original data or the models without collaboration from other parties. This is critical form preserving privacy in domains such as healthcare, finance, or other sensitive fields where data confidentiality is paramount. (Mohassel, ¶[0215])
However, Fan explicitly discloses:
a divisor used in the secret sharing method is added to the negative numerical value to
convert the negative numerical value to the positive numerical value. (Fan, Col. 15, Lines 44-47: “The first block to be processed is selected in step 440. The data is converted from its coding scheme ( e.g. ASCII, Unicode, etc) to a numerical value (in decimal or binary). 45 In step 450, a set of divisors are selected, at random, from the memory 208. There are n such divisors in the set.”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh and Fan. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Fan teaches method and system for securely storing data using a secret sharing scheme. One of ordinary skill would have motivation to combine Ghasemzadeh and Fan 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.
Regarding Claim 2, the combination of Ghasemzadeh, Goto, Fan and Mohassel discloses all the limitations of Claim 1 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel further discloses:
wherein the at least one parameter for performing the homogenization processing comprises a plurality of parameters, (Goto, Figure 18:
PNG
media_image10.png
224
649
media_image10.png
Greyscale
and [0111]: “Specifically, the standardization component 221 standardizes data values of attribute X as shown in FIG. 18. FIG. 18 shows an example of the prediction data that has been standardized by the prediction application in the exemplary embodiment of the present invention. In the example of FIG. 18, processing for normalizing data values of attribute X in a range of -1 to +1 is executed as standardization processing.”) [Figure 18 discloses multiple standardized values in range of -1 to 1 for attribute X. The examiner interprets the “homogenization processing” as the “standardization processing”]
the at least one parameter for performing the processing including the nonlinear processing is one parameter, and (Goto, [0111]: “The standardization component 221 transfers the prediction data in which the attribute targeted for standardization has been standardized (see FIG. 18) to the binarization component 222.”) [The examiner interprets the “nonlinear processing” as the “binarization processing”. The highlight indicates that at least one parameter of the standardization process includes the parameter of binarization process (i.e., the nonlinear processing)]
in the converting, the plurality of parameters for performing the homogenization processing is converted into the one parameter for performing the processing including the nonlinear processing. (Ghasemzadeh, Page 2, Col. 1, Section II. A, In the Binary dot product section, real-valued vectors with values like
PNG
media_image4.png
26
37
media_image4.png
Greyscale
and
PNG
media_image5.png
24
38
media_image5.png
Greyscale
are replaced by sign vectors (+- 1), and then encoded into binary vectors (e.g., -1
→
0 and +1
→
1). This is a direct conversion of negative values to positive binary representations, where a negative sign (-1) becomes 0. Figure 1 (top) in page 2 shows the original dot product which represents a homogenization processing:
PNG
media_image6.png
80
363
media_image6.png
Greyscale
because homogenization typically refers to linear or smoothing operations – like standard dot products, or typical batch normalization. Page 2, Col. 2, Section Binary batch-normalization: “It is often useful to normalize the result of the dot product
PNG
media_image7.png
25
106
media_image7.png
Greyscale
before feeding it to the binary activation function described above. A batch normalization layer converts each input y into
α
x y –
β
, where
α
a
n
d
β
are the parameters of the layer. Authors of [17] suggest combining batch-normalization and binary activation layers into a single thresholding layer. The cascade of the two layers computes the following:
PNG
media_image8.png
55
407
media_image8.png
Greyscale
”, Figure 1 (bottom):
PNG
media_image9.png
52
331
media_image9.png
Greyscale
) [Examiner’s note: The text and Figure 1 (bottom) show how the linear dot product (homogenization processing) is replaced by XNOR + Popcount, which is a nonlinear bitwise operation that maps to dot product results using a different computational mechanism. Additionally, the binary activation function replaces analog thresholding with a sign-based threshold – a nonlinear comparator-based transformation]
Regarding to Claim 5, the combination of Ghasemzadeh, Fan, Goto and Mohassel discloses all the limitations of Claim 1 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel further discloses:
wherein in the generating: the prediction model is encrypted by distributing, through a secret sharing method, the prediction model that has been converted, (Mohassel, [0215]: “Embodiments can hide the data, the model, the prediction, or any combinations of them, as they can all be secret shared.”)
and in the distributing of the prediction model, the at least one parameter for performing the processing including the nonlinear processing is distributed. (Mohassel, [0215]: “Embodiments can hide the data, the model, the prediction, or any combinations of them, as they can all be secret shared.”) [The examiner interprets that “any combinations of them” here is any combinations of the model, which will include the parameters or weights]
Regarding Claim 7, the combination of Ghasemzadeh, Fan, Goto and Mohassel discloses all the limitations of Claim 1 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel further discloses:
wherein the prediction model is a binarized neural network including a plurality of parameters each comprising a binary value of -1 or 1. (Ghasemzadeh, Page 2, Col. 1, Section II.A: “In case of binary CNNs, the elements of
PNG
media_image11.png
26
14
media_image11.png
Greyscale
and
PNG
media_image12.png
24
20
media_image12.png
Greyscale
are restricted to binary values
PNG
media_image4.png
26
37
media_image4.png
Greyscale
and
PNG
media_image5.png
24
38
media_image5.png
Greyscale
, respectively. The dot product of these vectors can be efficiently computed using XnorPopcount operations as suggested in [15], [16]. Let
PNG
media_image13.png
24
308
media_image13.png
Greyscale
are scalar values and
PNG
media_image14.png
22
59
media_image14.png
Greyscale
are sign vectors whose elements are either +1 or -1. If we encode the sign values (-1
→
0 and +1
→
1), we obtain binary vectors”)
Regarding to Claim 8, the combination of Ghasemzadeh, Fan, Goto and Mohassel discloses all the limitations of Claim 1 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel further discloses:
further comprising: training the prediction model using training data collected in advance, (Mohassel, [0004]: “After training data is obtained, a learning process can be used to train the model. Learning module 120 is shown receiving existing records 110 and providing model 130 after training has been performed”)
wherein a parameter obtained through the training as the at least one parameter for performing the homogenization processing is converted in the converting. ((Ghasemzadeh, Page 2, Col. 1, Section II. A, In the Binary dot product section, real-valued vectors with values like
PNG
media_image4.png
26
37
media_image4.png
Greyscale
and
PNG
media_image5.png
24
38
media_image5.png
Greyscale
are replaced by sign vectors (+- 1), and then encoded into binary vectors (e.g., -1
→
0 and +1
→
1). This is a direct conversion of negative values to positive binary representations, where a negative sign (-1) becomes 0. Figure 1 (top) in page 2 shows the original dot product which represents a homogenization processing:
PNG
media_image6.png
80
363
media_image6.png
Greyscale
because homogenization typically refers to linear or smoothing operations – like standard dot products, or typical batch normalization. Page 2, Col. 2, Section Binary batch-normalization: “It is often useful to normalize the result of the dot product
PNG
media_image7.png
25
106
media_image7.png
Greyscale
before feeding it to the binary activation function described above. A batch normalization layer converts each input y into
α
x y –
β
, where
α
a
n
d
β
are the parameters of the layer. Authors of [17] suggest combining batch-normalization and binary activation layers into a single thresholding layer. The cascade of the two layers computes the following:
PNG
media_image8.png
55
407
media_image8.png
Greyscale
”, Figure 1 (bottom):
PNG
media_image9.png
52
331
media_image9.png
Greyscale
) [Examiner’s note: The text and Figure 1 (bottom) show how the linear dot product (homogenization processing) is replaced by XNOR + Popcount, which is a nonlinear bitwise operation that maps to dot product results using a different computational mechanism. Additionally, the binary activation function replaces analog thresholding with a sign-based threshold – a nonlinear comparator-based transformation]
Regarding Claim 10, the combination of Ghasemzadeh, Fan, Goto and Mohassel discloses all the limitations of Claim 1 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel further discloses:
wherein in the converting, [[a]] the at least one negative numerical value is converted to [[a]] the positive numerical value by converting a numerical value in a plurality of parameters included in the prediction model to a set including a sign part indicating a sign of the numerical value as 0 or 1 and a numerical value part indicating an absolute value of the numerical value. (Goto, [0113]: “Specifically, as shown in FIG. 19, the binarization component 222 binarizes data values of attribute Y. FIG. 19 shows an example of the prediction data that has been binarized by the prediction application in the exemplary embodiment of the present invention. In the example of FIG. 19, processing for changing data values of attribute Y that are smaller than 50 to 0 (bin_ Y =0) and changing data values of attribute Y that are equal to or larger than 50 to 1 (bin_ Y = 1) is executed as binarization processing, similarly to the example of FIG. 10.”) [Plurality of parameters: data values of attribute Y that are smaller than 50; converted into one parameter: converted to 0]
Regarding Claim 13, Ghasemzadeh explicitly discloses:
converting, by the data providing device, a prediction model by converting at least one parameter which is included in the prediction model and is performing homogenization processing into at least one parameter for performing processing including nonlinear processing, (Ghasemzadeh, Page 2, Col. 1, Section II. A, In the Binary dot product section, real-valued vectors with values like
PNG
media_image4.png
26
37
media_image4.png
Greyscale
and
PNG
media_image5.png
24
38
media_image5.png
Greyscale
are replaced by sign vectors (+- 1), and then encoded into binary vectors (e.g., -1
→
0 and +1
→
1). This is a direct conversion of negative values to positive binary representations, where a negative sign (-1) becomes 0. Figure 1 (top) in page 2 shows the original dot product which represents a homogenization processing:
PNG
media_image6.png
80
363
media_image6.png
Greyscale
because homogenization typically refers to linear or smoothing operations – like standard dot products, or typical batch normalization. Page 2, Col. 2, Section Binary batch-normalization: “It is often useful to normalize the result of the dot product
PNG
media_image7.png
25
106
media_image7.png
Greyscale
before feeding it to the binary activation function described above. A batch normalization layer converts each input y into
α
x y –
β
, where
α
a
n
d
β
are the parameters of the layer. Authors of [17] suggest combining batch-normalization and binary activation layers into a single thresholding layer. The cascade of the two layers computes the following:
PNG
media_image8.png
55
407
media_image8.png
Greyscale
”, Figure 1 (bottom):
PNG
media_image9.png
52
331
media_image9.png
Greyscale
) [Examiner’s note: The text and Figure 1 (bottom) show how the linear dot product (homogenization processing) is replaced by XNOR + Popcount, which is a nonlinear bitwise operation that maps to dot product results using a different computational mechanism. Additionally, the binary activation function replaces analog thresholding with a sign-based threshold – a nonlinear comparator-based transformation]
the prediction model being a neural network; (Ghasemzadeh, pg. 2, col. 1, section II.A: “Neural networks are composed of multiple convolution, fully-connected, activation, batch-normalization, and max-pooling layers. Binarization enables the use of a simpler equivalent for each layer as explained in this section.”)
Ghasemzadeh fails to disclose:
A prediction model conversion system, comprising: a processor; and
a memory including a computer program, the computer program, when executed by the processor, causing the processor to perform functions, the function including:
receiving, as the input, user information, the user information being in the secret state;
performing the prediction processing using the encrypted prediction model and the user
information in the secret state; and
presenting, as at least one of an image or audio, prediction results to a user based on the prediction processing, the user corresponding to the user information,
generating, by the data providing device, an encrypted prediction model that performs prediction processing with input in a secret state remaining secret by encrypting the prediction model that has been converted.
wherein, in the generating: the prediction model is encrypted by distributing, through a secret sharing method, the prediction model that has been converted, to generate at least two distributed encrypted prediction models for keeping the prediction model secret and safely performing the prediction processing,
in the prediction processing, the at least two distributed encrypted prediction models are applied to at least two distributed data by the secret sharing method, and at least two secret prediction results are output in the secret state, and
the at least two secret prediction results, in the secret state, are necessary to obtain a decrypted prediction result.
and integerization processing is avoided for reducing an amount of computation of the
prediction processing, improving accuracy of the prediction processing, and reducing a drop in prediction accuracy, and
a divisor used in the secret sharing method is added to the negative numerical value to
convert the negative numerical value to the positive numerical value.
However, Goto explicitly discloses:
A prediction model conversion system, comprising: a processor; and (Goto, ¶[0170]: “A computer-readable recording medium having recorded therein a program for, using a computer, providing learning data to a system that generates a prediction model by performing machine learning, the program including an instruction that causes the computer to execute:”)
a memory including a computer program, the computer program, when executed by the processor, causing the processor to perform functions, the function including: (Goto, ¶[0170]: “A computer-readable recording medium having recorded therein a program for, using a computer, providing learning data to a system that generates a prediction model by performing machine learning, the program including an instruction that causes the computer to execute:”)
generating an encrypted prediction model that performs prediction processing with input in a secret state remaining secret by encrypting the prediction model that has been converted. (Goto, [0050]: “The encryption unit 20 encrypts the learning data so that a prediction model generated from the learning data in an unencrypted state and a prediction model generated from the learning data in an encrypted state have a corresponding relationship with each other in terms of parameters, numeric values, and operators. The data output unit 30 outputs the encrypted learning data to the cloud system 200.”, and [0051]: “Therefore, even when the learning data is encrypted, the cloud system 200 according to the present exemplary embodiment generates a prediction model that is similar to a prediction model generated when the learning data is not encrypted. Thus, the cloud system 200 according to the present exemplary embodiment can perform machine learning without executing decryption processing, even when data used in machine learning is encrypted.”) [The examiner interprets the process of “performing prediction without executing decryption processing” as “performs prediction processing with input in a secret state remaining secret”]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh and Goto. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Goto teaches generating prediction model in an encrypted state. One of ordinary skill would have motivation to combine Ghasemzadeh and Goto to protect user privacy and secure inference without trust. In many use cases (e.g., healthcare, finance, or personal data), the input data is sensitive; sending raw input data to a server for prediction exposes it to the model provider, so the input data should be encrypted to ensure that the server never sees the original data.
However, Mohassel explicitly discloses:
receiving, as the input, user information, the user information being in the secret state; (Mohassel, ¶[0015]: “According to an embodiment, the private input data can be represented as integers (e.g., by shifting bits of floating-point numbers), and the training can involve multiplying these integers (and other intermediate values) and integer-represented weights”)
performing the prediction processing using the encrypted prediction model and the user
information in the secret state; and (Mohassel, ¶[0054]: “This competition phase can include multiplication of input the data by weights to obtain a predicted output.”, ¶[0056]: “Thus, if the machine learning used profile data of a user to predict actions by the user, each data item in the profile can be split among the two servers.”, ¶[0057]: “The sharing can be done in a secure manner.”)
presenting, as at least one of an image or audio, prediction results to a user based on the prediction processing, the user corresponding to the user information, (Mohassel, ¶[0003]: “For an example data sample, the input data can be the pixel values of an image, and the output data can be a classification of what is in the image (e.g., that the image is of a dog).”
wherein, in the generating: the prediction model is encrypted by distributing, through a secret sharing method, the prediction model that has been converted, to generate at least two distributed encrypted prediction models, for keeping the prediction model secret and safely performing the prediction processing (Mohassel, ¶[0140]: “In some embodiments, the training data is secret shared between two servers S0 and S1 . We denote the shares of two values X and Y at the two servers by <X>0, <Y>0 and <X>1 , <Y>1 . In practice, the clients can distribute the shares between the two servers, or encrypt the first share using the public key of S0 and upload both the first encrypted share and the second plaintext share to S1 . S1 can then pass the encrypted shares to S0 to decrypt. Herein, both implementations are encompassed by secret sharing”, ¶0011]: “Using fully homomorphic encryption, the neural network model can make predictions on encrypted data. In this case, it is assumed that the neural network is trained on plaintext data and the model is known to one party who evaluates it on private data of another.”) [Examiner’s note: Mohassel discloses the neural network model is trained to make predictions using encrypted data, while the training data is secret shared to 2 different servers. This aligns with the concept of the prediction model is encrypted using secret sharing method then generate 2 separate encrypted prediction models]
in the prediction processing, the at least two distributed encrypted prediction models are applied to at least two distributed data by the secret sharing method, and at least two secret prediction results are output in the secret state, and (Mohassel, ¶[0039]: “As examples, the secret sharing can involve splitting a data item up into shares that require a sufficient number ( e.g., all) of training computers to reconstruct and/or encryption mechanisms where decryption requires collusion among the training computers.”, ¶[0140]: “In some embodiments, the training data is secret shared between two servers S0 and S1 . We denote the shares of two values X and Y at the two servers by <X>0, <Y>0 and <X>1 , <Y>1 . In practice, the clients can distribute the shares between the two servers, or encrypt the first share using the public key of S0 and upload both the first encrypted share and the second plaintext share to S1 . S1 can then pass the encrypted shares to S0 to decrypt. Herein, both implementations are encompassed by secret sharing”, ¶0011]: “Using fully homomorphic encryption, the neural network model can make predictions on encrypted data.) [Examiner’s note: Mohassel discloses training the neural network model to make predictions on encrypted data, wherein the training data is distributed for 2 servers using the secret sharing method which will end up with 2 encrypted distributed shares]
the at least two secret prediction results, in the secret state, are necessary to obtain a decrypted prediction result. (Mohassel, ¶0108]: “In some implementations, the sender prepares the garbled circuit by determining a truth table for each gate using the random numbers that replaced the two bits on the input wires. The output values are then encrypted ( e.g., using double-key symmetric encryption) with the random numbers from the truth table. Thus, one can only decrypt the gate only if one knows the two correct random numbers for a given output value.”) [Examiner’s note: “two secret prediction results” is being interpreted as the two random numbers needed to decrypt the gate]
and integerization processing is avoided for reducing an amount of computation of the
prediction processing, improving accuracy of the prediction processing, and reducing a drop in prediction accuracy, and (Mohassel, ¶[0015]: “A secret-shared result ( e.g., the delta value for updating the weights) can be truncated by truncating the secret-shared parts at the training computers, thereby allowing efficient computation and limiting the amount of memory for storing the integer values.”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh and Mohassel. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Mohassel teaches techniques for efficient implementations that allow multiple client computers to use their provided data in creating a machine learning model, without having to expose the private data. One of ordinary skill would have motivation to combine Ghasemzadeh and Mohassel to ensure that no single party can fully reconstruct the original data or the models without collaboration from other parties. This is critical form preserving privacy in domains such as healthcare, finance, or other sensitive fields where data confidentiality is paramount. (Mohassel, ¶[0215])
However, Fan explicitly discloses:
a divisor used in the secret sharing method is added to the negative numerical value to
convert the negative numerical value to the positive numerical value. (Fan, Col. 15, Lines 44-47: “The first block to be processed is selected in step 440. The data is converted from its coding scheme ( e.g. ASCII, Unicode, etc) to a numerical value (in decimal or binary). 45 In step 450, a set of divisors are selected, at random, from the memory 208. There are n such divisors in the set.”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh and Fan. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Fan teaches method and system for securely storing data using a secret sharing scheme. One of ordinary skill would have motivation to combine Ghasemzadeh and Fan 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.
Claim(s) 3 is rejected under 35 U.S.C. 103 as being unpatentable over Ghasemzadeh et al (“ReBNet: Residual Binarized Neural Network”) (hereafter referred to as “Ghasemzadeh”) in view of GOTO (US 2018/0089574 A1) (hereafter referred to as “Goto”), Mohassel et al. (US 2020/0242466 A1) (hereafter referred to as “Mohassel”), Fan et al. (US 11,245,522 B2) and further in view of Yamada et al. (US 2020/0279166 A1) (hereafter referred to as “Yamada”)
Regarding Claim 3, the combination of Ghasemzadeh, Goto, Fan and Mohassel discloses all the limitations of Claim 1 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel further discloses:
wherein the homogenization processing is processing performed by an equation yi = si xi + ti, where xi is an input and yi is an output, si and ti are the plurality of parameters for performing the homogenization processing, (Goto, [0102]: “Specifically, as shown in FIG. 15, the standardization attribute encryption unit 22 multiplies all samples of attribute X by a certain value (e.g., 10), and adds another certain value (e.g., 50) to values of the obtained products”) [The examiner interprets that the “homogenization processing” is the “standardization processing”, “attribute X” is “input xi”, “a certain value” that is multiplied with attribute X is parameter “si”, “another certain value” which is added is parameter “ti” and the “values of obtained products” is interpreted as “output yi” ]
ki is the at least one parameter for performing the processing including the nonlinear processing, and is determined using si and ti. (Goto, [0102]: “Specifically, as shown in FIG. 15, the standardization attribute encryption unit 22 multiplies all samples of attribute X by a certain value (e.g., 10), and adds another certain value (e.g., 50) to values of the obtained products, similarly to the example of step S304 shown in FIG. 3. FIG. 15 shows an example of the prediction data in which the specific attribute has been standardized in the exemplary embodiment of the present invention.”, and [0103]: “In step S403, the standardization attribute encryption unit 22 also transfers the prediction data in which the attribute targeted for standardization has been encrypted ( see FIG. 15) to the binarization attribute encryption unit 23.”) [The highlight indicates that attribute X obtained by the standardization process is transferred to the binarization process as the input, while X was calculated using si and ti (as shown above). The examiner interprets that the “homogenization processing” is the “standardization processing”, “attribute X” is “input xi”, “a certain value” that is multiplied with attribute X is parameter “si”, “another certain value” which is added is parameter “ti” and the “values of obtained products” is interpreted as “output yi” ]
Ghasemzadeh in view of Goto, Fan and Mohassel fails to disclose:
the processing including the nonlinear processing is processing performed by Equation
(1), and [Math 1]
PNG
media_image2.png
51
236
media_image2.png
Greyscale
However, Yamada explicitly discloses:
the processing including the nonlinear processing is processing performed by Equation
(1), and [Math 1]
PNG
media_image2.png
51
236
media_image2.png
Greyscale
(Yamada, [0094]: “An
activating function of the intermediate unit 213 is binarized to { -1, 1}, and an output Yj of the intermediate unit 213-j takes any one of the binary value of {-1, 1}.”, [0098]: “The input unit 211-i performs an operation f based on Wji (Xi, Wji) to the input Xi and outputs the result thereof to the intermediate unit 213-j. This operation is an operation for equalizing the positive/negative sign of Xi to the positive/ negative sign of Wji; and if Wji is 1, f(Xi, Wji)=Xi, and if Wji is -1, f(Xi, Wji)=-Xi.”) [The examiner interprets “ki” of Equation (1) as value of Xi in this context]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh, Goto, Mohassel, Fan and Yamada. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Goto teaches generating prediction model in an encrypted state. Mohassel teaches techniques for efficient implementations that allow multiple client computers to use their provided data in creating a machine learning model, without having to expose the private data. Fan teaches method and system for securely storing data using a secret sharing scheme. Yamada teaches performing binary calculation with the binarized neural network. One of ordinary skill would have motivation to combine Ghasemzadeh, Goto, Mohassel, Fan and Yamada to reduce calculation cost and hardware resources by applying binarized neural network (Yamada, [0175])
Claim(s) 6, 9 are rejected under 35 U.S.C. 103 as being unpatentable over Ghasemzadeh et al (“ReBNet: Residual Binarized Neural Network”) (hereafter referred to as “Ghasemzadeh”) in view of GOTO (US 2018/0089574 A1) (hereafter referred to as “Goto”), Mohassel et al. (US 2020/0242466 A1) (hereafter referred to as “Mohassel”), Fan et al (US 11,245,522 B2) and in further view of Matsuo (US 9,331,984 B2) (hereafter referred to as “Matsuo”)
Regarding to Claim 6, the combination of Ghasemzadeh, Fan, Goto and Mohassel discloses all the limitations of Claim 5 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel fails to disclose:
determining a divisor used in the secret sharing method in a range greater than an element of the prediction model.
However, Matsuo explicitly discloses:
determining a divisor used in the secret sharing method in a range greater than an element of the prediction model. (Matsuo, Col. 2, Lines 22-27: “The present invention eliminates such problems of the prior art by providing a secret sharing method in which secret data is shared as shared data parts equal to or greater than a threshold value in number such that the secret data cannot be reconstructed from shared data parts less than the threshold value in number,”) [The examiner interprets “a divisor” as the dividing the secret data into parts or pieces that contribute to the secret sharing, and “a possible value of an element” is interpreted as “a threshold value” here]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh, Goto, Mohassel , Fan and Matsuo. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Goto teaches generating prediction model in an encrypted state. Mohassel teaches techniques for efficient implementations that allow multiple client computers to use their provided data in creating a machine learning model, without having to expose the private data. Matsuo teaches a secret sharing method in which secret data is shared as shared data parts equal to or greater than a threshold value in number such that the secret data cannot be reconstructed from shared data parts less than the threshold value in number. Fan teaches method and system for securely storing data using a secret sharing scheme. One of ordinary skill would have motivation to combine Ghasemzadeh, Goto, Fan, Mohassel and Matsuo to preserve the privacy of the secret data because when less than k pairs of shared data parts are obtained, it is not possible to reconstruct the secret data (Matsuo, Col. 9, Lines 40-48)
Regarding to Claim 9, the combination of Ghasemzadeh, Fan, Goto and Mohassel discloses all the limitations of Claim 6 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel fails to disclose:
wherein in the converting, the divisor used in the secret sharing method is added to a negative numerical value in a plurality of parameters included in the prediction model to convert the negative numerical value to a positive numerical value.
However, Matsuo explicitly discloses:
wherein in the converting, the divisor used in the secret sharing method is added to a negative numerical value in a plurality of parameters included in the prediction model to convert the negative numerical value to a positive numerical value. (Matsuo, Col. 17, Lines 59-62: “the shared data piece B2 is created as a residue obtained by dividing the product of the random number data piece R2 and the secret data piece S 1 with a prime number P.”, Col. 18, Lines 61-62: “If the result of the subtraction is negative, a prime number is added thereto to make it positive.”) [The examiner interprets the “divisor” as the “prime number” in this context]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh, Goto, Mohassel, Fan and Matsuo. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Goto teaches generating prediction model in an encrypted state. Mohassel teaches techniques for efficient implementations that allow multiple client computers to use their provided data in creating a machine learning model, without having to expose the private data. Matsuo teaches a secret sharing method in which secret data is shared as shared data parts equal to or greater than a threshold value in number such that the secret data cannot be reconstructed from shared data parts less than the threshold value in number. Fan teaches method and system for securely storing data using a secret sharing scheme. One of ordinary skill would have motivation to combine Ghasemzadeh, Goto, Mohassel, Fan and Matsuo to preserve the privacy of the secret data because when less than k pairs of shared data parts are obtained, it is not possible to reconstruct the secret data (Matsuo, Col. 9, Lines 40-48)
Claim(s) 11 is rejected under 35 U.S.C. 103 as being unpatentable over Ghasemzadeh et al (“ReBNet: Residual Binarized Neural Network”) (hereafter referred to as “Ghasemzadeh”) in view of GOTO (US 2018/0089574 A1) (hereafter referred to as “Goto”), Mohassel et al. (US 2020/0242466 A1) (hereafter referred to as “Mohassel”), Fan et al (US 11,245,522 B2), and in further view of Unagami et al. (US 10,649,919 B2) (hereafter referred to as “Unagami”)
Regarding to Claim 11, the combination of Ghasemzadeh, Goto, Fan and Mohassel discloses all the limitations of Claim 5 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Fan and Mohassel fails to disclose:
wherein, in the prediction processing, a feature amount is calculated from data obtained by sensing; and
the feature amount that has been caculated is distributed, through the secret sharing method.
However, Unagami explicitly discloses:
wherein, in the prediction processing, a feature amount is calculated from data obtained by sensing; and (Unagami, Col. 9, Lines 41-45: “The feature value calculation unit 102 calculates a feature value from the privacy data, which is information acquired at the sensing unit 101.”)
the feature amount that has been caculated is distributed, through the secret sharing method. (Unagami, Col. 10, Lines 31-36: “The feature value encryption unit 104 encrypts the feature value calculated by the feature value calculation unit 102 by using a predetermined cryptosystem. In this encryption, the feature value encryption unit 104 uses a key stored in the key storage unit 103.”)
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh, Goto, Fan, Mohassel and Unagami. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Goto teaches generating prediction model in an encrypted state. Mohassel teaches techniques for efficient implementations that allow multiple client computers to use their provided data in creating a machine learning model, without having to expose the private data. Unagami teaches encrypting information using homomorphic encryption. Fan teaches method and system for securely storing data using a secret sharing scheme. One of ordinary skill would have motivation to combine Ghasemzadeh, Goto, Mohassel, Fan and Unagami to obtain relevant data similar to the feature extracted by sensing and make use of that reference data in training processing (Unagami, Col. 8, Lines 61-67)
Claim(s) 12 is rejected under 35 U.S.C. 103 as being unpatentable over Ghasemzadeh et al (“ReBNet: Residual Binarized Neural Network”) (hereafter referred to as “Ghasemzadeh”) in view of GOTO (US 2018/0089574 A1) (hereafter referred to as “Goto”), Mohassel et al. (US 2020/0242466 A1) (hereafter referred to as “Mohassel”) , Unagami et al. (US 10,649,919 B2) (hereafter referred to as “Unagami”), Fan et al (US 11,245,522 B2) and further in view of Yamada et al. (US 2020/0279166 A1) (hereafter referred to as “Yamada”)
Regarding to Claim 12, the combination of Ghasemzadeh, Goto, Fan, Mohassel and Unagami discloses all the limitations of Claim 11 (as shown in the rejection above).
Ghasemzadeh in view of Goto, Mohassel, Fan and Unagami further discloses:
wherein, in the prediction processing, the prediction processing is executed by the prediction model that has been distributed, by inputting, to the prediction model that has been distributed, the feature amount that has been distributed, (Goto, [0045]: “In the present invention, the analysis application of the cloud service generates a prediction model by applying preprocessing and analysis processing to encrypted input data.”, [0126]: “In the case of text data analysis processing in which text data is used as input data and the frequency of appearance of each character or word is analyzed as a feature amount,”) [Prediction model is generated by preprocessing the input data which is also defined as the input feature amount]
the prediction processing includes the nonlinear processing, (Goto, [0057]: “The analysis engine 223 predicts data by applying the prediction data that has been standardized and binarized to the prediction model.”) [The examiner interprets the “nonlinear processing” as the binarization process in this context]
Ghasemzadeh in view of Goto, Mohassel, Fan and Unagami fails to disclose:
the nonlinear processing is processing of converting an input to the nonlinear processing into 1 when the input is 0 or a numerical value corresponding to a positive, and into a positive numerical value corresponding to -1 when the input is a numerical value corresponding to a negative.
However, Yamada explicitly discloses:
the nonlinear processing is processing of converting an input to the nonlinear processing into 1 when the input is 0 or a numerical value corresponding to a positive, and into a positive numerical value corresponding to -1 when the input is a numerical value corresponding to a negative. (Yamada, [0094]: “An activating function of the intermediate unit 213 is binarized to { -1, 1}, and an output Yj of the intermediate unit 213-j takes any one of the binary value of {-1, 1}.”, [0098]: “The input unit 211-i performs an operation f based on Wji (Xi, Wji) to the input Xi and outputs the result thereof to the intermediate unit 213-j. This operation is an operation for equalizing the positive/negative sign of Xi to the positive/ negative sign of Wji; and if Wji is 1, f(Xi, Wji)=Xi, and if Wji is -1, f(Xi, Wji)=-Xi.”) [The inputs being binarized to (-1) or (1) in response to whether the data value is positive or negative]
It would have obvious to one of ordinary skill in the art before the effective filing date of the present application to combine Ghasemzadeh, Goto, Mohassel, Unagami and Yamada. Ghasemzadeh teaches an end-to-end framework for training reconfigurable binary neural networks on software. Fan teaches method and system for securely storing data using a secret sharing scheme .Goto teaches generating prediction model in an encrypted state. Mohassel teaches techniques for efficient implementations that allow multiple client computers to use their provided data in creating a machine learning model, without having to expose the private data. Unagami teaches encrypting information using homomorphic encryption. Yamada teaches performing binary calculation with the binarized neural network. One of ordinary skill would have motivation to combine Ghasemzadeh, Goto, Mohassel, Unagami, Fan and Yamada to reduce calculation cost and hardware resources by applying binarized neural network (Yamada, [0175])
Allowable Subject Matter
Claim 4 would be allowable if rewritten or amended to overcome the rejection(s) under 35 U.S.C 101 set forth in this office action.
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 on (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
/DAVID YI/Supervisory Patent Examiner, Art Unit 2126
Prosecution Insights