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 .
This action is in response to the claims filed 12/01/2025. Claims 1-3, 5-11, and 13-23 are pending. Claims 1 (a method), and 11 (a machine) are independent.
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 12/01/2025 has been entered.
Response to Arguments
On pages 7-9 of the remarks Applicant states that the term “homomorphic conjugation” is known in the art and notes Cheon’s discussion of a “slot-wise conjugation”. This is persuasive and the 112(b) rejection is withdrawn. The term homomorphic conjugation will be interpreted to be any conjugation related to homomorphic cryptography.
Applicant's arguments filed 12/01/2025 have been fully considered but they are not persuasive.
On pages 11 and 12 of the remarks Applicant states: “Cheon is related to a method of CKKS (HEAAN) without a neural network”. This is not persuasive as Cheon is combined with other references and explicitly contemplates its application to neural networks on page 4, last paragraph.
On pages 11-12 of the Remarks Applicant states: “In Cheon, the extracted imaginary part mj is then used to recover the original plaintext message from the ciphertext’s decryption structure, completing the bootstrapping procedure …”
This argument is not persuasive.
With reference to Cheon Fig. 2, and page 376, the only act after the “Extraction of the Imaginary Part” is the “Switching Back to the Coefficient Representation”. There is no explicit evidence within Cheon that the switching step includes the imaginary part that was previously extracted.
Note the newly added reference Lee “Low-Complexity Deep Convolutional Neural Networks on Fully Homomorphic Encryption Using Multiplexed Parallel Convolutions” discloses that even with “imaginary-removing bootstrapping” that imaginary data persists. See Lee § 6.
Applicant’s asserts that removal and discarding of the imaginary part of the encrypted data distinguishes over Cheon. However, Cheon also performs “Extraction of the Imaginary Part”. And the claim sets forth no limitations or actions after the extraction that would distinguish over the Cheon reference.
Any distinction between the claimed invention and Cheon with regard to the imaginary component cannot be discerned from the presently presented claims that perform no acts beyond the separation of the imaginary part.
The claim does not set forth further specific acts that are performed with the absence of the imaginary part making it impossible to articulate a distinction over the art of record.
On page 13 Applicant states: “Rather, only the present application discloses example reasons for (claim 1).”
In response to applicant's argument that the examiner's conclusion of obviousness is based upon improper hindsight reasoning, it must be recognized that any judgment on obviousness is in a sense necessarily a reconstruction based upon hindsight reasoning. But so long as it takes into account only knowledge which was within the level of ordinary skill at the time the claimed invention was made, and does not include knowledge gleaned only from the applicant's disclosure, such a reconstruction is proper. See In re McLaughlin, 443 F.2d 1392, 170 USPQ 209 (CCPA 1971).
In this instance, “It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine Lee with Cheon as Lee explicitly contemplates such a use, (“we implement boot-strapping of the RNS-CKKS scheme in the SEAL library according to [6]-[10] to support a large number of homo-morphic operations for a deep neural network,” Lee § A. [6] is Cheon).”
Regarding Applicant’s arguments on pages 14-20 with respect to the § 101 rejection. Note the updated discussion in the § 101 rejection below.
Specifically, the independent claims 1 and 11 are directed exclusively to Applicant’s “Equation 3”. Equation 3 is unarguably a mathematical equation.
Although there is a discussion of homographic cryptography, no data is encrypted, decrypted, or otherwise used by the claim to accomplish a task or benefit. This could be viewed as a non-limiting intended use, or a non-limiting descriptive material. However, it is sufficient to note that mere intended use of Equation 3 does not transform an ineligible mathematical process (Equation 3) into a patentable one. MPEP 2106.05(h): “As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible “simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use.” Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981).”
Applicant mentions Ex Parte Desjardins in reference to the patentability of AI inventions.
The pending claims have nothing to do with AI.
The claims do not train a model, implement a model, or use a model to make a determination.
Rather the claims are exclusively directed to a particular mathematical formulae (Equation 3) that is suggested for use in homomorphically encrypted data, such as that in neural networks. However, for the purposes of the claim, the particular data is irrelevant, as is the machine learning elements, and even the homomorphic cryptography is not used in any application practical or otherwise. This is because the claim is narrowly drafted to cover only a specific calculation involving imaginary numbers.
Improvements to a computer, improvements to AI, and improvements to homomorphic cryptography may be patent eligible. However, when an improvement is restricted to be exclusively a single mathematical equation it is ineligible.
Applicant’s further remarks are dependent on those addressed and are not persuasive for the reasons discussed above.
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 1-2, 5-11, and 14-23 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The claim(s) recite(s) a mathematical algorithm. This judicial exception is not integrated into a practical application because the mathematical algorithm is recited only in the abstract and not actually used in an application. The claim(s) does/do not include additional elements that are sufficient to amount to significantly more than the judicial exception because there are no additional elements beyond the performance of math.
In more detail, Applicant’s specification ¶¶ 86-87 state: In operation 330, the neural network operation apparatus 10 may remove noise of the encrypted data based on encrypted data x and homomorphic conjugation data x …. according to Equation 3 below,
PNG
media_image1.png
94
841
media_image1.png
Greyscale
”
The independent claims 1 and 11, require:
“generate homomorphic conjugation of data …
… Removing noise of the encrypted data …
… remove an imaginary part of the encrypted data…”
Every significant step within claims 1 and 11 is embodied in Equation 3. The claims are a mathematical algorithm.
Although claim 1 further requires one or more processors and a memory, this is merely a statement to ‘apply it’ to a machine. See MPEP 2106.04(d).I.
Note that the independent claims 1 and 11 merely performs Applicant’s Equation 3. While dependent claims set forth further mathematical steps for the purpose of homomorphic cryptography of a neural network, no data is encrypted/received, decrypted/presented, nor is any particular learning performed. Thus, the further dependent claims merely suggest a ‘field of use’ and does not transform the abstract mathematical algorithm into a patent eligible one. See MPEP 2106.05.I.A.iv and MPEP 2106.05(h): “As explained by the Supreme Court, a claim directed to a judicial exception cannot be made eligible “simply by having the applicant acquiesce to limiting the reach of the patent for the formula to a particular technological use.” Diamond v. Diehr, 450 U.S. 175, 192 n.14, 209 USPQ 1, 10 n. 14 (1981).”
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.
Claim(s) 1-2, 5-11, and 14-23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al., “Privacy-Preserving Machine Learning With Fully Homomorphic Encryption for Deep Neural Network” (published 3/22/2022 and submitted in IDS filed 10/31/2023), in view of Lee et al., “Low-Complexity Deep Convolutional Neural Networks on Fully Homomorphic encryption Using Multiplexed Parallel Convolutions” (published May, 2022); hereafter Lee2.
Note with regard to the similarities in inventorship between this application and the Lee2 reference: See MPEP 2153.01(a): “If, however, the application names fewer joint inventors than a publication (e.g., the application names as joint inventors A and B, and the publication names as authors A, B and C), it would not be readily apparent from the publication that it is an inventor-originated disclosure and the publication would be treated as prior art under AIA 35 U.S.C. 102(a)(1)”.
As to claims 1, 11, and 20 Lee discloses the method/machine/CRM comprising:
one or more processors configured to: (see Lee abstract)
a memory comprising one or more non-transitory storage media that store instructions that, when executed by the one or more processors, configures the apparatus to: (see Lee abstract)
generate homomorphic conjugation data of encrypted data by performing a homomorphic conjugation operation on the encrypted data, wherein the encrypted data corresponds to an output of each of a plurality of layers included in a neural network; and (see Lee § II.A, “homomorphic…. conjugation”)
generate an updated encrypted data for [further homomorphic encryption operation (non-limiting intended future use)] by removing noise of the encrypted data based on the encrypted data and the homomorphic conjugation data. (see Lee §§ II.C and IV, “bootstrapping”)
Lee does not disclose:
wherein, for the removing of the noise, the execution of the instructions further configures the apparatus to remove an imaginary part of the encrypted data.
Lee2 discloses:
wherein, for the removing of the noise, the execution of the instructions further configures the apparatus to remove an imaginary part of the encrypted data.
(“Hence, to stably perform ResNet with many layers, it is important to remove the imaginary part of the input of each APR…. Figure 6 shows the mean of absolute values of imaginary parts after each layer using normal and imaginary-removing bootstrappings for one instance of ResNet-110 inference…. The proposed imaginary-removing bootstrapping makes the noise of imaginary parts remain much smaller during deeper ResNet inference,” Lee2 § 5. Lee Figure 6 showing a reduced imaginary component relative to the prior non-removing method.)
A person of ordinary skill in the art before the effective filing date of the claimed invention would have combined Lee with Lee2 by performing the imaginary removing bootstrapping discussed in Lee2. It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine Lee with Lee2 in order to prevent catastrophic divergence that causes failure in the model, Lee2 § 5.
As to claims 5, 14 Lee in view of Lee2 discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
for the removing of the noise, the one or more processors are configured to: perform an addition operation of the encrypted data and the homomorphic conjugation data; (“We homomorphically compute the formula Re(x) = x/2 + x/2” Lee2 § 5) and multiply a result of the addition operation by a predetermined value. (“Since the bootstrapping and APR work only for input values in [−1, 1], it is required to do scaling by 1/B before bootstrapping and by B after the APR. We set sufficiently large B to maintain all the computed values within [−B,B]. We set B = 40 and B = 65 for the CIFAR-10” Lee2 § 6)
As to claim 23 Lee in view of Lee2 discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
wherein the homomorphic conjugation operation changes sign of the imaginary part of the encrypted data to generate the homomorphic conjugation data, (“to remove the imaginary part of the input of each APR. We propose to apply the imaginary-removing bootstrapping operation before the APR. We homomorphically compute the formula Re(x) = x/2 + x/2 by halving all coefficient values in SLOTTOCOEFF operation in the bootstrapping and homomorphically computing v + ¯v.” Lee2 § 5. This is a changed sign of the imaginary part.)
and wherein the removal of the imaginary part is implemented by performing an arithmetic operation based on the encrypted data and the homomorphic conjugation data. (“and homomorphically computing v + ¯v.” Lee2 § 5)
Regarding claim 2, see Lee generally.
Regarding claims 6 and 15 (see Lee §§ II.C and IV, “bootstrapping”).
Regarding claims 7 and 16 (see Lee § II.C, “FFT”)
Regarding claims 8 and 17 (see Lee Algorithm 5)
Regarding claims 9 and 18 (see Lee Algorithm 5)
Regarding claims 10 and 19 (see Lee § V.D, “RELU”. and see Lee Algorithm 5)
Regarding claims 21 and 22 (see Lee § V.A.3 discussing error and bounding operations)
Claim(s) 1-2, 5-11, and 14-23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Lee et al., “Privacy-Preserving Machine Learning With Fully Homomorphic Encryption for Deep Neural Network” (published 3/22/2022 and submitted in IDS filed 10/31/2023), in view of Cheon et al., “Bootsrapping for Approximate Homomorphic Encryption” (published 2018).
While the effective filing date of the present Application is 2/10/2023, less than a year after Lee et al., Lee has eleven (11) authors implying that some of the content therein may be authored by non-inventors of the present application.
As to claims 1, 11, and 20 Lee discloses the method/machine/CRM comprising:
one or more processors configured to: (see Lee abstract)
a memory comprising one or more non-transitory storage media that store instructions that, when executed by the one or more processors, configures the apparatus to: (see Lee abstract)
generate homomorphic conjugation data of encrypted data by performing a homomorphic conjugation operation on the encrypted data, wherein the encrypted data corresponds to an output of each of a plurality of layers included in a neural network; and (see Lee § II.A, “homomorphic…. conjugation”)
generate an updated encrypted data for [further homomorphic encryption operation (non-limiting intended future use)] by removing noise of the encrypted data based on the encrypted data and the homomorphic conjugation data. (see Lee §§ II.C and IV, “bootstrapping”)
Lee does not disclose:
wherein, for the removing of the noise, the execution of the instructions further configures the apparatus to remove an imaginary part of the encrypted data.
Cheon discloses:
wherein, for the removing of the noise, the execution of the instructions further configures the apparatus to remove an imaginary part of the encrypted data.
(“we can extract the imaginary (sine) part by conjugation operation (i.e., 2 sin θ = exp(iθ) − exp(−iθ)), which will be described in the next section.” Cheon § 3.2)
A person of ordinary skill in the art before the effective filing date of the claimed invention would have combined Lee with Cheon in order to implement bootstrapping for homomorphic encryption. It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine Lee with Cheon as Lee explicitly contemplates such a use, (“we implement boot-strapping of the RNS-CKKS scheme in the SEAL library according to [6]-[10] to support a large number of homo-morphic operations for a deep neural network,” Lee § A. [6] is Cheon)
As to claims 5, 14 Lee in view of Cheon discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
for the removing of the noise, the one or more processors are configured to: perform an addition operation of the encrypted data and the homomorphic conjugation data; and multiply a result of the addition operation by a predetermined value.
(“ct ← Add(ct , ctj ) (mod q)” Cheon Algorithm 1 and 2. See Cheon § 5.1 using algorithm 1 in recryption.).
As to claim 23 Lee in view of Cheon discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
wherein the homomorphic conjugation operation changes sign of the imaginary part of the encrypted data to generate the homomorphic conjugation data, and (Cheon § 3.2)
wherein the removal of the imaginary part is implemented by performing an arithmetic operation based on the encrypted data and the homomorphic conjugation data. (“we can extract the imaginary (sine) part by conjugation operation (i.e., 2 sin θ = exp(iθ) − exp(−iθ)), which will be described in the next section.” Cheon § 3.2)
Regarding claim 2, see Lee generally.
Regarding claims 6 and 15 (see Lee §§ II.C and IV, “bootstrapping”).
Regarding claims 7 and 16 (see Lee § II.C, “FFT”)
Regarding claims 8 and 17 (see Lee Algorithm 5)
Regarding claims 9 and 18 (see Lee Algorithm 5)
Regarding claims 10 and 19 (see Lee § V.D, “RELU”. and see Lee Algorithm 5)
Regarding claims 21 and 22 (see Lee § V.A.3 discussing error and bounding operations)
Claim(s) 1-2, 5, 6, 11, 14-15, and 20-21, and 23 is/are rejected under 35 U.S.C. 103 as being unpatentable over Cheon et al., “Bootsrapping for Approximate Homomorphic Encryption” (published 2018), in view of Pulido-Gaytan et al., “Privacy-preserving neural networks with Homomorphic encryption: Challenges and opportunities” (published 2021).
As to claims 1, 11, and 20 Cheon discloses the method/machine/CRM comprising:
An apparatus with a neural network operation of homomorphic encrypted data, the apparatus comprising: one or more processors a memory comprising one or more non-transitory storage media that store instructions that, when executed by the one or more processors, configures the apparatus to: (“We employ the ciphertext packing method and combine it with our efficient evaluation strategy to achieve a better performance in terms of memory and computation cost.” Cheon § 5.1)
generate homomorphic conjugation data of encrypted data by performing a homomorphic conjugation operation on the encrypted data, (“applying the evaluation method of slot-wise conjugation described in Sect. 4.2.” Cheon § 5.1. discussing a homomorphic computation, see Cheon title.) wherein the encrypted data corresponds to an output of each of
generate an updated encrypted data for [further homomorphic encryption operation (non-limiting intended future use)] by removing noise of the encrypted data based on the encrypted data and the homomorphic conjugation data. (“4.2 Rotation and Conjugation The purpose of the key-switching operation is to convert a ciphertext under a secret s into a ciphertext of the same message with respect to another secret key…. The following lemma shows the correctness of key-switching procedure and estimates a noise bound.” Cheon § 4.2)
wherein, for the removing of the noise, the execution of the instructions further configures the apparatus to remove an imaginary part of the encrypted data. (“we can extract the imaginary (sine) part by conjugation operation (i.e., 2 sin θ = exp(iθ) − exp(−iθ)), which will be described in the next section.” Cheon § 3.2)
Cheon does not explicitly disclose:
a plurality of layers included in a neural network
Pulido-Gaytan discloses:
one or more processors configured to: (“clouds provide services to facilitate this process, but it introduces new security threats of data breaches. Modern encryption techniques ensure security and are considered as the best option to protect stored data and data in transit from an unauthorized third-party. However, a decryption process is necessary when the data must be processed or analyzed, falling into the initial problem of data vulnerability. Fully Homomorphic Encryption (FHE) is considered the holy grail of cryptography. It allows a non-trustworthy third-party resource to process encrypted information without disclosing confidential data.” Pulido-Gaytan abstract)
a plurality of layers included in a neural network
(“clouds provide services to facilitate this process, but it introduces new security threats of data breaches. Modern encryption techniques ensure security and are considered as the best option to protect stored data and data in transit from an unauthorized third-party. However, a decryption process is necessary when the data must be processed or analyzed, falling into the initial problem of data vulnerability. Fully Homomorphic Encryption (FHE) is considered the holy grail of cryptography. It allows a non-trustworthy third-party resource to process encrypted information without disclosing confidential data.” Pulido-Gaytan abstract. “The interaction between neurons is essential for NN performance, the network structure defines the interaction between layers, subsets of grouped neurons. In the network sequence, the layer position establishes a specific role in the processing of data. Typically, NN is composed of three types of layers. The first layer is the input layer. It receives information from outside the network. Internal layers, also called hidden, are not directly accessible from the exterior. The last segment in the NN is the output layer. It transfers information outside of the network” Pulido-Gaytan § 5.1. Pulido-Gaytan Fig. 3)
A person of ordinary skill in the art before the effective filing date of the claimed invention would have combined Cheon with Pulido-Gaytan in order to implement a neural network with homomorphically encrypted data. It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine Cheon with Pulido-Gaytan as Cheon explicitly contemplates such a use, (“It has a number of applications in machine learning such as logistic regression and neural networks.” Cheon § 1)
As to claim 2 Cheon in view of Pulido-Gaytan discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
further comprising a receiver configured to receive the encrypted data, wherein, for the generating of the homomorphic conjugation data, the one or more processors are configured to generate the homomorphic conjugation data based on the received encrypted data. (Pulido-Gaytan Fig. 3, the “evaluate” function being the NN comprising the conjugation. “clouds provide services to facilitate this process, but it introduces new security threats of data breaches. Modern encryption techniques ensure security and are considered as the best option to protect stored data and data in transit from an unauthorized third-party. However, a decryption process is necessary when the data must be processed or analyzed, falling into the initial problem of data vulnerability. Fully Homomorphic Encryption (FHE) is considered the holy grail of cryptography. It allows a non-trustworthy third-party resource to process encrypted information without disclosing confidential data.” Pulido-Gaytan abstract)
As to claims 5, 14 Cheon in view of Pulido-Gaytan discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
for the removing of the noise, the one or more processors are configured to: perform an addition operation of the encrypted data and the homomorphic conjugation data; and multiply a result of the addition operation by a predetermined value.
(“ct ← Add(ct , ctj ) (mod q)” Cheon Algorithm 1 and 2. See Cheon § 5.1 using algorithm 1 in recryption.)
As to claim 6, Cheon in view of Pulido-Gaytan discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
wherein the encrypted data comprises data having a ciphertext level less than or equal to a predetermined threshold value, and the one or more processors are configured to: (“In summary, for an input ciphertext ct ∈ R2 q satisfying ct,sk = m (mod q), our bootstrapping process returns a ciphertext ct such that ct ,sk = m + e (mod Q1) for a modulus Q1 q and some error e with e can ∞ ≤ O( √ N · Brs). It consists of the initial/final linear transformations and the evaluation of the complex exponential function, so the total depth (number of levels consumed) for bootstrapping is O(log(Kq)) = O(log λ).” Cheon § 5.3, performing bootstrapping to constrain the noise/error)
perform a bootstrapping operation on the encrypted data; and (“The parameter log p is the bit size of plaintexts and the plaintext precision denotes the number of significant bits of plaintexts after bootstrapping” Cheon § 6.1)
for the removing of the noise, remove noise of data on which the bootstrapping operation is completed. (“The main idea is to consider an encryption error as part of a computational error that occurs during approximate computations. For an encryption ct of a message m with a secret key sk, the decryption algorithm [ct,sk]q outputs an approximate value m+e of the original message with a small error e. The main advantage of HEAAN comes from the rescaling procedure for managing the magnitude of plaintexts.” Cheon § 1. Constraining size and noise, see generally Cheon § 5.3).
As to claim 15, Cheon in view of Pulido-Gaytan discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
a bootstrapping operation on the encrypted data, (“Fig. 2. Pipeline of our bootstrapping process” Cheon § 5.1) wherein the encrypted data comprises data having a ciphertext level less than or equal to a predetermined threshold value (“In summary, for an input ciphertext ct ∈ R2 q satisfying ct,sk = m (mod q), our bootstrapping process returns a ciphertext ct such that ct ,sk = m + e (mod Q1) for a modulus Q1 q and some error e with e can ∞ ≤ O( √ N · Brs). It consists of the initial/final linear transformations and the evaluation of the complex exponential function, so the total depth (number of levels consumed) for bootstrapping is O(log(Kq)) = O(log λ).” Cheon § 5.3, performing bootstrapping to constrain the noise/error)
wherein the obtaining of the homomorphic conjugation data comprises obtaining homomorphic conjugation data of data on which the bootstrapping operation is completed, (“applying the evaluation method of slot-wise conjugation described in Sect. 4.2.” Cheon § 5.1. discussing a homomorphic computation, see Cheon title.)
wherein the removing of the noise comprises removing noise of the data on which the bootstrapping operation is completed, based on the data on which the bootstrapping operation is completed and the homomorphic conjugation data of the data on which the bootstrapping operation is completed.
(“The main idea is to consider an encryption error as part of a computational error that occurs during approximate computations. For an encryption ct of a message m with a secret key sk, the decryption algorithm [ct,sk]q outputs an approximate value m+e of the original message with a small error e. The main advantage of HEAAN comes from the rescaling procedure for managing the magnitude of plaintexts.” Cheon § 1. Constraining size and noise, see generally Cheon § 5.3).
As to claim 21 Cheon in view of Pulido-Gaytan discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
Wherein the removal of noise prevents resulting value of polynomials from diverging. (“We now consider the error growth during homomorphic evaluation of a linear transformation….” Cheon § 5.3)
As to claim 23 Cheon in view of Pulido-Gaytan discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
wherein the homomorphic conjugation operation changes sign of the imaginary part of the encrypted data to generate the homomorphic conjugation data, and (Cheon § 3.2)
wherein the removal of the imaginary part is implemented by performing an arithmetic operation based on the encrypted data and the homomorphic conjugation data. (“we can extract the imaginary (sine) part by conjugation operation (i.e., 2 sin θ = exp(iθ) − exp(−iθ)), which will be described in the next section.” Cheon § 3.2)
Claim(s) 7, 8, 10, 16, 17, 19, and 22 is/are rejected under 35 U.S.C. 103 as being unpatentable over Cheon et al., “Bootsrapping for Approximate Homomorphic Encryption” (published 2018), in view of Pulido-Gaytan et al., “Privacy-preserving neural networks with Homomorphic encryption: Challenges and opportunities” (published 2021), and Chen et al., “Improved Bootstrapping for Approximate Homomorphic Encryption” (published 2018).
As to claims 7, 16 Cheon in view of Pulido-Gaytan discloses the method/machine/CRM of claims 1, 11, and 20 but does not disclose:
adjust a last layer value of a fast Fourier transform (FFT) coefficient to a predetermined value; and
for the performing of the bootstrapping operation, perform the bootstrapping operation based on the adjusted last layer value of the FFT coefficient.
Chen discloses:
adjust a last layer value of a fast Fourier transform (FFT) coefficient to a predetermined value; and (“In this section, we present a method to improve the performance of linear transformations coeffToSlot and slotToCoeff. 3.1 FFT-like Algorithms for coeffToSlot and slotToCoef The coeffToSlot and slotToCoeff steps in the original bootstrapping algorithm amounts to two linear transforms that are mutually inverses to each other.” Chen § 3.1.) “We can then use a dynamic programming algorithm to compute the optimal strategy as a list of splitting points (a1, . . . , ak). Given this optimal level collapsing strategy, we can generate the collapsed levels by merging the individual layers.” Chen § 3.3)
for the performing of the bootstrapping operation, perform the bootstrapping operation based on the adjusted last layer value of the FFT coefficient. (“The coeffToSlot and slotToCoeff steps in the original bootstrapping algorithm amounts to two linear transforms that are mutually inverses to each other.” Chen § 3.1.)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine Cheon in view of Pulido-Gaytan with Chen as Chen states it improves upon Cheon, (“A follow up work by Cheon et al. (Eurocrypt ’18) described an approximate bootstrapping procedure for the scheme. In this work, we improve upon the previous bootstrapping result.” Chen Abstract)
As to claims 8, 17, Cheon in view of Pulido-Gaytan and Chen discloses the method/machine/CRM of claims 7, 11, and 20 and further discloses:
generate a composite polynomial of approximate polynomials; and (“In the EvalExp step, we take two ciphertexts to be homomorphically evaluated by the approximate polynomial Pr(·) of the complex exponential polynomial.” Cheon § 5.3)
perform the neural network operation on the data (Pulido-Gaytan generally, and Fig. 3) on which the bootstrapping operation is completed, (“This section gives a high level structure of the bootstrapping process for the HEAAN scheme.”Cheon § 5.1) based on the composite polynomial. (Cheon § 5.3).
As to claims 10, 19, Cheon in view of Pulido-Gaytan and Chen discloses the method/machine/CRM of claims 7, 11, and 20 but does not disclose:
wherein the neural network operation comprises a rectified linear unit (ReLU) function operation.
Pulido-Gaytan further discloses:
wherein the neural network operation comprises a rectified linear unit (ReLU) function operation.
(“The selection of an activation function is an essential element in the construction of an effective NN model. It determines the reaction of a neuron to the inputs and information forwarded to the following layers. The most common activation functions are step, sign, sigmoid, ReLU, and Tanh [72].” Pulido-Gaytan § 5.1)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to combine Cheon in view of Pulido-Gaytan with Chen as Chen states it improves upon Cheon, (“A follow up work by Cheon et al. (Eurocrypt ’18) described an approximate bootstrapping procedure for the scheme. In this work, we improve upon the previous bootstrapping result.” Chen Abstract)
As to claim 22 Cheon in view of Pulido-Gaytan discloses the method/machine/CRM of claims 1, 11, and 20 and further discloses:
Wherein the removal of noise prevents resulting value of polynomials from diverging. (“We now consider the error growth during homomorphic evaluation of a linear transformation….” Cheon § 5.3)
Claim(s) 9, 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Cheon et al., “Bootsrapping for Approximate Homomorphic Encryption” (published 2018), in view of Pulido-Gaytan et al., “Privacy-preserving neural networks with Homomorphic encryption: Challenges and opportunities” (published 2021), Chen et al., “Improved Bootstrapping for Approximate Homomorphic Encryption” (published 2018), and Lee et al., “Precise Approximation of Convolutional Neural Networks for Homomorphically Encrypted Data” (published 2021).
As to claims 9, 18, Cheon in view of Pulido-Gaytan and Chen discloses the method/machine/CRM of claims 8, 11, and 20 but does not disclose:
wherein the composite polynomial comprises a composite polynomial of minimax approximate polynomials.
Lee discloses:
wherein the composite polynomial comprises a composite polynomial of minimax approximate polynomials.
(“In order to implement deep learning on word-wise homomorphic encryption (HE), the ReLU and max-pooling functions should be approximated by some polynomials for homomorphic operations…. Thus, we propose a method to approximate the ReLU and max-pooling functions accurately using a composition of minimax approximate polynomials of small degrees.” Lee Abstract)
A person of ordinary skill in the art before the effective filing date of the claimed invention would have modified Cheon in view of Pulido-Gaytan and Chen with Lee by utilizing the Minimax of Lee as an alternate to the relu function of Pulido-Gaytan (Pulido-Gaytan § 5.1). It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify Cheon in view of Pulido-Gaytan and Chen with Lee as Lee explicitly contemplates its use for the purpose of homomorphic machine learning such as that of Pulido-Gaytan, (“In order to implement deep learning on word-wise homomorphic encryption (HE), the ReLU and max-pooling functions should be approximated by some polynomials for homomorphic operations.” Lee Abstract)
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure. See PTO-892, particularly:
Sav et al., US 12,488,141, disclosing privacy preserving distributed training of neural networks.
Shpurov et al., US 12,512,960, discloses homomorphic computations on encrypted data.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to MICHAEL W CHAO whose telephone number is (571)272-5165. The examiner can normally be reached M, W-F 8-5.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Rupal Dharia can be reached at (571) 272-3880. 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.
/MICHAEL W CHAO/ Primary Examiner, Art Unit 2492
Prosecution Insights