DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Arguments
Applicant's arguments filed 12/10/2025 have been fully considered but they are not fully persuasive.
Regarding the 101 rejections,
On pages 6-7 of “Remarks” applicant contends that the amended claim 1 provides a practical application under Step 2A Prong 2. The examiner respectfully disagrees. As noted in the previous office action, under the broadest reasonable interpretation, performing a Hermitic expansion is interpreted as a mathematical calculation which is a judicial exception (MPEP 2106). Applicant asserts that the use of the Hermitic polynomial expansion for activation functions in neural networks provides a technical improvement to the field of neural networks by reducing execution time, reducing network communication, and providing higher classification accuracy. However, the claims do not reflect the improvement and instead merely recite using a generic convolutional neural network as a tool to perform a judicial exception of a Hermitic expansion which, under the broadest reasonable interpretation, merely recite steps that apply using a generic convolutional neural network as a tool to perform a judicial exception, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Additionally, it appears that the proposed improvement is only realized because of the specific mathematical concepts (Hermitic polynomial expansion) used in the claim. The judicial exception itself cannot provide the improvement. See MPEP 2106.05(a): “It is important to note, the judicial exception alone cannot provide the improvement. The improvement can be provided by one or more additional elements.”
In response to applicant’s use of the August 2025 memorandum on pages 7-8 of “Remarks”, the examiner notes that the memorandum provides reminders of existing policy, which reminds the examiner to carefully distinguish between claim limitations that merely involve a judicial exception versus those that recite an exception. It states that it is not intended to announce any new subject matter evaluation. This analysis is consistent with the USPTO’s August 2025 memorandum on 35 USC § 101.
On pages 8-9 of “Remarks” applicant cites supporting federal court decisions as support for eligibility. The examiner notes that examiners determine eligibility based on the guidance in the MPEP. Therefore, applicant’s arguments regarding the 101 rejections are not persuasive.
Regarding the 103 rejections, applicant's arguments filed with respect to the prior art Park are persuasive and therefore the rejections under Park are withdrawn. Applicant has amended the claims to recite new combinations of limitations and a new combination of references is applied below. Please see below for new grounds of rejection, necessitated by Amendment.
Claim Interpretation
The following is a quotation of 35 U.S.C. 112(f):
(f) Element in Claim for a Combination. – An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The following is a quotation of pre-AIA 35 U.S.C. 112, sixth paragraph:
An element in a claim for a combination may be expressed as a means or step for performing a specified function without the recital of structure, material, or acts in support thereof, and such claim shall be construed to cover the corresponding structure, material, or acts described in the specification and equivalents thereof.
The claims in this application are given their broadest reasonable interpretation using the plain meaning of the claim language in light of the specification as it would be understood by one of ordinary skill in the art. The broadest reasonable interpretation of a claim element (also commonly referred to as a claim limitation) is limited by the description in the specification when 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is invoked.
As explained in MPEP § 2181, subsection I, claim limitations that meet the following three-prong test will be interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph:
(A) the claim limitation uses the term “means” or “step” or a term used as a substitute for “means” that is a generic placeholder (also called a nonce term or a non-structural term having no specific structural meaning) for performing the claimed function;
(B) the term “means” or “step” or the generic placeholder is modified by functional language, typically, but not always linked by the transition word “for” (e.g., “means for”) or another linking word or phrase, such as “configured to” or “so that”; and
(C) the term “means” or “step” or the generic placeholder is not modified by sufficient structure, material, or acts for performing the claimed function.
Use of the word “means” (or “step”) in a claim with functional language creates a rebuttable presumption that the claim limitation is to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites sufficient structure, material, or acts to entirely perform the recited function.
Absence of the word “means” (or “step”) in a claim creates a rebuttable presumption that the claim limitation is not to be treated in accordance with 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph. The presumption that the claim limitation is not interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, is rebutted when the claim limitation recites function without reciting sufficient structure, material or acts to entirely perform the recited function.
Claim limitations in this application that use the word “means” (or “step”) are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action. Conversely, claim limitations in this application that do not use the word “means” (or “step”) are not being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, except as otherwise indicated in an Office action.
This application includes one or more claim limitations that do not use the word “means,” but are nonetheless being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, because the claim limitation(s) uses a generic placeholder that is coupled with functional language without reciting sufficient structure to perform the recited function and the generic placeholder is not preceded by a structural modifier. Such claim limitation(s) is/are:
a Hermite block configured to apply… in claim 1.
Because this/these claim limitation(s) is/are being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, it/they is/are being interpreted to cover the corresponding structure described in the specification as performing the claimed function, and equivalents thereof.
If applicant does not intend to have this/these limitation(s) interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph, applicant may: (1) amend the claim limitation(s) to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph (e.g., by reciting sufficient structure to perform the claimed function); or (2) present a sufficient showing that the claim limitation(s) recite(s) sufficient structure to perform the claimed function so as to avoid it/them being interpreted under 35 U.S.C. 112(f) or pre-AIA 35 U.S.C. 112, sixth paragraph.
Claim Rejections - 35 USC § 112: Indefiniteness
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 1-10 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Regarding claim 1, the claim recites the limitation and providing an output of the convolutional neural network as an inference result for the input values. There is insufficient antecedent basis for this limitation in the claim because the term “the input values” lacks antecedent basis. For the purposes of examination, the term “the input values” are interpreted as being an input.
Regarding claims 2-10, they are rejected for at least their dependence on claim 1.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1, in step 1 of the 101 analysis set forth in MPEP 2106, the claim recites An operating method of a deep learning system configured to perform a private inferences, comprising:. The claim recites a method. A method is one of the four statutory categories of invention.
In Step 2A, Prong 1 of the 101 analysis set forth in MPEP 2106, the examiner has determined that the following limitations recite a process that, under broadest reasonable interpretation, covers a mental process or mathematical concept but for the recitation of generic computer components:
and a Hermite block configured to apply an activation function based on a Hermitic polynomial expansion; (i.e., the broadest reasonable interpretation includes a mathematical calculation using hermitic polynomial expansion, a mathematical calculation is considered a mathematical concept (MPEP 2106)).
If the claim limitations, under their broadest reasonable interpretation, covers activities classified under Mental processes: concepts performed in the human mind (including observation, evaluation, judgement, or opinion) (see MPEP 2106.04(a)(2), subsection (III)) or Mathematical concepts: mathematical relationships, mathematical formulas or equations, or mathematical calculations (see MPEP 2106.04(a)(2), subsection (I)). Accordingly, the claim recites an abstract idea.
In Step 2A, Prong 2 of the 101 analysis, set forth in MPEP 2106, the examiner has determined that the following additional elements do not integrate this judicial exception into a practical application:
obtaining encrypted input values; (i.e., the broadest reasonable interpretation of receiving a data instance is mere data gathering, which is an insignificant extra solution activity (MPEP 2106.05(g))).
executing a convolutional neural network with respect to the encrypted input values, (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
wherein the convolutional neural network comprises: a convolution layer configured to perform a convolution operation, (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
and providing an output of the convolutional neural network as an inference result for the input values (i.e., the broadest reasonable interpretation of outputting a value is mere data outputting, which is an insignificant extra solution activity (MPEP 2106.05(g))).
Since the claim does not contain any other additional elements, that amount to integration into a practical application, the claim is directed to an abstract idea.
In Step 2B of the 101 analysis set forth in the 2019 PEG, the examiner has determined that the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception:
Regarding limitations (II) and (V), under the broadest reasonable interpretation, recite steps of mere data gathering/outputting, which has been recognized by the courts as being well-understood, routine, and conventional functions. Specifically, the courts have recognized computer functions directed to mere data gathering/outputting as well-understood, routine, and conventional functions when they are claimed in a merely generic manner or as insignificant extra-solution activity when considering evidence in view of Berkheimer v. HP, Inc., 881 F.3d 1360, 1368, 125 USPQ2d 1649, 1654 (Fed. Cir. 2018), see USPTO Berkheimer Memorandum (April 2018)).
Examiner uses Berkheimer: Option 2, a citation to one or more of the court decisions discussed in MPEP 2106.05(d)(II) as noting well-understood, routine, and conventional nature of the additional elements:
Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016) (using a telephone for image transmission); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network); buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network). See MPEP 2106.05(d)(II).
Further, limitations (III) and (IV), under the broadest reasonable interpretation, merely recite steps that apply using a generic convolutional neural network as a tool to perform a judicial exception, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Considering additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Regarding claim 2, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 2 recites wherein the encrypted input values are values encrypted based on a homomorphic encryption scheme. Under the broadest reasonable interpretation, the limitations recite encrypting inputs using homomorphic encryption which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 2 does not solve the deficiencies of claim 1.
Regarding claim 3, it is dependent upon claim 2 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 3 recites wherein the homomorphic encryption scheme is at least one of BGV (Brakerski-Gentry-Vaikuntanathan), BFV (Brakerski-Fan-Vercauteren), or CKKS (Cheon-Kim-Kim-Song). Under the broadest reasonable interpretation, the limitations recite encrypting inputs using homomorphic encryption which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 3 does not solve the deficiencies of claim 2.
Regarding claim 4, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 4 recites wherein the Hermitic polynomial expansion is a Hermitic expansion of a rectified linear unit (ReLU) activation function. Under the broadest reasonable interpretation, the limitations recite performing Hermitic polynomial expansion on a function which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 4 does not solve the deficiencies of claim 1.
Regarding claim 5, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 5 recites wherein the Hermite polynomial expansion comprises a second-order Hermite polynomial. Under the broadest reasonable interpretation, the limitations recite performing Hermitic polynomial expansion with a second order polynomial which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 5 does not solve the deficiencies of claim 1.
Regarding claim 6, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 6 recites wherein the activation function is based on a batch-normalized Hermite polynomial. Under the broadest reasonable interpretation, the limitations recite performing batch normalization which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 6 does not solve the deficiencies of claim 1.
Regarding claim 7, it is dependent upon claim 6 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 7 recites wherein the activation function comprises a sum of terms of the batch-normalized Hermite polynomial multiplied by respective coefficients. Under the broadest reasonable interpretation, the limitations recite performing mathematical calculations with the Hermite polynomial which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 7 does not solve the deficiencies of claim 6.
Regarding claim 8, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 8 recites wherein the Hermitic polynomial expansion comprises a Hermitic expansion of a non-linear activation function. Under the broadest reasonable interpretation, the limitations recite performing Hermitic expansion on a function which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 8 does not solve the deficiencies of claim 1.
Regarding claim 9, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 9 recites wherein the deep learning system uses at least of one of a visual geometry group (VGG), residual neural network (ResNet), or Preactivation ResNet. Under the broadest reasonable interpretation, the limitations merely recite steps that apply a generic type of deep learning model, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Therefore, claim 9 does not solve the deficiencies of claim 1.
Regarding claim 10, it is dependent upon claim 1 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 10 recites wherein the activation function is limited to addition operations and multiplication operations. Under the broadest reasonable interpretation, the limitations recite performing multiplication and addition which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 10 does not solve the deficiencies of claim 1.
Regarding claim 11, in step 1 of the 101 analysis set forth in MPEP 2106, the claim recites A deep learning system including a neural network trained to perform private inferences, comprising: a client device. The claim recites a system. A system comprising a device is interpreted as an machine and a machine is one of the four statutory categories of invention.
In Step 2A, Prong 1 of the 101 analysis set forth in MPEP 2106, the examiner has determined that the following limitations recite a process that, under broadest reasonable interpretation, covers a mental process or mathematical concept but for the recitation of generic computer components:
pre-calculate randomly generated data, to obtain pre-calculated data; (i.e., the broadest reasonable interpretation includes a step of evaluation and judgement and could be performed mentally like determining values of random numbers, which is either a mental process of evaluation/judgement (MPEP 2106)).
generate…operated values by performing a homomorphic encryption operation with respect to the received pre-calculated data, (i.e., the broadest reasonable interpretation includes a mathematical calculation using homomorphic encryption, a mathematical calculation is considered a mathematical concept (MPEP 2106)).
an activation function based on a Hermitic polynomial expansion, (i.e., the broadest reasonable interpretation a includes mathematical calculation using hermitic polynomial expansion, a mathematical calculation is considered a mathematical concept (MPEP 2106)).
If the claim limitations, under their broadest reasonable interpretation, covers activities classified under Mental processes: concepts performed in the human mind (including observation, evaluation, judgement, or opinion) (see MPEP 2106.04(a)(2), subsection (III)) or Mathematical concepts: mathematical relationships, mathematical formulas or equations, or mathematical calculations (see MPEP 2106.04(a)(2), subsection (I)). Accordingly, the claim recites an abstract idea.
In Step 2A, Prong 2 of the 101 analysis, set forth in MPEP 2106, the examiner has determined that the following additional elements do not integrate this judicial exception into a practical application:
a client device configured to (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
and a cloud server configured to: receive the pre-calculated data from the client device, (i.e., the broadest reasonable interpretation of receiving data is mere data gathering, which is an insignificant extra solution activity (MPEP 2106.05(g))).
using the neural network (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
wherein the neural network is configured to apply (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
and output the operated values. (i.e., the broadest reasonable interpretation of outputting a value is mere data outputting, which is an insignificant extra solution activity (MPEP 2106.05(g))).
Since the claim does not contain any other additional elements, that amount to integration into a practical application, the claim is directed to an abstract idea.
In Step 2B of the 101 analysis set forth in the 2019 PEG, the examiner has determined that the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception:
Regarding limitations (V) and (VIII), under the broadest reasonable interpretation, recite steps of mere data gathering/outputting, which has been recognized by the courts as being well-understood, routine, and conventional functions. Specifically, the courts have recognized computer functions directed to mere data gathering/outputting as well-understood, routine, and conventional functions when they are claimed in a merely generic manner or as insignificant extra-solution activity when considering evidence in view of Berkheimer v. HP, Inc., 881 F.3d 1360, 1368, 125 USPQ2d 1649, 1654 (Fed. Cir. 2018), see USPTO Berkheimer Memorandum (April 2018)).
Examiner uses Berkheimer: Option 2, a citation to one or more of the court decisions discussed in MPEP 2106.05(d)(II) as noting well-understood, routine, and conventional nature of the additional elements:
Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016) (using a telephone for image transmission); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network); buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network). See MPEP 2106.05(d)(II).
Further, limitation (IV), under the broadest reasonable interpretation, merely recite steps that apply generic computer components to perform a judicial exception which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Similarly, limitations (VI) and (VII), under the broadest reasonable interpretation, merely recite steps that apply a generic neural network as a tool to perform a judicial exception which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Considering additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Regarding claim 12, it is dependent upon claim 11 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 12 recites wherein performing the homomorphic encryption operation includes performing a convolution operation based on a homomorphic encryption scheme,. Under the broadest reasonable interpretation, the limitations recite performing a convolution operation which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Claim 12 also recites and wherein outputting the operated values includes transmitting the operated values to the client device. Under the broadest reasonable interpretation, the limitations recite steps of mere data outputting, which has been recognized by the courts as being well-understood, routine, and conventional functions. Specifically, the courts have recognized computer functions directed to mere data outputting as well-understood, routine, and conventional functions when they are claimed in a merely generic manner or as insignificant extra-solution activity (MPEP 2106.05(g)). Therefore, claim 12 does not solve the deficiencies of claim 11.
Regarding claim 13, it is dependent upon claim 12 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 13 recites wherein the client device is configured to obtain a result value using the activation function and the operated values. Under the broadest reasonable interpretation, the limitations recite steps of mere data outputting, which has been recognized by the courts as being well-understood, routine, and conventional functions. Specifically, the courts have recognized computer functions directed to mere data outputting as well-understood, routine, and conventional functions when they are claimed in a merely generic manner or as insignificant extra-solution activity (MPEP 2106.05(g)). Therefore, claim 13 does not solve the deficiencies of claim 12.
Regarding claim 14, it is dependent upon claim 13 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 14 recites wherein the client device is configured to obtain the result value by performing the homomorphic encryption operation on plaintext on which the pre-calculated data is based. Under the broadest reasonable interpretation, the limitations recite performing a homomorphic encryption operation on plaintext which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 14 does not solve the deficiencies of claim 13.
Regarding claim 15, it is dependent upon claim 11 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 15 recites wherein the cloud server is configured to perform an operation based on a multi-party computation technique using the activation function. Under the broadest reasonable interpretation, the limitations merely recite steps that apply a generic multi-party computation technique, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Therefore, claim 15 does not solve the deficiencies in claim 11.
Regarding claim 16, in step 1 of the 101 analysis set forth in MPEP 2106, the claim recites An operating method of a deep learning system, comprising:. The claim recites a method. A method is one of the four statutory categories of invention.
In Step 2A, Prong 1 of the 101 analysis set forth in MPEP 2106, the examiner has determined that the following limitations recite a process that, under broadest reasonable interpretation, covers a mental process or mathematical concept but for the recitation of generic computer components:
wherein the private inference includes a homomorphic encryption operation and a multi-party computation, (i.e., the broadest reasonable interpretation includes a mathematical calculation of using a homomorphic encryption operation, a mathematical calculation is considered a mathematical concept (MPEP 2106)).
and wherein the multi-party computation includes…an activation function based on a Hermitic polynomial expansion. (i.e., the broadest reasonable interpretation includes a mathematical calculation using hermitic polynomial expansion, a mathematical calculation is considered a mathematical concept (MPEP 2106)).
If the claim limitations, under their broadest reasonable interpretation, covers activities classified under Mental processes: concepts performed in the human mind (including observation, evaluation, judgement, or opinion) (see MPEP 2106.04(a)(2), subsection (III)) or Mathematical concepts: mathematical relationships, mathematical formulas or equations, or mathematical calculations (see MPEP 2106.04(a)(2), subsection (I)). Accordingly, the claim recites an abstract idea.
In Step 2A, Prong 2 of the 101 analysis, set forth in MPEP 2106, the examiner has determined that the following additional elements do not integrate this judicial exception into a practical application:
collecting data; (i.e., the broadest reasonable interpretation of collecting data is mere data gathering, which is an insignificant extra solution activity (MPEP 2106.05(g))).
training a prediction model based on the collected data; and performing a private inference using the prediction model, (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
executing a neural network configured to apply (i.e., the generic computer components recited in this limitation merely add the words “apply it”, or an equivalent, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea (MPEP 2106.05(f))).
Since the claim does not contain any other additional elements, that amount to integration into a practical application, the claim is directed to an abstract idea.
In Step 2B of the 101 analysis set forth in the 2019 PEG, the examiner has determined that the claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception:
Regarding limitation (III), under the broadest reasonable interpretation, recite steps of mere data gathering, which has been recognized by the courts as being well-understood, routine, and conventional functions. Specifically, the courts have recognized computer functions directed to mere data gathering as well-understood, routine, and conventional functions when they are claimed in a merely generic manner or as insignificant extra-solution activity when considering evidence in view of Berkheimer v. HP, Inc., 881 F.3d 1360, 1368, 125 USPQ2d 1649, 1654 (Fed. Cir. 2018), see USPTO Berkheimer Memorandum (April 2018)).
Examiner uses Berkheimer: Option 2, a citation to one or more of the court decisions discussed in MPEP 2106.05(d)(II) as noting well-understood, routine, and conventional nature of the additional elements:
Receiving or transmitting data over a network, e.g., using the Internet to gather data, Symantec, 838 F.3d at 1321, 120 USPQ2d at 1362 (utilizing an intermediary computer to forward information); TLI Communications LLC v. AV Auto. LLC, 823 F.3d 607, 610, 118 USPQ2d 1744, 1745 (Fed. Cir. 2016) (using a telephone for image transmission); OIP Techs., Inc., v. Amazon.com, Inc., 788 F.3d 1359, 1363, 115 USPQ2d 1090, 1093 (Fed. Cir. 2015) (sending messages over a network); buySAFE, Inc. v. Google, Inc., 765 F.3d 1350, 1355, 112 USPQ2d 1093, 1096 (Fed. Cir. 2014) (computer receives and sends information over a network). See MPEP 2106.05(d)(II).
Further, limitation (IV), under the broadest reasonable interpretation, merely recite steps that apply generic training and predicting of a machine learning model which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Similarly, limitation (V), under the broadest reasonable interpretation, merely recite steps that apply a generic neural network as a tool to perform a judicial exception which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Considering additional elements individually and in combination, and the claim as a whole, the additional elements do not provide significantly more than the abstract idea. Therefore, the claim is not patent eligible.
Regarding claim 17, it is dependent upon claim 16 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 17 recites wherein the collecting the data comprises generating a ciphertext using a homomorphic encryption scheme on plaintext data. Under the broadest reasonable interpretation, the limitations recite performing a homomorphic encryption operation which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 17 does not solve the deficiencies of claim 16.
Regarding claim 18, it is dependent upon claim 16 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 18 recites wherein training the prediction model comprises performing at least one of a feed-forward learning or a backpropagation learning. Under the broadest reasonable interpretation, the limitations merely recite steps that apply a generic feed-forward or backpropagation learning, which represents merely adding the words “apply it”, or an equivalent, which are not indicative of an inventive concept (MPEP 2106.05(f)). Therefore, claim 18 does not solve the deficiencies in claim 16.
Regarding claim 19, it is dependent upon claim 16 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 19 recites wherein performing the private inference comprises calculating a polynomial activation function using a Beaver Triple (BT) protocol. Under the broadest reasonable interpretation, the limitations recite calculating a beaver triple operation which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 19 does not solve the deficiencies of claim 16.
Regarding claim 20, it is dependent upon claim 16 and fails to resolve the deficiencies identified above by integrating the judicial exception into a practical application, or introducing significantly more than the judicial exception. For example, claim 20 recites wherein the neural network comprises a convolution layer configured to perform a convolution operation. Under the broadest reasonable interpretation, the limitations recite performing a convolution operation which is interpreted as using a mathematical calculation. A mathematical calculation is interpreted as a mathematical concept. Therefore, claim 20 does not solve the deficiencies of claim 16.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-14 are rejected under 35 U.S.C. 103 as being unpatentable over Liu, et al., Non-Patent Literature “Privacy-Preserving All Convolutional Net Based on Homomorphic Encryption” (“Liu”) in view of Lokhande, et al., Non-Patent Literature “Generating Accurate Pseudo-labels in Semi-Supervised Learning and Avoiding Overconfident Predictions via Hermite Polynomial Activations” (“Lokhande”).
Regarding claim 1, Liu discloses:
An operating method of a deep learning system configured to perform a private inferences, comprising: obtaining encrypted input values; executing a convolutional neural network with respect to the encrypted input values, (Liu, pg. 753, “Our work is to construct efficient privacy-preserving protocols for an All Convolutional Networks [An operating method of a deep learning system configured to perform a private inferences,]. In the proposed protocol, the private information will be encrypted and sent to the cloud service provider in an encrypted form, and the cloud can calculate the predicted result of the encrypted data [comprising: obtaining encrypted input values; executing a convolutional neural network with respect to the encrypted input values,].”).
wherein the convolutional neural network comprises: a convolution layer configured to perform a convolution operation,… (Liu, pg. 754, “Convolution processing is a method for extracting some primary signal features by local connecting and weight sharing to simulate neural cells with local receptive fields [configured to perform a convolution operation,…]. Local connection means that each neuron on the convolutional layer [wherein the convolutional neural network comprises: a convolution layer] establishes a connection with the neurons in the fixed area of the previous layer.”).
and providing an output of the convolutional neural network as an inference result for the input values. (Liu, abstract, “In this paper, we proposed an encrypted all convolutional net that transformed traditional all convolutional net into a net based on homomorphic encryption. This scheme allows different data holders to send their encrypted data to cloud service, complete predictions, and return them in encrypted form as the cloud service provider does not have a secret key [and providing an output of the convolutional neural network as an inference result for the input values.].”).
While Liu teaches a system of private inference using homomorphic encryption and activation functions, Liu does not explicitly teach:
and a Hermite block configured to apply an activation function based on a Hermitic polynomial expansion;
Lokhande teaches and a Hermite block configured to apply an activation function based on a Hermitic polynomial expansion; (Lokhande, pg. 3, “We will use an expansion based on Hermite polynomials as a substitute for ReLU activations [and a Hermite block configured to apply an activation function based on a Hermitic polynomial expansion;].”).
Liu and Lokhande are both in the same field of endeavor (i.e. neural networks). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Liu and Lokhande to teach the above limitation(s). The motivation for doing so is using Hermite polynomials for activation functions improves the convergence rate of the model (cf. Lokhande, pg. 11, “we show how incorporating Hermite activations within a network makes the loss landscape smoother relative to ReLUs. Smoother landscapes implying faster convergence is not controversial [7]. One difference between ReLUs and Hermites is the nonsmooth behavior: for ReLU networks, standard first order methods require O(1/ϵ2) (versus O(1/ϵ) iterations for Hermite nets) to find a local minima.”).
Regarding claim 2, Liu in view of Lokhande teaches the method of claim 1. Liu further teaches wherein the encrypted input values are values encrypted based on a homomorphic encryption scheme. (Liu, pg. 756, “Fully homomorphic encryption is known as the Holy Grail in cryptography. Fully homomorphic encryption means that the ciphertext can be arbitrarily calculated without knowing the secret key, that is, for any function f and plaintext m, f(Enc(m)) = Enc(f(m)) [wherein the encrypted input values are values encrypted based on a homomorphic encryption scheme.].”).
Regarding claim 3, Liu in view of Lokhande teaches the method of claim 2. Liu further teaches wherein the homomorphic encryption scheme is at least one of BGV (Brakerski-Gentry-Vaikuntanathan), BFV (Brakerski-Fan-Vercauteren), or CKKS (Cheon-Kim-Kim-Song). (Liu, pg. 757, “Our scheme is based on the BGV [17] homomorphic cryptosystem using keyswitch technology and modulus-switch technology, which is one of the most efficient all-homomorphic encryption schemes [wherein the homomorphic encryption scheme is at least one of BGV (Brakerski-Gentry-Vaikuntanathan),].”).
Regarding claim 4, Liu in view of Lokhande teaches the method of claim 1. Lokhande further teaches wherein the Hermitic polynomial expansion is a Hermitic expansion of a rectified linear unit (ReLU) activation function. (Lokhande, pg. 4, “et x = (x1, …, xn) be an input to a neuron in a neural network and y be the output. Let w = (w1, w2, …, wn) be the weights associated with the neuron. Let σ be the non-linear activation applied to wT x. Often we set σ = ReLU. Here, we investigate the scenario where σ(x) = ∑i =0ciℎi(x), denoted as σhermite, were hi’s are as defined previously and ci’s are trainable parameters. As [6] suggests, we initialized the parameters ci’s associated with hermites to be ci = σi, where σi = <ReLU, ℎi> [wherein the Hermitic polynomial expansion is a Hermitic expansion of a rectified linear unit (ReLU) activation function.].”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Lokhande with the teachings of Liu for the same reasons disclosed in claim 1.
Regarding claim 5, Liu in view of Lokhande teaches the method of claim 1. Lokhande further teaches wherein the Hermitic polynomial expansion comprises a second-order Hermite polynomial. (Lokhande, pg. 5, “When implemented naively, Hermite activations do not work well for deeper architectures directly, which may be a reason that they have not been carefully explored so far. Basically, with no adjustments, we encounter a number of numerical issues that are not easy to fix. In fact, [8] explicitly notes that higher-order polynomials tend to make the activations unbounded making the training unstable. Fortunately, a trick mentioned in [1] in the context of quadratic functions, addresses the problem; quadratic hermitic activations is interpreted as a second-order Hermite polynomial (i.e. wherein the Hermitic polynomial expansion comprises a second-order Hermite polynomial.).”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Lokhande with the teachings of Liu for the same reasons disclosed in claim 1.
Regarding claim 6, Liu in view of Lokhande teaches the method of claim 1. Lokhande further teaches wherein the activation function is based on a batch-normalized Hermite polynomial. (Lokhande, pg. 22 and Figure 3, “We introduce softsign function to handle the numerical issues from the unbounded nature of Hermite polynomials. W denotes the weight, BN denotes batch normalization [wherein the activation function is based on a batch-normalized Hermite polynomial.], σ is the Hermite activation and SS is the softsign function.”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Lokhande with the teachings of Liu for the same reasons disclosed in claim 1.
Regarding claim 7, Liu in view of Lokhande teaches the method of claim 6. Lokhande further teaches wherein the activation function comprises a sum of terms of the batch-normalized Hermite polynomial multiplied by respective coefficients. (Lokhande, pg. 14, “Let fk(x)=∑jakj∑i=0∞ ciℎiwj Tx, be a one-hidden-layer network with the sum of infinite series of hermite polynomials as an activation function [wherein the activation function comprises a sum of terms of the batch-normalized Hermite polynomial multiplied by respective coefficients.]”, and Lokhande, pg. 22 and Figure 3, “We introduce softsign function to handle the numerical issues from the unbounded nature of Hermite polynomials. W denotes the weight, BN denotes batch normalization [batch-normalized Hermite polynomial], σ is the Hermite activation and SS is the softsign function.”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Lokhande with the teachings of Liu for the same reasons disclosed in claim 1.
Regarding claim 8, Liu in view of Lokhande teaches the method of claim 1. Lokhande further teaches wherein the Hermitic polynomial expansion comprises a Hermitic expansion of a non-linear activation function. (Lokhande, pg. 4, “et x = (x1, …, xn) be an input to a neuron in a neural network and y be the output. Let w = (w1, w2, …, wn) be the weights associated with the neuron. Let σ be the non-linear activation applied to wT x. Often we set σ = ReLU. Here, we investigate the scenario where σ(x) = ∑i =0ciℎi(x), denoted as σhermite, were hi’s are as defined previously and ci’s are trainable parameters. As [6] suggests, we initialized the parameters ci’s associated with hermites to be ci = σi, where σi = <ReLU, ℎi>; a ReLU is interpreted as a non-linear activation function (i.e. wherein the Hermitic polynomial expansion comprises a Hermitic expansion of a non-linear activation function.).”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Lokhande with the teachings of Liu for the same reasons disclosed in claim 1.
Regarding claim 9, Liu in view of Lokhande teaches the method of claim 1. Lokhande further teaches wherein the deep learning system uses at least of one of a visual geometry group (VGG), residual neural network (ResNet), or Preactivation ResNet. (Lokhande, pg. 5, “With the above modification in hand, we can use our activation function within Resnet18 using Pre-activation Blocks [10, 11]. In the preactivation block, we found that having the second softsign function after the weight layer is useful. The slightly modified preactivation block of ResNets is shown in Figure 3. We train ResNets with Hermite activations on CIFAR10 to assess the general behavior of our substitution [wherein the deep learning system uses at least of one of a visual geometry group (VGG), residual neural network (ResNet), or Preactivation ResNet.].”).
Liu and Lokhande are all in the same field of endeavor (i.e. neural networks). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Liu and Lokhande to teach the above limitation(s). The motivation for doing so is that using a ResNet removes the common vanishing gradient problem with deep neural networks to improve training.
Regarding claim 10, Liu in view of Lokhande teaches the method of claim 1. Lokhande further teaches wherein the activation function is limited to addition operations and multiplication operations. (Lokhande, pg. 3, “We will use an expansion based on Hermite polynomials as a substitute for ReLU activations; using polynomials for the activation function is interpreted as being limited to addition and multiplication operations (i.e. wherein the activation function is limited to addition operations and multiplication operations.).”).
It would have been obvious to one of ordinary skill in the art before the effective filling date of the present application to combine the teachings of Lokhande with the teachings of Liu for the same reasons disclosed in claim 1.
Regarding claim 11, Liu discloses:
A deep learning system including a neural network trained to perform private inferences, comprising: a client device configured to pre-calculate randomly generated data, to obtain pre-calculated data; and a cloud server configured to: receive the pre-calculated data from the client device, (Liu, pg. 753, “Our work is to construct efficient privacy-preserving protocols for an All Convolutional Networks [A deep learning system including a neural network trained to perform private inferences,]. In the proposed protocol, the private information will be encrypted [comprising: a client device configured to pre-calculate randomly generated data, to obtain pre-calculated data;] and sent to the cloud service provider in an encrypted form, and the cloud can calculate the predicted result of the encrypted data [and a cloud server configured to: receive the pre-calculated data from the client device,].”).
generate, using the neural network, operated values by performing a homomorphic encryption operation with respect to the received pre-calculated data,…and output the operated values. (Liu, abstract, “In this paper, we proposed an encrypted all convolutional net that transformed traditional all convolutional net into a net based on homomorphic encryption [generate, using the neural network, operated values by performing a homomorphic encryption operation]. This scheme allows different data holders to send their encrypted data to cloud service [with respect to the received pre-calculated data,…], complete predictions, and return them in encrypted form as the cloud service provider does not have a secret key [and output the operated values.].”).
While Liu teaches a system of private inference using homomorphic encryption and activation functions, Liu does not explicitly teach:
wherein the neural network is configured to apply an activation function based on a Hermitic polynomial expansion,
Lokhande teaches wherein the neural network is configured to apply an activation function based on a Hermitic polynomial expansion, (Lokhande, pg. 3, “We will use an expansion based on Hermite polynomials as a substitute for ReLU activations [wherein the neural network is configured to apply an activation function based on a Hermitic polynomial expansion,].”).
Liu and Lokhande are both in the same field of endeavor (i.e. neural networks). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Liu and Lokhande to teach the above limitation(s). The motivation for doing so is using Hermite polynomials for activation functions improves the convergence rate of the model (cf. Lokhande, pg. 11, “we show how incorporating Hermite activations within a network makes the loss landscape smoother relative to ReLUs. Smoother landscapes implying faster convergence is not controversial [7]. One difference between ReLUs and Hermites is the nonsmooth behavior: for ReLU networks, standard first order methods require O(1/ϵ2) (versus O(1/ϵ) iterations for Hermite nets) to find a local minima.”).
Regarding claim 12, Liu in view of Lokhande teaches the deep learning system of claim 11. Liu further teaches wherein performing the homomorphic encryption operation includes performing a convolution operation based on a homomorphic encryption scheme, and wherein outputting the operated values includes transmitting the operated values to the client device. (Liu, abstract, “In this paper, we proposed an encrypted all convolutional net that transformed traditional all convolutional net into a net based on homomorphic encryption [wherein performing the homomorphic encryption operation includes performing a convolution operation based on a homomorphic encryption scheme,]. This scheme allows different data holders to send their encrypted data to cloud service, complete predictions, and return them in encrypted form as the cloud service provider does not have a secret key [and wherein outputting the operated values includes transmitting the operated values to the client device.].”).
Regarding claim 13, Liu in view of Lokhande teaches the deep learning system of claim 12. Liu further teaches wherein the client device is configured to obtain a result value using the activation function and the operated values. (Liu, pg. 753, “Our work is to construct efficient privacy-preserving protocols for an All Convolutional networks. In the proposed protocol, the private information will be encrypted and sent to the cloud service provider in an encrypted form, and the cloud can calculate the predicted result of the encrypted data [and the operated values.]. The public key of scheme is used for encryption and the private key of scheme is used for decryption [wherein the client device is configured to obtain a result value].”, and Liu, pg. 758, “The accuracy of full convolutional net depends on the ReLU function. To compute the ReLU function homomorphically, we can use the polynomial approximation of the ReLU [using the activation function]”).
Regarding claim 14, Liu in view of Lokhande teaches the deep learning system of claim 13. Liu further teaches wherein the client device is configured to obtain the result value by performing the homomorphic encryption operation on plaintext on which the pre-calculated data is based. (Liu, pg. 757, “Our scheme is based on the BGV [17] homomorphic cryptosystem using keyswitch technology and modulus-switch technology, which is one of the most efficient all-homomorphic encryption schemes… The plaintext space of the BGV scheme operates on a polynomial ring, which is modulo the cyclotomic polynomial At :Zt[x]/ðm(x), where ðm(x) is the m-th polynomial. It contains five operations: KeyGen represents key generation algorithms, Encpk represents encryption [on which the pre-calculated data is based.], Decsk represents decryption [wherein the client device is configured to obtain the result value by performing the homomorphic encryption operation on plaintext], homAdd represents homomorphic addition and homMul represents homomorphic multiplication.”).
Claims 15-18 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Liu, et al., Non-Patent Literature “Privacy-Preserving All Convolutional Net Based on Homomorphic Encryption” (“Liu”) in view of Lokhande, et al., Non-Patent Literature “Generating Accurate Pseudo-labels in Semi-Supervised Learning and Avoiding Overconfident Predictions via Hermite Polynomial Activations” (“Lokhande”) and further in view of Nandakumar, et al., US Pre-Grant Publication US20200366459A1 (“Nandakumar”).
Regarding claim 15, Liu in view of Lokhande teaches the deep learning system of claim 11. While the combination teaches cloud server performing operations using an activation function, the combination does not explicitly teach wherein the cloud server is configured to perform an operation based on a multi-party computation technique using the activation function.
Nandakumar teaches wherein the cloud server is configured to perform an operation based on a multi-party computation technique using the activation function. (Nandakumar, ⁋95 and Figure 6A, “Referring to FIG. 6A, this figure is a flowchart of an exemplary method for secure multi-party learning from encrypted data, in accordance with an exemplary embodiment; Figure 6A shows that a service provider uses MPC with an AI model thus a server performs a MPC using an activation function (i.e. wherein the cloud server is configured to perform an operation based on a multi-party computation technique using the activation function.).”).
Liu, in view of Lokhande, and Nandakumar are both in the same field of endeavor (i.e. homomorphic encryption). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Liu, in view of Lokhande, and Nandakumar to teach the above limitation(s). The motivation for doing so is using multi-party computation keeps raw data confidential between multiple users (cf. Nandakumar, ⁋7-8).
Regarding claim 16, Liu discloses:
An operating method of a deep learning system, comprising: collecting data; training a prediction model based on the collected data; and performing a private inference using the prediction model, (Liu, pg. 753, “Our work is to construct efficient privacy-preserving protocols for an All Convolutional Networks [training a prediction model based on the collected data;]. In the proposed protocol, the private information [An operating method of a deep learning system, comprising: collecting data;] will be encrypted and sent to the cloud service provider in an encrypted form, and the cloud can calculate the predicted result of the encrypted data [and performing a private inference using the prediction model,].”).
wherein the private inference includes a homomorphic encryption operation…, (Liu, abstract, “In this paper, we proposed an encrypted all convolutional net that transformed traditional all convolutional net into a net based on homomorphic encryption [wherein the private inference includes a homomorphic encryption operation…,].”). and…includes executing a neural network configured to apply an activation function…. (Liu, pg. 758, “The accuracy of full convolutional net depends on the ReLU function. To compute the ReLU function homomorphically, we can use the polynomial approximation of the ReLU [and…includes executing a neural network configured to apply an activation function….]”).
While Liu teaches a system of private inference using homomorphic encryption and activation functions, Liu does not explicitly teach:
and a multi-party computation
wherein the multi-party computation includes a neural network with an activation function
activation function based on a Hermitic polynomial expansion
Lokhande teaches an activation function based on a Hermitic polynomial expansion (Lokhande, pg. 3, “We will use an expansion based on Hermite polynomials as a substitute for ReLU activations [activation function based on a Hermitic polynomial expansion].”).
Liu and Lokhande are both in the same field of endeavor (i.e. neural networks). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Liu and Lokhande to teach the above limitation(s). The motivation for doing so is using Hermite polynomials for activation functions improves the convergence rate of the model (cf. Lokhande, pg. 11, “we show how incorporating Hermite activations within a network makes the loss landscape smoother relative to ReLUs. Smoother landscapes implying faster convergence is not controversial [7]. One difference between ReLUs and Hermites is the nonsmooth behavior: for ReLU networks, standard first order methods require O(1/ϵ2) (versus O(1/ϵ) iterations for Hermite nets) to find a local minima.”).
While Liu in view of Lokhande teaches cloud server performing operations using an activation function, the combination does not explicitly teach:
and a multi-party computation
wherein the multi-party computation includes a neural network with an activation function
Nandakumar teaches:
and a multi-party computation (Nandakumar, ⁋95 and Figure 6A, “Referring to FIG. 6A, this figure is a flowchart of an exemplary method for secure multi-party learning [and a multi-party computation] from encrypted data, in accordance with an exemplary embodiment.”).
wherein the multi-party computation includes a neural network with an activation function (Nandakumar, ⁋95 and Figure 6A, “Referring to FIG. 6A, this figure is a flowchart of an exemplary method for secure multi-party learning from encrypted data, in accordance with an exemplary embodiment; Figure 6A shows that a service provider uses MPC with an AI model thus a server performs a MPC using an activation function (i.e. wherein the multi-party computation includes a neural network with an activation function).”).
Liu, in view of Lokhande, and Nandakumar are both in the same field of endeavor (i.e. homomorphic encryption). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Liu, in view of Lokhande, and Nandakumar to teach the above limitation(s). The motivation for doing so is using multi-party computation keeps raw data confidential between multiple users (cf. Nandakumar, ⁋7-8).
Regarding claim 17, Liu in view of Lokhande and Nandakumar teaches the method of claim 16. Liu further teaches wherein the collecting the data comprises generating a ciphertext using a homomorphic encryption scheme on plaintext data. (Liu, pg. 757, “Our scheme is based on the BGV [17] homomorphic cryptosystem using keyswitch technology and modulus-switch technology, which is one of the most efficient all-homomorphic encryption schemes… The plaintext space of the BGV scheme operates on a polynomial ring, which is modulo the cyclotomic polynomial At :Zt[x]/ðm(x), where ðm(x) is the m-th polynomial. It contains five operations: KeyGen represents key generation algorithms, Encpk represents encryption [wherein the collecting the data comprises generating a ciphertext using a homomorphic encryption scheme on plaintext data.], Decsk represents decryption, homAdd represents homomorphic addition and homMul represents homomorphic multiplication.”).
Regarding claim 18, Liu in view of Lokhande and Nandakumar teaches the method of claim 16. Liu further teaches wherein training the prediction model comprises performing at least one of a feed-forward learning or a backpropagation learning. (Liu, pg. 753, “The Convolutional Neural Network (CNN) is a feed-forward neural network [wherein training the prediction model comprises performing at least one of a feed-forward learning].”).
Regarding claim 20, Liu in view of Lokhande and Nandakumar teaches the method of claim 16. Liu further teaches wherein the neural network comprises a convolution layer configured to perform a convolution operation. (Liu, pg. 754, “Convolution processing is a method for extracting some primary signal features by local connecting and weight sharing to simulate neural cells with local receptive fields [configured to perform a convolution operation.]. Local connection means that each neuron on the convolutional layer [wherein the neural network comprises a convolution layer] establishes a connection with the neurons in the fixed area of the previous layer.”).
Claim 19 is rejected under 35 U.S.C. 103 as being unpatentable over Liu, et al., Non-Patent Literature “Privacy-Preserving All Convolutional Net Based on Homomorphic Encryption” (“Liu”) in view of Lokhande, et al., Non-Patent Literature “Generating Accurate Pseudo-labels in Semi-Supervised Learning and Avoiding Overconfident Predictions via Hermite Polynomial Activations” (“Lokhande”) and further in view of Nandakumar, et al., US Pre-Grant Publication US20200366459A1 (“Nandakumar”) and Mishra, et al., Non-Patent Literature “Delphi: A Cryptographic Inference Service for Neural Networks” (“Mishra”).
Regarding claim 19, Liu in view of Lokhande and Nandakumar teaches the method of claim 16. While the combination teaches cloud server performing operations using an activation function and multi-party computations, the combination does not explicitly teach wherein performing the private inference comprises calculating a polynomial activation function using a Beaver Triple (BT) protocol.
Mishra teaches wherein performing the private inference comprises calculating a polynomial activation function using a Beaver Triple (BT) protocol. (Mishra, pg. 2510 col. 1, “This step depends on the activation type: (a) ReLU: The server constructs C by garbling the circuit C described in Fig. 5. It sends C to the client and simultaneously, the server and the client exchange labels for the input wires corresponding to ri+1 and Mi·ri−si via an Oblivious Transfer (OT). (b) Polynomial approximaitons: The client and the server run the Beaver’s triples generation protocol to generate a number of Beaver’s multiplication triples [wherein performing the private inference comprises calculating a polynomial activation function using a Beaver Triple (BT) protocol.].”).
Liu, in view of Lokhande and Nandakumar, and Mishra are both in the same field of endeavor (i.e. private inferencing). It would have been obvious for a person having ordinary skill in the art before the effective filing date of the claimed invention to combine Liu, in view of Lokhande and Nandakumar, and Mishra to teach the above limitation(s). The motivation for doing so is that Beaver Triple protocol improves the security of polynomial calculations (cf. Mishra, pg. 2509 col. 1, “Beaver’s multiplication procedure is a secure protocol such that at the end of the protocol, parties P1 and P2 hold an additive secret sharing of xy. We provide details of this protocol in Appendix A but note here that this protocol can be used to securely evaluate any polynomial.”).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. Chandran, et al., US20230032519A1 discloses performing private inferencing using secure two-party computations.
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 NICHOLAS S WU whose telephone number is (571)270-0939. The examiner can normally be reached Monday - Friday 8:00 am - 4:00 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, Michelle Bechtold can be reached at 571-431-0762. 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.
/N.S.W./Examiner, Art Unit 2148 /MICHELLE T BECHTOLD/Supervisory Patent Examiner, Art Unit 2148
Prosecution Insights