DETAILED ACTION
This Office Action is in response to the communication filed on 5/7/2026.
Claims 3, 6 and 12 have been canceled.
Claims 1-2, 4-5, 7-11 and 13-14 are pending.
Claims 1-2, 4-5, 7-11 and 13-14 are rejected.
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and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114.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 05/20/2026 has been entered.
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 with respect to claims 1-2, 4-5, 7-11 and 13-14 filed 1/12/2026 have been fully considered but they are not persuasive.
Applicant argues that Qian in view of Lai does not determine a noise power based on a clipping threshold and the privacy budget. Examiner disagrees.
In response to applicant's argument that the references fail to show certain features of the invention, specifically noise power. However, the references do disclose noise power because noise power can be broadly interpreted as any parameter controlling the magnitude of noise. Lai discloses noise scale which is used to determine a boundary for noise. Therefore, Lai broadly teaches noise power. The claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993).
In response to applicant's argument that the references fail to show certain features of the invention specifically, privacy budget. Privacy budget can be broadly interpreted as anything which limits privacy loss in differential privacy. Lai teaches creating a boundary for total privacy loss, which limits the privacy loss at the boundary created. Therefore, Lai teaches privacy budget. The claims are interpreted in light of the specification, limitations from the specification are not read into the claims. See In re Van Geuns, 988 F.2d 1181, 26 USPQ2d 1057 (Fed. Cir. 1993).
In response to applicant’s argument that there is no teaching, suggestion, or motivation to combine the references, the examiner recognizes that obviousness may be established by combining or modifying the teachings of the prior art to produce the claimed invention where there is some teaching, suggestion, or motivation to do so found either in the references themselves or in the knowledge generally available to one of ordinary skill in the art. See In re Fine, 837 F.2d 1071, 5 USPQ2d 1596 (Fed. Cir. 1988), In re Jones, 958 F.2d 347, 21 USPQ2d 1941 (Fed. Cir. 1992), and KSR International Co. v. Teleflex, Inc., 550 U.S. 398, 82 USPQ2d 1385 (2007). In this case, all references are related to natural langue processing and the references teach or suggest all of the claim features and one of ordinary skill would be motivated to combine in order to increase efficiency and improve security.
Applicant’s arguments with respect to claims 1-2, 4-5, 7-11 and 13-14, specifically that Qian in view of Lai does not teach character level embeddings or perform clipping per-character, has been considered but is/are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Applicant argues that dependent claims are allowable based on respective allowable base claims, however the base claims are not in condition for allowance therefore the dependent claim are not allowable.
Examiner’s Note:
Examiner is interpreting clipping operation to be an obfuscation operation (adding noise) where plaintext data is blurred. The noise power determines how much noise to add (how much clipping to perform) which is set according to a “budget”, a budget is set according to an amount of resources allocated, an amount of resources allocated is includes cost to perform operations (adding noise) (i.e. more important data = more money/resources are allocated to protecting it = more noise can be added). See instant application [0015-0018], [0063], [0065-0073], [0076-0080].
Examiner is interpreting this to mean that noise will be added more frequently on more important data. For example, given sentence “jointly training a natural language processing model”
Less important with lower budget: jointly training (noise) a natural language (noise) processing model
More important with higher budget: jointly (noise) training (noise) a (noise) natural (noise) language (noise) processing (noise) model
Here we have noise being inserted more frequently when the budget allows for it.
Therefore, applicant is performing more secure encryption when more security is needed. Applicant is performing this “more secure encryption” by adding noise more frequently.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-2, 4, 7-11, and 13-14 are rejected under 35 U.S.C. 103 as being unpatentable over Qian (U.S. 20220269816), in view of Lai (US 20230059367), and in further view of Kang (U.S. 20200380964).
Regarding Claims 1, 13 and 14,
Qian discloses: A method for jointly training a natural language processing (NLP) model based on privacy protection, wherein the NLP model comprises an encoding network located at a first party and a processing network located at a second party, and (Qian [0002-0007] Method for obtaining, by an application executing on a processor of an electronic device, user data of a user… sending the transform of the representation of the user data, to a service provider via a network and receiving, from the service provider, via the network, service data based on the transform of the user data)
the method is performed by the first party and comprises: obtaining a target training statement; (Qian [0005-0007] obtaining, by an application executing on a processor of an electronic device, user data of a user; [0092] user data may comprise emergency commands (i.e., statements such as, “Call an ambulance.”))
inputting the target training statement to the encoding network,; and (Qian [0048] According to certain embodiments, at block 307, device 301 generates a representation of the set of user data 305... obtaining a vector representation of user data 305… encoding the user data 305; [Fig. 8, [0090-0092] The operations described with reference to FIG. 8 may be performed on any suitable apparatus on which user data is generated and collected (for example, instances of device 100 in FIG. 1, device 301 in FIG. 3, device 401 in FIG. 4, device 501 in FIG. 5 or device 701 in FIG. 7), and which passes the collected user data to one or more service provider platforms; [0092] As shown in the illustrative example of FIG. 8, at operation 810, the processor generates a representation of the user data obtained at operation 805. According to some embodiments, generating the representation of the data comprises expressing the data in a small vector format; [0092] user data may comprise emergency commands (i.e., statements such as, “Call an ambulance.”) Note: when “Call an ambulance.” Is vectorized, the vector will represent the sentence “Call an ambulance.”
Qian does not explicitly teach: determining noise power for the target training statement based on a preset privacy budget and the clipping threshold; obtaining a target noise that conforms to differential privacy through sampling from a noise distribution determined based on the noise power; and adding the target noise to the sentence representation vector, to obtain a target noise addition representation, wherein the target noise addition representation is sent to the second party for training of the processing network. determining a clipping threshold and
However, in the same field of endeavor, Lai teaches determining noise power for the target training statement based on a preset privacy budget and the clipping threshold; (Lai [0002-0022]; [0043-0063]; [0074]; [0080-0086, Algorithm 1]; [0090-0097]; [0123-0127] teaches determining a noise scale based on a privacy budget and a clipping model clips the one or more gradients determines for a user so that its l2-norm is bounded by a pre-defined gradient clipping bound β)
obtaining a target noise that conforms to differential privacy through sampling from a noise distribution determined based on the noise power; and (Lai [0002-0022]; [0043-0063]; [0074]; [0080-0086, Algorithm 1]; [0090-0097]; [0123-0127] teaches an iterative training process (current iteration round) where data (samples) is used in the current round and a probability including a probability relating to the sensitive user samples; determining a noise scale based on a privacy budget and a clipping model clips the one or more gradients determines for a user so that its l2-norm is bounded by a pre-defined gradient clipping bound β; At each iteration t, the user-entity differential privacy system 106 randomly samples U.sup.t users from U and E.sup.t sensitive entities from E with sampling rates q.sub.u and q.sub.e, respectively (lines 7 and 9)
adding the target noise to the sentence representation vector, to obtain a target noise addition representation, wherein the target noise addition representation is sent to the second party for training of the processing network. (Qian [0003-0007]; sending the transform of the representation of the user data, to a service provider via a network and receiving, from the service provider, via the network, service data based on the transform of the user data. The service data includes a user-specific output based on the transform of the user data; [0057] a hash salt, wherein a set of user-specific nonce data (also referred to as a “salt”) is added to the user data prior to hashing; [Fig. 3-309, 3-311, 0050-0056, Claims 4, 6, 13 and 20] applying local differential privacy comprises adding noise to the locally encoded data)
determining a clipping threshold and (Lai [0015-0022]; [0043]; [0068-0070]; [0090-0097]; [0074 & Algorithm 1, 0080-0094] teaches differential privacy system 106 determines the bounded gradients by clipping the one or more gradients determined for each user… model that clips (e.g., bounds) a value to satisfy a value limit. In particular, in some embodiments, a clipping model utilizes a value that exceeds a value limit to generate a new value within that value limit. For instance, in some implementations, a clipping model clips the one or more gradients determines for a user so that its l2-norm is bounded by a pre-defined gradient clipping bound β)
Qian and Lai are analogous art because they are from the same field of endeavor privacy preserving natural langue processing.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Qian and Lai before him or her, to modify the method of Qian to include the privacy budget, noise scale and noise power of Lai because it will allow for budgets to be considered when inserting noise for protecting data and increase the data that can be protected in differential privacy.
The motivation for doing so would be [“user-entity differential privacy system produces a natural language model that performs a natural language task with an accuracy that approaches comparability with a model that provides no data security”] (Paragraph 0054 and 0097 by Lai)] and [“user-entity differential privacy system produces a natural language model that performs a natural language task with an accuracy that approaches comparability with a model that provides no data security”] (Paragraph 0054 and 0097 by Lai)].
Therefore, it would have been obvious to combine Qian and Lai to obtain the invention as specified in the instant claim.
While Qian in view of Lai teaches device 301 generates a representation of the set of user data 305... obtaining a vector representation of user data 305… encoding the user data 305, Qian in view of Lai does not explicitly teach: obtaining a character representation vector based on an encoding output of the encoding network encoding each character in the target training statement; forming a sentence representation vector based on the clipped character representation vector; clipping the character representation vector of each character
However, in the same field of endeavor Kang teaches: obtaining a character representation vector based on an encoding output of the encoding network encoding each character in the target training statement; (Kang [0065-0083] teaches generating vector representations of data elements, defined segment of the user input data (clipping), which can be characters or groups or characters)
forming a sentence representation vector based on the clipped character representation vector; (Kang [0065-0083] teaches generating vector representations of data elements, defined segment of the user input data (clipping), which can be characters or groups or characters and using those defined segments to generate a representation of a sentence)
clipping the character representation vector of each character (Kang [0065-0083] teaches defined segment of the user input data. For example, a user may provide as text input into a system implementing the method , the query: “what is my balance today”; in such example, each of the terms “what”, “is”, “my”, “balance”, and “today” may be considered data elements of the user input data and generating vector representations of data elements, defined segment of the user input data (clipping), which can be characters or groups or characters)
Qian in view of Lai and Kang are analogous art because they are from the same field of endeavor natural language modeling.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Qian in view of Lai and Kang before him or her, to modify the method of Qian in view of Lai to include clipping per-character representation vectors to generate character level embeddings of Kang because it will improve efficiency by reducing the dimensionality.
The motivation for doing so would be [“Generating a representation of the user data may improve efficiency by reducing the dimensionality”] (Paragraph 0089 by Kang)].
Therefore, it would have been obvious to combine Qian in view of Lai and Kang to obtain the invention as specified in the instant claim.
Claim 13 is rejected under the same rationale as claim 1 above, in addition, Qian further discloses: A computing device (Qian [0002-0010] a processor of an electronic device)
Claim 14 is rejected under the same rationale as claim 1 above, in addition, Qian r discloses: A non-transitory computer-readable storage medium (Qian [0002-0010] a non-transitory, computer-readable medium contains instructions, that when executed by a processor, cause an apparatus)
Regarding Claim 2,
Qian in view of Lai and Kang discloses: The method according to claim 1, wherein the obtaining a local target training statement comprises:
performing sampling from a total local sample set based on a sampling probability p, to obtain a sample subset used for a current iteration round; and (Qian [0048, 0060] According to certain embodiments, at block 307, device 301 generates a representation of the set of user data 305. Depending on embodiments, generating a representation of the user data may comprise… performing t-distributed stochastic neighborhood embedding (t-SNE)
reading the target training statement from the sample subset. (Qian [0080] While some service provider platforms generate service data based on aggregative analyses across a full set of user data, other service provider platforms generate service data based on AI/ML analyses, which perform targeted analyses to obtain a result from only a subset of the available data)
Regarding Claim 4,
Qian in view of Lai and Kang discloses: The method according to claim 1, wherein clipping -the character representation vector comprises:
Qian does not explicitly teach: upon determining that a current norm value of the
However, in the same field of endeavor Lai teaches: upon determining that a current norm value of the (Lai [0015-0022]; [0043]; [0068-0070]; [0080-0086, Algorithm 1]; [0090-0097] teaches In one or more embodiments, a clipping model includes a computer implemented model that clips (e.g., bounds) a value to satisfy a value limit. In particular, in some embodiments, a clipping model utilizes a value that exceeds a value limit to generate a new value within that value limit. For instance, in some implementations, a clipping model clips the one or more gradients determines for a user so that its l2-norm is bounded by a pre-defined gradient clipping bound β)
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 with Lai for similar reasons as cited in claim 1.
While Qian in view of Lai teaches determining when thresholds are crossed and clips (e.g. bounds) values accordingly, Qian in view of Lai does not explicitly teach: character representation vector
However, in the same field of endeavor Kang teaches: character representation vector (Kang [0065-0083] teaches defined segment of the user input data. For example, a user may provide as text input into a system implementing the method , the query: “what is my balance today”; in such example, each of the terms “what”, “is”, “my”, “balance”, and “today” may be considered data elements of the user input data and generating vector representations of data elements, defined segment of the user input data (clipping), which can be characters or groups or characters)
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 with Kang for similar reasons as cited in claim 1.
Regarding Claim 7,
Qian in view of Lai and Kang discloses: The method according to claim 1, wherein determining noise power for the target training statement comprises:
determining, based on the clipping threshold, sensitivity corresponding to the target training statement; and (Qian [0052-0054]; [0071-079]; [0081-0092] teaches determining the sensitivity of the target data according to privacy policies which are based on performing transformations)
Qian does not explicitly teach: determining the noise power for the target training statement based on a preset single-sentence privacy budget and the sensitivity.
However, in the same field of endeavor Lai teaches: determining the noise power for the target training statement based on a preset single-sentence privacy budget and the sensitivity. (Lai [0002-0022]; [0043-0063]; [0074]; [0080-0086, Algorithm 1]; [0090-0097]; [0123-0127] teaches determining a noise scale based on a privacy budget and a clipping model clips the one or more gradients determines for a user so that its l2-norm is bounded by a pre-defined gradient clipping bound β)
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 with Lai for similar reasons as cited in claim 1.
Regarding Claim 8,
Qian in view of Lai and Kang discloses: The method according to claim 1, wherein the noise power for the target training statement is determined further based on target budget information,
Qian does not explicitly teach: wherein the target budget information for a current iteration round t is determined based on a preset total privacy budget used for a total quantity T of iteration rounds;
However, in the same field of endeavor Lai teaches: wherein the target budget information for a current iteration round t is determined based on a preset total privacy budget used for a total quantity T of iteration rounds; (Lai [0052-0056]; [0080-0086, Algorithm 1]; [0090-0097] teaches determining a target for privacy budget and noise scale for the current iteration of the iterative process based on the pre-determined privacy budget)
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 with Lai for similar reasons as cited in claim 1.
Regarding Claim 9,
Qian in view of Lai and Kang discloses: The method according to claim 8,
Qian does not explicitly teach: wherein the target training statement is obtained through sequential reading from a sample subset used for the current iteration round t, and the sample subset is obtained through sampling from a total local sample set based on a sampling probability p; wherein the target budget information for the current iteration round t is determined at least by converting the total privacy budget into a total privacy parameter value in Gaussian differential privacy space; and determining a target privacy parameter value for the current iteration round t in the Gaussian differential privacy space based on the total privacy parameter value, the total quantity T of iteration rounds, and the sampling probability p; and wherein the noise power for the target training statement is determined further based on the target privacy parameter value and a quantity of characters in each training sentence in the sample subset.
However, in the same field of endeavor Lai teaches: wherein the target training statement is obtained through sequential reading from a sample subset used for the current iteration round t, and the sample subset is obtained through sampling from a total local sample set based on a sampling probability p; (Lai [0052-0063]; [0080-0086, Algorithm 1]; [0090-0097] teaches an iterative training process (current iteration round) where data (samples) is used in the current round and a probability including a probability relating to the sensitive user samples)
wherein the target budget information for the current iteration round t is determined at least by converting the total privacy budget into a total privacy parameter value in Gaussian differential privacy space; and (Lai [0002-0022]; [0052-0063]; [0080-0086, Algorithm 1]; [0090-0097]; [0123-0127] teaches a noise scale derived from both user information and sensitive entity information to inject random Gaussian noise into the parameters of the natural language model; At each iteration t, the user-entity differential privacy system 106 randomly samples U.sup.t users from U and E.sup.t sensitive entities from E with sampling rates q.sub.u and q.sub.e, respectively (lines 7 and 9))
determining a target privacy parameter value for the current iteration round t in the Gaussian differential privacy space based on the total privacy parameter value, the total quantity T of iteration rounds, and the sampling probability p; and (Lai [0002-0022]; [0052-0063]; [0080-0086, Algorithm 1]; [0090-0097]; [0123-0127] teaches determining a target for privacy budget and noise scale for the current iteration of the iterative process based on the pre-determined privacy budget)
wherein the noise power for the target training statement is determined further based on the target privacy parameter value and a quantity of characters in each training sentence in the sample subset. (Lai [0002-0022]; [0043-0063]; [0074]; [0080-0086, Algorithm 1]; [0090-0097]; [0123-0127] teaches determining a noise scale based on a privacy budget and a clipping model clips the one or more gradients determines for a user so that its l2-norm is bounded by a pre-defined gradient clipping bound β; At each iteration t, the user-entity differential privacy system 106 randomly samples U.sup.t users from U and E.sup.t sensitive entities from E with sampling rates q.sub.u and q.sub.e, respectively (lines 7 and 9))
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 with Lai for similar reasons as cited in claim 1.
Regarding Claim 10,
Qian in view of Lai and Kang discloses: The method according to claim 9, wherein the determining a target privacy parameter value for the current iteration round t comprises:
Qian does not explicitly teach: deriving the target privacy parameter value based on a first relational expression for calculating the total privacy parameter value in the Gaussian differential privacy space, wherein the first relational expression shows that the total privacy parameter value is directly proportional to the sampling probability p and a square root of the total quantity T of iteration rounds, and depends on a result of a power operation in which a natural exponent e is used as a base and the target privacy parameter value is used as an exponent.
However, in the same field of endeavor Lai teaches: deriving the target privacy parameter value based on a first relational expression for calculating the total privacy parameter value in the Gaussian differential privacy space, wherein the first relational expression shows that the total privacy parameter value is directly proportional to the sampling probability p and a square root of the total quantity T of iteration rounds, and depends on a result of a power operation in which a natural exponent e is used as a base and the target privacy parameter value is used as an exponent. (Lai [0002-0022]; [0043-0063]; [0074]; [0080-0086, Algorithm 1]; [0090-0097]; [0123-0127] teaches determining a noise scale based on a privacy budget and a clipping model clips the one or more gradients determines for a user so that its l2-norm is bounded by a pre-defined gradient clipping bound β and the user-entity differential privacy system generates the one or more parameters using the average gradient and the noise scale (e.g., the Gaussian noise generated from the noise scale)
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 with Lai for similar reasons as cited in claim 1.
Regarding Claim 11,
Qian in view of Lai and Kang discloses: The method according to claim 1, wherein the encoding network is implemented by using one of the following neural networks:
a long short-term memory network (LSTM), a bidirectional LSTM, and a transformer network. (Quai [0059]; [0083] As shown in the explanatory example of FIG. 4, set of representation functionalities may also include a transformer functionality of the type often used in natural language processing (NLP))
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Qian (U.S. 20220269816), in in view of Lai (US 20230059367), in view of Sathaye (U.S. 20150205939) and in further view of Kang (U.S. 20200380964)
Regarding Claim 5,
Qian in view of Lai discloses: The method according to claim 3, wherein the forming the sentence representation vector based on a
vectors of all the characters to form the sentence representation vector. (Qian [0052-0054]; [0071-079]; [0081-0092] teaches transforming a representation of user data to a “sentence” representation, in addition to teaching perturbation to protect user’s privacy; Qian [Equation 1]; 0081-0092] teaches adjusting the “noisiness of the LDP”)
While Qian in view of Lai teaches sentence representation, Qian in view of Lai does not explicitly teach: clipped character representation vector
However, in the same field of endeavor Kang teaches: clipped character representation vector (Kang [0065-0083] teaches defined segment of the user input data. For example, a user may provide as text input into a system implementing the method , the query: “what is my balance today”; in such example, each of the terms “what”, “is”, “my”, “balance”, and “today” may be considered data elements of the user input data and generating vector representations of data elements, defined segment of the user input data (clipping), which can be characters or groups or characters)
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 with Lai for similar reasons as cited in claim 1.
While Qian in view of Lai discloses character representations as part of sentence vectors and Lai discloses clipping characters and injecting noise according to a budget and noise scale Qian in view of does not explicitly disclose: splicing
However, in the same field of endeavor Sathaye discloses: splicing (Sathaye [Fig. 3-5]; [0055] In an implementation, the shielded data 130 may be further shielded by splitting it into N segments of shielded data (432, 434))
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 with Lai for similar reasons as cited in claim 1.
Qian in view of Lai and Kang and Sathaye are analogous art because they are from the same field of endeavor data protection.
Before the effective filing date of the claimed invention, it would have been obvious to one of ordinary skill in the art, having the teachings of Qian in view of Lai and Kang and Sathaye before him or her, to modify the method of Qian in view of Lai and Kang to include the data splitting of Sathaye because it will noise to be inserted between the splits to increase security in untrusted environments.
The motivation for doing so would be [“A communications agent securely coupled to the trusted agent is operable to securely transfer one or more of the N segments of shielded data to one or more storage devices in untrusted computing environments.”] (Sathaye [Abstract]).
Therefore, it would have been obvious to combine Qian in view of Lai and Kang and Sathaye to obtain the invention as specified in the instant claim.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's
disclosure.
GE 9/30/2020 (CN 112199717) teaches a privacy model training method based on a small amount of public data and electronic device, comprising: using several training to obtain N neural network teacher models; respectively inputting a small amount of public data xi into N neural network teacher models; obtaining the statistical voting result of each common data xi for each tag k; adding noise to each statistical voting result; obtaining public data xi satisfying differential privacy principle and corresponding label; judging the neural network by a large amount of random noise vector and a pre-training, optimizing the resistance generating network, and generating a lot of non-marked data; by satisfying the differential privacy principle of public data xi and corresponding label, a lot of non-marked data for pre-training the self-encoder combined training student model to obtain the privacy student model. The invention only needs a small amount of public data to train a private student model, realizing the physical isolation and network isolation of the sensitive data, solving the problem that the precision of the privacy student model is not high.
Stephenson 4/11/2021 (US 20210314140 ) teaches a system for an artificial intelligence synchronized distributed ledger. The system includes a computing device containing a receiving module, the receiving module designed and configured to receive an input from a remote device, parse the input to identify protected and non-protected data contained within the input, transform the protected data into a digitally signed assertion and convert the non-protected into an encrypted datastore. The computing device containing a processing module, the processing module designed and configured to receive the digitally signed assertion from the receiving module, insert the digitally signed assertion into an immutable sequential data structure, receive the encrypted datastore, retrieve at least an input, generate a record utilizing the at least a retrieved input, and perform a first machine-learning process utilizing the at least a retrieved input.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to THOMAS A CARNES whose telephone number is (571)272-4378. The examiner can normally be reached Monday-Friday.
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, Shewaye Gelagay can be reached at (571) 272-4219. 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.
THOMAS A. CARNES
Examiner
Art Unit 2436
/THOMAS A CARNES/Examiner, Art Unit 2436
Prosecution Insights