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 .
In the event the determination of the status of the application as subject to AIA 35 U.S.C 102 and 103 (or as subject to pre-AIA 35 U.S.C 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
Claims 1-20 are subject to review.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 10/20/2023 and 12/18/2025 are being considered by the examiner.
Claim Rejections - 35 USC § 112
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.
Claim 13 is 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.
Claim 13 recites the limitation “the program supports fine-tuning after being combined online with an input network and an output network of a client requesting fine-tuning by being connected with the client in a state in which a first hidden layer of an encoder network, a last hidden layer of a decoder network, and an input layer of the decoder network, which are transmitted from the client requesting fine-tuning, are combined with the encoder network and the decoder network,” and this limitation is unclear and infinite. Specifically, it is unclear how exactly the first hidden layer of an encoder is connected to the last hidden layer of the decoder, and finally connected to the input layer of the decoder in order to perform the computation of data as recited in the subsequent limitations of the claim. Furthermore, it is unclear how the requested fine tuning is achieved in combination with these layers. In addition, it is unclear where exactly the first input vector goes to and its transition through which exact layers and results in a lack of clarity. Therefore, the metes and bounds of this claim are unclear and is found to be indefinite.
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.
Claim 1-20 rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without reciting significantly more.
Step 1 – is the claim directed to a process, machine, manufacture, or composition of matter?
Claims 1-8 are directed to an “apparatus” which describes one of the four statutory categories of patentable subject matter, i.e., a machine.
Claims 9-13 are directed to an “apparatus” which describes one of the four statutory categories of patentable subject matter, i.e., a machine.
Claims 14-20 are directed to a “method” which describes one of the four statutory categories of patentable subject matter, i.e., a process.
Regarding Claim 1:
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 1 recites an abstract idea, substantially as follows:
As per Claim 1, the claim recites limitations of:
“wherein the program performs generating an input vector by passing user data through the secure input network,” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via an input network, which is considered to be a mathematical calculation.
“receiving an output vector, acquired by passing the input vector through a hidden network of the deep-learning model, from the server, and” ,” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via a hidden network, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 1 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“memory in which a secure input network and a secure output network generated in advance based on a deep-learning model provided from a server and at least one program are recorded; and” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
“a processor for executing the program,” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
“transmitting the input vector to the server,” – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
“generating output data by passing the received output vector through the secure output network.” – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 1 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“memory in which a secure input network and a secure output network generated in advance based on a deep-learning model provided from a server and at least one program are recorded; and” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
“a processor for executing the program,” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
“transmitting the input vector to the server,” – the broadest reasonable interpretation of this imitation is found to be merely outputting data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
“generating output data by passing the received output vector through the secure output network.” – the broadest reasonable interpretation of this imitation is found to be merely outputting data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 2
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 2 recites an abstract idea, substantially as follows:
“fine-tuning the deep-learning model using user-private data in a state in which the input network and the output network are changed, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing fine-tuning a model, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 2 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“fixing the hidden network of the deep-learning model provided from the server and changing an input network and an output network thereof to be user-private,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“storing the input network and the output network of the fine-tuned deep-learning model in the memory as the secure input network and the secure output network.” – is merely storing data to a process which is considered to be insignificant extra-solution activity (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 2 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“fixing the hidden network of the deep-learning model provided from the server and changing an input network and an output network thereof to be user-private,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“storing the input network and the output network of the fine-tuned deep-learning model in the memory as the secure input network and the secure output network.” – the broadest reasonable interpretation of this imitation is found to be merely receiving data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 3:
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
No, Claim 3 does not recite an abstract idea.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 3 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“when changing the input network and the output network, the program adds a user-private network to each of the input network and the output network and changes an output unit of the output network to an output unit set by the user.” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 3 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“when changing the input network and the output network, the program adds a user-private network to each of the input network and the output network and changes an output unit of the output network to an output unit set by the user.” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 4:
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
No, Claim 4 does not recite an abstract idea.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 4 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“changing the deep-learning model to be user-private,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“transmitting a hidden network of the deep-learning model changed to be user-private to the server, and” – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
“storing an input network and an output network of the user-private deep-learning model in the memory as the secure input network and the secure output network.” – is merely storing data to a process which is considered to be insignificant extra-solution activity (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 4 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“changing the deep-learning model to be user-private,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“transmitting a hidden network of the deep-learning model changed to be user-private to the server, and” – the broadest reasonable interpretation of this imitation is found to be merely transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
“storing an input network and an output network of the user-private deep-learning model in the memory as the secure input network and the secure output network.” – the broadest reasonable interpretation of this imitation is found to be merely receiving data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 5:
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 5 recites an abstract idea, substantially as follows:
“when changing the deep-learning model, the program changes an order of vector elements of the deep-learning model to an order set by a user and changes an order of weight elements by synchronizing the order of the weight elements with the changed order of the vector elements.” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing changing the order or the permutation of vector elements, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 5 does not include additional limitations that integrate the judicial exception into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 5 does not include additional limitations that amount to significantly more than the judicial exception.
Regarding Claim 6:
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 6 recites an abstract idea, substantially as follows:
“wherein the program scales all of weights of the deep-learning model by multiplying the weights by constant A and adding constant B.” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing scaling weights, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 6 does not include additional limitations that integrate the judicial exception into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 6 does not include additional limitations that amount to significantly more than the judicial exception.
Regarding Claim 7:
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 7 recites an abstract idea, substantially as follows:
“changing an order of vector elements of an input network of the deep-learning model and an order of weight elements of a first hidden layer to an order set by a user by synchronizing the order of the vector elements with the order of the weight elements and reducing a number of nodes at which the input network meets the first hidden layer,” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing changing the order or the permutation of vector elements, which is considered to be a mathematical calculation.
“changing an order of weight elements of a last hidden layer and an order of vector elements of an output network of the deep-learning model to an order set by the user by synchronizing the order of the weight elements with the order of the vector elements and reducing a number of nodes at which the last hidden layer meets the output network,” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing changing the order or the permutation of weight elements, which is considered to be a mathematical calculation.
“fine-tuning the deep-learning model using user data in a state in which the hidden network, excluding the first hidden layer and the last hidden layer, is fixed, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing fine-tuning a deep learning model, which is considered to be a mathematical calculation.
“when receiving the output vector, the output vector is acquired by passing through the hidden network with which the first hidden layer and last hidden layer of the fine-tuned deep-learning model are combined.” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via a hidden network, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 7 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“changing an output unit of the output network to an output unit set by the user and generating a user-private output network in which an order of output nodes is an order set by the user,” – is merely selecting a particular data source or type or data to be manipulated, in this case the output unit, and falls under insignificant extra-solution activity (see MPEP 2106.05(g)).
“storing the input network and the output network of the fine-tuned deep-learning model in the memory as the secure input network and the secure output network, and” – is merely storing data to a process, which is considered to be insignificant extra-solution activity (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 7 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“changing an output unit of the output network to an output unit set by the user and generating a user-private output network in which an order of output nodes is an order set by the user,” – is merely a recitation of an insignificant extra-solution data gathering (see MPEP 2106.05(g)). Further, the insignificant extra-solution data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. iii. Electronic recordkeeping”. [Examiner Note: changing an output unit based on user selection amounts to updating the output which falls under electronic recordkeeping]
“storing the input network and the output network of the fine-tuned deep-learning model in the memory as the secure input network and the secure output network, and” .” – the broadest reasonable interpretation of this imitation is found to be merely receiving data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 8
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 8 recites an abstract idea, substantially as follows:
“the program further performs performing homomorphic encryption on the input vector before transmitting the input vector to the server, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing homomorphic encryption on data, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 8 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“the program uses an input network and an output network of the deep-learning model provided from the server, rather than using the secure input network and the secure output network, and” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“decrypting the received output vector. “ – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 8 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“the program uses an input network and an output network of the deep-learning model provided from the server, rather than using the secure input network and the secure output network, and” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“decrypting the received output vector.” – the broadest reasonable interpretation of this imitation is found to be merely outputting data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 9
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 9 recites an abstract idea, substantially as follows:
“generating an output vector by passing the received input vector through a hidden network of the deep-learning model, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via a hidden network, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 9 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“memory in which at least one program and a deep-learning model are recorded; and” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
“a processor for executing the program,” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
“wherein the program performs receiving a securely processed input vector from a client,” – is merely a recitation of an insignificant extra-solution data gathering (see MPEP 2106.05(g)).
“transmitting the generated output vector to the corresponding client.” – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 9 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“memory in which at least one program and a deep-learning model are recorded; and” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
“a processor for executing the program,” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
“wherein the program performs receiving a securely processed input vector from a client,” – the broadest reasonable interpretation of this imitation is found to be merely receiving data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
“transmitting the generated output vector to the corresponding client.” – the broadest reasonable interpretation of this imitation is found to be merely outputting data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 10
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
No, Claim 10 does not recite an abstract idea.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 10 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“wherein the program is connected with a client requesting fine-tuning, thereby supporting fine-tuning in a state in which an input network and an output network of the client are combined with the hidden network online.” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 10 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“wherein the program is connected with a client requesting fine-tuning, thereby supporting fine-tuning in a state in which an input network and an output network of the client are combined with the hidden network online.” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
Regarding Claim 11
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 11 recites an abstract idea, substantially as follows:
“when generating the output vector, the output vector is generated through a hidden network transmitted by a client that transmits the input vector.” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via a hidden network, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 11 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“the memory stores a hidden network transmitted from each of one or more clients, and” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 11 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“the memory stores a hidden network transmitted from each of one or more clients, and” – is directed to merely applying an abstract idea using a generic computer as a tool (see MPEP 2106.05(f)(2), 2106.04(d)).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 12
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 12 recites an abstract idea, substantially as follows:
“when generating the output vector, the program generates the output vector through the hidden network, including the first hidden layer and the last hidden layer that are fine-tuned by the client that transmits the input vector.” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via a hidden network, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 12 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“the program combines a first hidden layer and a last hidden layer, transmitted from a client requesting fine-tuning, with remaining part of the hidden network and supports fine-tuning in a state in which an input network and an output network of the client requesting fine-tuning are combined with the hidden network online by being connected with the client, and” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 12 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“the program combines a first hidden layer and a last hidden layer, transmitted from a client requesting fine-tuning, with remaining part of the hidden network and supports fine-tuning in a state in which an input network and an output network of the client requesting fine-tuning are combined with the hidden network online by being connected with the client, and” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
Regarding Claim 13
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 13 recites an abstract idea, substantially as follows:
“when generating the output vector, the program performs first generating the output vector through the encoder network and the decoder network combined with the first hidden layer of the encoder network and the last hidden layer of the decoder network that are fine-tuned by the client that transmits a first input vector, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via hidden layers of encoder/decoder networks, which is considered to be a mathematical calculation.
“again generating the output vector through the decoder network combined with the input layer of the decoder network and the last hidden layer of the decoder network that are fined-tuned by the client that transmits a second input vector generated from the previously generated output vector, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via hidden layers of encoder/decoder networks, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 13 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“the deep-learning model is a transformer model,” – is merely indicating a field of use or technological environment directed towards the technology of deep learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“the program supports fine-tuning after being combined online with an input network and an output network of a client requesting fine-tuning by being connected with the client in a state in which a first hidden layer of an encoder network, a last hidden layer of a decoder network, and an input layer of the decoder network, which are transmitted from the client requesting fine-tuning, are combined with the encoder network and the decoder network,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“again generating the output vector is repeatedly performed.” – is merely a recitation of repeatedly manipulating data, being insignificant extra-solution activity (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 13 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“the deep-learning model is a transformer model,” – is merely indicating a field of use or technological environment directed towards the technology of deep learning neural networks (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“the program supports fine-tuning after being combined online with an input network and an output network of a client requesting fine-tuning by being connected with the client in a state in which a first hidden layer of an encoder network, a last hidden layer of a decoder network, and an input layer of the decoder network, which are transmitted from the client requesting fine-tuning, are combined with the encoder network and the decoder network,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“again generating the output vector is repeatedly performed.” – the broadest reasonable interpretation of this imitation is found to be performing repetitive calculations, considered WURC under MPEP2106.05(d)(II)(ii).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 14:
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 14 recites an abstract idea, substantially as follows:
“generating an input vector by passing user data through the secure input network;” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via an input network, which is considered to be a mathematical calculation.
“receiving an output vector acquired by passing the input vector through a hidden network of the deep-learning model from the server; and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via a hidden network, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 14 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“transmitting the input vector to the server;” – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
“generating output data by passing the received output vector through the secure output network.” – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 14 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“transmitting the input vector to the server;” – the broadest reasonable interpretation of this imitation is found to be merely outputting data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
“generating output data by passing the received output vector through the secure output network.” – the broadest reasonable interpretation of this imitation is found to be merely outputting data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 15
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 15 recites an abstract idea, substantially as follows:
“fine-tuning the deep-learning model using user private data in a state in which the input network and the output network are changed, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing fine-tuning a model, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 15 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“fixing the hidden network of the deep-learning model provided from the server and changing an input network and an output network of the deep-learning model to be user-private,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“storing the input network and the output network of the fine-tuned deep-learning model in the memory as the secure input network and the secure output network.” – is merely storing data to a process which is considered to be insignificant extra-solution activity (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 15 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“fixing the hidden network of the deep-learning model provided from the server and changing an input network and an output network of the deep-learning model to be user-private,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“storing the input network and the output network of the fine-tuned deep-learning model in the memory as the secure input network and the secure output network.” – the broadest reasonable interpretation of this imitation is found to be merely receiving data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 16
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
No, Claim 16 does not recite an abstract idea.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 16 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“wherein changing the input network and the output network comprises adding a user private network to each of the input network and the output network and changing an output unit of the output network to an output unit set by a user.” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 16 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“wherein changing the input network and the output network comprises adding a user private network to each of the input network and the output network and changing an output unit of the output network to an output unit set by a user.” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 17
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
No, Claim 17 does not recite an abstract idea.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 17 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“changing the deep-learning model to be user-private,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“transmitting a hidden network of the deep-learning model changed to be user-private to the server, and” – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
“storing an input network and an output network of the user-private deep-learning model in memory as the secure input network and the secure output network.” – is merely storing data to a process (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 17 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“changing the deep-learning model to be user-private,” – is merely indicating a field of use or technological environment directed towards the technology of Privacy-Preserving Machine Learning (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“transmitting a hidden network of the deep-learning model changed to be user-private to the server, and” – the broadest reasonable interpretation of this imitation is found to be merely transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
“storing an input network and an output network of the user-private deep-learning model in the memory as the secure input network and the secure output network.” – the broadest reasonable interpretation of this imitation is found to be merely receiving data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 18
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 18 recites an abstract idea, substantially as follows:
“wherein changing the deep-learning model comprises changing an order of vector elements of the deep-learning model to an order set by a user and changing an order of weight elements by synchronizing the order of the weight elements with the changed order of the vector elements.” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing changing the order or the permutation of vector elements, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 18 does not include additional limitations that integrate the judicial exception into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 18 does not include additional limitations that amount to significantly more than the judicial exception.
Regarding Claim 19
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 19 recites an abstract idea, substantially as follows:
“the secure input network and the secure output network are generated in advance by performing changing an order of vector elements of an input network of the deep-learning model and an order of weight elements of a first hidden layer to an order set by a user by synchronizing the order of the vector elements with the order of the weight elements and reducing a number of nodes at which the input network meets the first hidden layer,” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing changing the order or the permutation of vector elements, which is considered to be a mathematical calculation.
“changing an order of weight elements of a last hidden layer and an order of vector elements of an output network of the deep-learning model to an order set by the user by synchronizing the order of the weight elements with the order of the vector elements and reducing a number of nodes at which the last hidden layer meets the output network,” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing changing the order or the permutation of weight elements, which is considered to be a mathematical calculation.
“fine-tuning the deep-learning model using user data in a state in which a hidden network, excluding the first hidden layer and the last hidden layer, is fixed, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing fine-tuning, which is considered to be a mathematical calculation.
“when receiving the output vector, the output vector is acquired by passing through the hidden network with which the first hidden layer and last hidden layer of the fine-tuned deep-learning model are combined.” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing generating a vector through computation via a hidden network, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 19 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“changing an output unit of the output network to an output unit set by the user and generating a user-private output network in which an order of output nodes is an order set by the user,” – is merely selecting a particular data source of type or data to be manipulated, in this case the output unit, and falls under insignificant extra-solution activity (see MPEP 2106.05(g)).
“storing the input network and the output network of the fine-tuned deep-learning model in memory as the secure input network and the secure output network, and” – is merely storing data to a process, which is considered to be insignificant extra-solution activity (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 19 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“changing an output unit of the output network to an output unit set by the user and generating a user-private output network in which an order of output nodes is an order set by the user,” – is merely a recitation of an insignificant extra-solution data gathering (see MPEP 2106.05(g)). Further, the insignificant extra-solution data gathering is also WURC, see MPEP 2106.05(d)(II) “The courts have recognized the following computer functions as well-understood, routine, and conventional functions when they are claimed in a merely generic manner (e.g., at a high level of generality) or as insignificant extra-solution activity. iii. Electronic recordkeeping”. [Examiner Note: changing an output unit based on user selection amount to updating the output which falls under electronic recordkeeping]
“storing the input network and the output network of the fine-tuned deep-learning model in memory as the secure input network and the secure output network, and.” – the broadest reasonable interpretation of this imitation is found to be merely receiving data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
Regarding Claim 20
Steps 2A Prong 1 – is the claim directed to a law of nature, a natural phenomenon (product of nature) or an abstract idea?
Yes, Claim 20 recites an abstract idea, substantially as follows:
“the method further comprising: performing homomorphic encryption on the input vector before transmitting the input vector to the server, and” – is directed to the abstract idea of mathematical concepts (See MPEP 2106.04(a)(2)) as it is describing homomorphic encryption on data, which is considered to be a mathematical calculation.
Step 2A Prong 2: Does the claim recite additional elements that integrate the judicial exception into a practical application?
No, Claim 20 does not include additional limitations that integrate the judicial exception into a practical application. The additional limitation(s):
“an input network and an output network of the deep-learning model provided from the server, rather than the secure input network and the secure output network, are used,” – is merely indicating a field of use or technological environment directed towards the technology of data privacy (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“decrypting the received output vector. “ – is merely a recitation of an insignificant extra-solution data outputting (see MPEP 2106.05(g)).
Therefore, the additional elements, alone or in combination, do not integrate the abstract idea into a practical application (See MPEP 2106.04).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
No, Claim 20 does not include additional limitations that amount to significantly more than the judicial exception. The additional limitation(s):
“the program uses an input network and an output network of the deep-learning model provided from the server, rather than using the secure input network and the secure output network, and” – is merely indicating a field of use or technological environment directed towards the technology of data privacy (see MPEP 2106.06(h)) and fails to integrate the judicial exception.
“decrypting the received output vector.” – the broadest reasonable interpretation of this imitation is found to be merely outputting data, which is analogous to receiving or transmitting data over a network, considered WURC under MPEP2106.05(d)(II)(i).
Therefore, the additional elements, alone or in combination, do not amount to significantly more than the judicial exception (See MPEP 2106.05).
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.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claim(s) 1, 2, 9-12, 14 and 15 are rejected under 35. U.S.C 103 as being unpatentable over NPL reference Poirot et al. “Split Learning for collaborative deep learning in healthcare” hereinafter referred to as Poirot, in view of Bercich et al. (US 20210089682 A1) hereinafter referred to as Bercich.
Regarding Claim 1
Poirot discloses:
“a secure input network and a secure output network generated in advance based on a deep-learning model provided from a server” (Poirot at Pg. 2, Paragraph 6: The paradigm of split learning revolves around splitting up a conventional neural network [Examiner Note: mapped to networks generated in advance based on a deep-learning model] into several elements that can have different accessibility properties. […] The mentioned accessibility properties of these links can either be ‘central’, which means they are hosted on the central server location and accessible as black box to all clients, or ‘local’, in which case they can only be accessed by the proprietary client; Poirot at Pg. 2, Paragraph 7: The chain in this configuration consists of three links. As considered from a forward propagation point of view, the first is called ‘front’, and is local [Examiner Note: mapped to secure input network]. It receives raw input data during forward propagation, and returns an obfuscated intermediate representation.; Poirot at Pg. 2, Paragraph 7: […] to return another intermediate representation to the final link called ‘back’. The back is again local [Examiner Note: mapped to secure output network] and performs the final decoding computation on its input. [Further see Figure 1, which shows the input (front) and output (back) network secured separately from the shared server]
PNG
media_image1.png
242
614
media_image1.png
Greyscale
);
“wherein the program performs generating an input [...] by passing user data through the secure input network,” (Poirot at Pg. 2, Paragraph 7: As considered from a forward propagation point of view, the first is called ‘front’, and is local [Examiner Note: mapped to secure input network]. It receives raw input data [user data] during forward propagation, and returns an obfuscated intermediate representation.)
“transmitting the input [...] to the server,” (Poirot at Pg. 2, Paragraph 7: The second link is called ‘center’ and is centrally hosted [Examiner Note: mapped to the server] . It takes the intermediate representation from the front [Examiner Note: mapped to the input], and performs most of the computation to return another intermediate representation to the final link called ‘back’.)
“receiving an output [...], acquired by passing the input [...] through a hidden network of the deep-learning model, from the server, and” (Poirot at Pg. 2, Paragraph 7: The second link is called ‘center’ [Examiner Note: mapped to hidden network of the deep-learning model] and is centrally hosted [Examiner Note: mapped to the server] . It takes the intermediate representation from the front [Examiner Note: mapped to the input], and performs most of the computation to return another intermediate representation [Examiner Note: mapped to output] to the final link called ‘back’.; Poirot at Pg. 2, Paragraph 9: Typically but not necessarily, the largest part of trainable networks layers can be found in the central link [Examiner Note: the largest part of a trainable networks are the hidden layers, which sit between the input and output and perform majority of computations])
“generating output data by passing the received output [...] through the secure output network.” (Poirot at Pg. 2, Paragraph 7: […] to return another intermediate representation [Examiner Note: the output] to the final link called ‘back’. The back is again local [Examiner Note: mapped to secure output network] and performs the final decoding computation on its input. This local stage is where gradients are computed from the decoded output and labels [Examiner Note: generates output data].)
However, Poirot does not disclose:
“memory in which […] at least program are recorded; and”
“a processor for executing the program,”
“an input vector”
“an output vector”
On the other hand, Bercich discloses:
“memory in which [...] at least program are recorded” (Bercich at [0009]: The method includes providing a computer system with a memory storing specific computer-executable instructions for a neural network.)
“a processor for executing the program,” (Bercich at [0011]: The system further includes one or more processors in communication with the computer-readable memory.)
“an input vector” (Bercich at [0011]: The training includes a plurality of training cycles wherein the training cycles comprise: inputting one of the input vectors at the input nodes; processing said input vector with the neural network to provide an encoded output vector at the output node;)
“an output vector” (Bercich at [0011]: The training includes a plurality of training cycles wherein the training cycles comprise: inputting one of the input vectors at the input nodes; processing said input vector with the neural network to provide an encoded output vector at the output node;)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Poirot with the above teachings of Bercich by using a method of propagating data through secured input and output networks to and from the hidden network of a model hosted in a server, as taught by Poirot, and a system that propagates data as vectors, as taught by Bercich. The modification would have been obvious because one of ordinary skill in the art would be motivated to perform operations without sharing raw data as suggested by Bercich at [0008]: “the inventions disclosed herein may provide data that may be analyzed in such situations without access to the original data that has not been encoded”.
As per Claim 14, this a method claim corresponding to apparatus Claim 1, and is rejected for similar reasons.
Regarding Claim 2
The combination of Poirot and Bercich discloses: “The client apparatus of claim 1, wherein, when generating the secure input network and the secure output network, the program performs” and the limitations are shown in the rejection above.
The combination of Poirot and Bercich further discloses:
“fixing the hidden network of the deep-learning model provided from the server and” (Bercich at [0035]: the one or more processors may fix the weights and biases [Examiner Note: mapped to fixing a network] in one or more of the nodes in the neural network.)
“changing an input network and an output network thereof to be user-private,” (Poirot at Pg. 2, Paragraph 6: The paradigm of split learning revolves around splitting up a conventional neural network into several elements [Examiner Note: mapped to the input and output networks] that can have different accessibility properties. […] The mentioned accessibility properties of these links can either be ‘central’, which means they are hosted on the central server location and accessible as black box to all clients, or ‘local’, in which case they can only be accessed by the proprietary client [Examiner Note: mapped to user-private])
“fine-tuning the deep-learning model using user-private data in a state in which the input network and the output network are changed, and” (Poirot at Pg. 2, Paragraph 8: When training the model one or more mini-batches can iteratively be forwarded through the chain thereby training [Examiner Note: mapped to fine-tuning] both the local [Examiner Note: the input and output networks which have been secured i.e. changed], as well as the central links.) [Examiner Note: training both the local and central links equates to training or fine-tuning the entire model]
“storing the input network and the output network of the fine-tuned deep-learning model in the memory as the secure input network and the secure output network.” (Poirot at Pg. 2, Paragraph 8: When training is switched from one client to another, the state of the local links [Examiner Note: the input network and output network of the fine-tuned deep learning model] from one clients is downloaded and updated at the next.)
As per Claim 15, this a method claim corresponding to apparatus Claim 2, and is rejected for similar reasons.
Regarding Claim 9
Poirot discloses:
“wherein the program performs receiving a securely processed input […] from a client,” (Poirot at Pg. 2, Paragraph 7: As considered from a forward propagation point of view, the first is called ‘front’, and is local [Examiner Note: local is equivalent to being contained within the client]. It receives raw input data during forward propagation, and returns an obfuscated intermediate representation [Examine Note: mapped to securely processed input]. The second link is called ‘center’ and is centrally hosted. It takes the intermediate representation from the front, and […])
“generating an output […] by passing the received input […] through a hidden network of the deep-learning model, and” (Poirot at Pg. 2, Paragraph 7: The second link is called ‘center’ [Examiner Note: mapped to hidden network] and is centrally hosted. It takes the intermediate representation from the front [Examiner Note: mapped to the input], and performs most of the computation [Examiner Note: mapped to passing the input through the hidden network] to return another intermediate representation [Examiner Note: mapped to the output] to the final link called ‘back’.)
“transmitting the generated output […] to the corresponding client.” (Poirot at Pg. 2, Paragraph 7: It takes the intermediate representation from the front, and performs most of the computation to return another intermediate representation [Examiner Note: mapped to the generated output] to the final link called ‘back’. The back is again local and […]; [Further see Figure 1, which shows propagation of data between clients and a shared server]
PNG
media_image1.png
242
614
media_image1.png
Greyscale
);
However, Poirot does not disclose:
“memory in which at least one program and a deep-learning model are recorded; and”
“a processor for executing the program,”
“an input vector”
“an output vector”
On the other hand, Bercich discloses:
“memory in which at least one program and a deep-learning model are recorded; and” (Bercich at [0009]: The method includes providing a computer system with a memory storing specific computer-executable instructions for a neural network.)
“a processor for executing the program,” (Bercich at [0011]: The system further includes one or more processors in communication with the computer-readable memory.)
“an input vector” (Bercich at [0011]: The training includes a plurality of training cycles wherein the training cycles comprise: inputting one of the input vectors at the input nodes; processing said input vector with the neural network to provide an encoded output vector at the output node;)
“an output vector” (Bercich at [0011]: The training includes a plurality of training cycles wherein the training cycles comprise: inputting one of the input vectors at the input nodes; processing said input vector with the neural network to provide an encoded output vector at the output node;)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Poirot with the above teachings of Bercich by using a method of propagating data through secured input and output networks to and from the hidden network of a model hosted in a server, as taught by Poirot, and a system that propagates data as vectors, as taught by Bercich. The modification would have been obvious because one of ordinary skill in the art would be motivated to perform operations without sharing raw data as suggested by Bercich at [0008]: “the inventions disclosed herein may provide data that may be analyzed in such situations without access to the original data that has not been encoded”.
Regarding Claim 10
The combination of Poirot and Bercich discloses: “The server apparatus of claim 9,” and the limitations are shown in the rejection above.
The combination of Poirot and Bercich further discloses:
“wherein the program is connected with a client requesting fine-tuning, thereby supporting fine-tuning in a state in which an input network and an output network of the client are combined with the hidden network online.” (Poirot at Pg. 2, Paragraph 8: When training the model one or more mini-batches can iteratively be forwarded through the chain thereby training [Examiner Note: mapped to fine-tuning] both the local [Examiner Note: mapped to the client], as well as the central links.)
Poirot at Pg. 2, Paragraph 9: Typically but not necessarily, the largest part of trainable networks layers can be found in the central link [Examiner Note: the largest part of a trainable networks are the hidden layers, which sit between the input and output and perform majority of computations]
Regarding Claim 11
The combination of Poirot and Bercich discloses: “The server apparatus of claim 9, wherein:” and the limitations are shown in the rejection above.
The combination of Poirot and Bercich further discloses:
“when generating the output vector, the output vector is generated through a hidden network transmitted by a client that transmits the input vector. (Poirot at Pg. 2, Paragraph 6: The paradigm of split learning revolves around splitting up a conventional neural network into several elements that can have different accessibility properties. […] The mentioned accessibility properties of these links
can either be ‘central’, […], or ‘‘local’, in which case they can only be accessed by the proprietary client; Poirot at Pg. 2, Paragraph 7: The second link is called ‘center’ [Examiner Note: mapped to hidden network] and is centrally hosted. It takes the intermediate representation from the front [Examiner Note: mapped to the input], and performs most of the computation [Examiner Note: mapped to generating] to return another intermediate representation [Examiner Note: mapped to the output] to the final link called ‘back’.; Poirot at Pg. 2, Paragraph 9: Typically but not necessarily, the largest part of trainable networks layers can be found in the central link; [Examiner Note: the largest part of a trainable networks are the hidden layers, which sit between the input and output and perform majority of computations]).
Regarding Claim 12
The combination of Poirot and Bercich discloses: “The server apparatus of claim 9, wherein:” and the limitations can be found in the rejection above.
The combination of Poirot and Bercich further discloses:
“the program combines a first hidden layer and a last hidden layer, transmitted from a client requesting fine-tuning, with remaining part of the hidden network and supports fine-tuning in a state in which an input network and an output network of the client requesting fine-tuning are combined with the hidden network online by being connected with the client, and” (Poirot at Pg. 2, Paragraph 6: The paradigm of split learning revolves around splitting up a conventional neural network into several elements that can have different accessibility properties. […] The mentioned accessibility properties of these links can either be ‘central’, […], or ‘‘local’, in which case they can only be accessed by the proprietary client; Poirot at Pg. 2, Paragraph 8: When training the model one or more mini-batches can iteratively be forwarded through the chain thereby training [Examiner Note: mapped to fine-tuning] both the local [Examiner Note: mapped to the client], as well as the central links.; Poirot at Pg. 2, Paragraph 9: Typically but not necessarily, the largest part of trainable networks layers can be found in the central link [Examiner Note: the largest part of a trainable networks are the hidden layers, which sit between the input and output and perform majority of computations])
“when generating the output vector, the program generates the output vector through the hidden network, including the first hidden layer and the last hidden layer that are fine-tuned by the client that transmits the input vector.” (Poirot at Pg. 2, Paragraph 8: When training the model one or more mini-batches can iteratively be forwarded through the chain thereby training [Examiner Note: mapped to fine-tuning] both the local [Examiner Note: mapped to the client], as well as the central links.; Poirot at Pg. 2, Paragraph 7: The second link is called ‘center’ [Examiner Note: mapped to hidden network] and is centrally hosted. It takes the intermediate representation from the front [Examiner Note: mapped to the input], and performs most of the computation [Examiner Note: mapped to generating] to return another intermediate representation [Examiner Note: mapped to the output] to the final link called ‘back’.) [Examiner Note: hidden networks, mapped here as the ‘central’ link, consist of multiple hidden layers]
Claim(s) 3 and 16 are rejected under 35. U.S.C 103 as being unpatentable over Poirot in view of Bercich further in view of Hsieh et al. (US 20180144465 A1) hereinafter referred to as Hsieh.
Regarding Claim 3
The combination of Poirot and Bercich discloses: “The client apparatus of claim 2,” and the limitations are shown in the rejection above.
The combination of Poirot and Bercich further discloses:
“when changing the input network and the output network, the program adds a user-private network to each of the input network and the output network and “ (Poirot at Pg. 2, Paragraph 6: The paradigm of split learning revolves around splitting up a conventional neural network into several elements [Examiner Note: mapped to the input and output networks] that can have different accessibility properties. […] The mentioned accessibility properties of these links can either be ‘central’, which means they are hosted on the central server location and accessible as black box to all clients, or ‘local’, in which case they can only be accessed by the proprietary client;) [Examiner Note: adding a network equates to changing the overall architecture, and splitting up a conventional neural network is equivalent to changing the overall architecture]
However, the combination of Poirot and Bercich does not disclose:
“changes an output unit of the output network to an output unit set by the user”
On the other hand, Hsieh discloses:
“changes an output unit of the output network to an output unit set by the user” (Hsieh at [165]: The images are obtained according to the settings provided by the acquisition engine 1430. The settings can be automatically configured [Examiner Note: mapped to changes an output unit] at the imaging device 1410 by the acquisition engine 1430 and/or manually input/overridden by the user 1404 (e.g., a clinician, radiologist, technician, etc.))
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Poirot and Bercich with the above teachings of Hsieh by using a method that adds user-private networks to a model as taught by Poirot and Bercich, and a system that allows a user to configure changes to the output, as taught by Hsieh. The modification would have been obvious because one of ordinary skill in the art would be motivated to improve deep learning systems and methods as suggested by Hsieh at [0001]: “This disclosure relates generally to improved medical systems and, more particularly, to improved deep learning medical systems and methods for medical procedures”.
As per Claim 16, this a method claim corresponding to apparatus Claim 3, and is rejected for similar reasons.
Claim(s) 4 and 17 are rejected under 35. U.S.C 103 as being unpatentable over Poirot in view of Bercich and further in view of Qian et al. (US 20210374605 A1) hereinafter referred to as Qian.
Regarding Claim 4
The combination of Poirot and Bercich discloses: “The client apparatus of claim 1, wherein, when generating the secure input network and the secure output network, the program performs” and the limitations are shown in the rejection above.
The combination of Poirot and Bercich further discloses:
“transmitting a hidden network of the deep-learning model […] to the server, and” (Poirot at Pg. 2, Paragraph 6: The paradigm of split learning revolves around splitting up [Examiner Note: mapped to transmitting] a conventional neural network into several elements that can have different accessibility properties. Poirot at Pg. 2, Paragraph 7: The second link is called ‘center’ [Examiner Note: mapped to hidden network] and is centrally hosted [Examiner Note: mapped to the server].) Poirot at Pg. 2, Paragraph 9: Typically but not necessarily, the largest part of trainable networks layers can be found in the central link; [Examiner Note: the largest part of a trainable networks are the hidden layers, which sit between the input and output and perform majority of computations])
“storing an input network and an output network of the […] deep-learning model in the memory as the secure input network and the secure output network.” (Poirot at Pg. 2, Paragraph 8: When training is switched from one client to another, the state of the local links [Examiner Note: the input network and output network of the deep learning model] from one clients is downloaded and updated at the next.)
However, Poirot in view of Bercich does not disclose:
“changing the deep-learning model to be user-private,”
On the other hand, Qian discloses:
“changing the deep-learning model to be user-private,” (Qian at [0052] These models may then go through local private perturbation [Examiner Note: mapped to changing] 640, which may lead to private model A 650, private model B 655, and private model Z 660.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Poirot and Bercich with the above teachings of Qian by using a method that partitions a user-private model between memory and a server as taught by Poirot and Bercich, and a method of model privatization as taught by Qian. The modification would have been obvious because one of ordinary skill in the art would be motivated to improve performance and privacy as suggested by Qian at [0027]: “embodiments disclosed herein can not only achieve a better performance but also provide a strong privacy guarantee at the same time”.
As per Claim 17, this a method claim corresponding to apparatus Claim 4, and is rejected for similar reasons.
Claim(s) 5 and 18 are rejected under 35. U.S.C 103 as being unpatentable over Poirot in view of Bercich in view of Qian and further in view of Hsieh.
Regarding Claim 5
The combination of Poirot, Bercich, and Qian discloses: “The client apparatus of claim 4, wherein, when changing the deep-learning model,” and the limitations are shown in the rejection above.
The combination of Poirot, Bercich, and Qian further discloses:
“changes an order of weight elements” (Qian at [0054]: The splitting may separate the weights associated with each local model while the shuffling may change them to random order.)
However, the combination of Poirot, Bercich, and Qian does not disclose:
“the program changes an order of vector elements of the deep-learning model to an order set by a user and […] synchronizing the order of the weight elements with the changed order of the vector elements.”
On the other hand, Hsieh discloses:
“the program changes an order of vector elements of the deep-learning model to an order set by a user and […] synchronizing the order of the weight elements with the changed order of the vector elements.” (Hsieh at [0289]: For example, the data is used to form and/or modify nodes [Examiner Note: mapped to vector elements of the deep learning model] and/or connections between nodes in a deep learning network; Hsieh at [0230]: A classifier 2350 (e.g., a softmax classifier, etc.) associates weights with nodes [Examiner Note: mapped to synchronizing weight elements with vector elements] representing features of interest).
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Poirot, Bercich, and Qian with the above teachings of Hsieh by using a method that permutes weight elements as taught by the combination of Poirot, Bercich, and Qian, and a system that synchronizes the changes between weight and vector elements of a deep learning model, as taught by Hsieh. The modification would have been obvious because one of ordinary skill in the art would be motivated to improve deep learning systems and methods as suggested by Hsieh at [0001]: “This disclosure relates generally to improved medical systems and, more particularly, to improved deep learning medical systems and methods for medical procedures”.
As per Claim 18, this a method claim corresponding to apparatus Claim 5, and is rejected for similar reasons.
Claim(s) 6 is rejected under 35. U.S.C 103 as being unpatentable over Poirot in view of Bercich in view of Qian in view of Hsieh and further in view of Cheng et al. (US 20230023520 A1) herein after referred to as Cheng.
Regarding Claim 6
The combination of Poirot, Bercich, Qian, and Hsieh discloses: “The client apparatus of claim 5,” and the limitations are shown in the rejection above.
However, the combination of Poirot, Bercich, Qian, and Hsieh does not disclose:
“wherein the program scales all of weights of the deep-learning model by multiplying the weights by constant A and adding constant B.”
On the other hand, Cheng discloses:
“wherein the program scales all of weights of the deep-learning model by multiplying the weights by constant A and adding constant B.” (Cheng at [0012]: During the training, randomized iterative affine cipher-based encryption may be performed on a model parameter [Examiner Note: weights of the deep learning model] of the interaction layer; Cheng at [0155] A process of encrypting the affine cipher is operated based on affine transformation (which is also known as linear transformation [Examiner Note: mapped to scaling]). For example, a plaintext m is inputted, a ciphertext c is outputted, c=(a*m+b) [Examiner Note: mapped to multiplying by constant A and adding constant B] %n, or c=(a*(m+b))%n, where an encryption key is (a, b, n). *represents a multiplication, % represents a modulus operation, n being a modulus.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Poirot, Bercich, Qian and Hsieh by using a method that permutes weight and vector elements of deep learning model as taught by the combination of Poirot, Bercich, Qian and Hsieh, and a method that scales the weights of a deep learning model, as taught by Cheng. The modification would have been obvious because one of ordinary skill in the art would be motivated to improve model training efficiency as suggested by Cheng at [0005]: “thereby improving computing efficiency of a training process”.
Claim(s) 8 and 20 are rejected under 35. U.S.C 103 as being unpatentable over Poirot in view of Bercich and further in view of Gomez et al. (US 11575500 B2) hereinafter referred to as Gomez.
Regarding Claim 8
The combination of Poirot and Bercich discloses: “The client apparatus of claim 1, wherein:” and the limitations are shown in the rejection above.
The combination of Poirot and Bercich further discloses:
“transmitting the input vector to the server, and” (Poirot at Pg. 2, Paragraph 7: The second link is called ‘center’ and is centrally hosted [Examiner Note: mapped to the server]. It takes the intermediate representation [Examiner Note: mapped to input] from the front, and performs most of the computation to return another intermediate representation to the final link called ‘back’.)
“decrypting the received output vector.” (Poirot at Pg. 2, Paragraph 7: The back is again local and performs the final decoding computation on its input.)
However, Poirot in view of Bercich does not disclose:
“the program uses an input network and an output network of the deep-learning model provided from the server, rather than using the secure input network and the secure output network, and”
“the program further performs performing homomorphic encryption on the input vector before”
On the other hand, Gomez discloses:
“the program uses an input network and an output network of the deep-learning model provided from the server, rather than using the secure input network and the secure output network, and” (Gomez at Col 6, Lines 45-49: The example DCNN architecture in FIG. 3 shows an input layer 302. (e.g., representing the image, text, or other data to be input into the model), a number of hidden layers 304, and an output layer 308 (e.g., representing the output, such as objects recognized in the image))
“performing homomorphic encryption on the input vector” (Gomez at Col 11, Lines 52-56: The vectorized input data is then encrypted homomorphically, using the public key associated with the encrypted NN model to generate the encrypted input data as the first layer of the NN encrypted homomorphically)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Poirot and Bercich with the above teachings of Gomez by using a method that transmits data to be decrypted, as taught by Poirot and Bercich, and a method of performing homomorphic encryption on data, as taught by Gomez. The modification would have been obvious because one of ordinary skill in the art would be motivated to ensure data and model confidentiality as suggested by Gomez at Col 2, Lines 58-60: “Confidentiality of the trained NN, input data and inference results are therefore guaranteed.”).
As per Claim 20, this a method claim corresponding to apparatus Claim 8, and is rejected for similar reasons.
Allowable Subject Matter
Claims 7 and 19 would be allowable if rewritten to overcome the rejection(s) under 35 U.S.C. 101, set forth in this Office action and to include all of the limitations of the base claim and any intervening claims.
The closest prior art of record, Hsieh et al. (US 20180144465 A1) discloses modifying nodes and connections between nodes in a deep learning network (Hsieh at [0289]). Additionally, Hsieh recites that nodes of each layer of a neural network are connected, resulting in an input layer having a connection to a hidden layer, a hidden layer having a connection to an output layer, etc., (Hsieh at [0075]). Lastly, Hsieh recites associating weights with nodes (Hsieh at [0230]).
However, the examiner has found that the distinct feature of the applicant’s claimed invention over the prior art is the explicit claiming of the aforementioned limitations as specified in Claims 7 and 19. When viewed individually or in as a combination with other prior art of record, the limitations specified in Claims 7 and 19 are distinct.
Claim 13 would be allowable if rewritten to overcome the rejection(s) under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), 2nd paragraph, and under 35 U.S.C. 101, set forth in this Office action and to include all of the limitations of the base claim and any intervening claims.
The closest prior art of record, Clement et al. (US 20210357187 A1) discloses a transformer model based on encoder and decoder blocks, wherein the blocks consist of layers. (Clement at [0030-0031]). Additionally Clement recites the output of the encoder block is a set of hidden representations which are then propagated to the decoder blocks (Clement at [0063]).
However, the examiner has found that the distinct feature of the applicant’s claimed invention over the prior art is the explicit claiming of the aforementioned limitations as specified in Claim 13. When viewed individually or in as a combination with other prior art of record, the limitations specified in Claim 13 are distinct.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
US 20200082272 A1 – recites Privacy-Preserving Machine Learning methods using an autoencoder to ensure data privacy in remote deep learning services.
US 20240177049 A1 – recites methods of confidential tuning of machine learning models
Any inquiry concerning this communication or earlier communications from the examiner should be directed to SAMIYAH KABIR whose telephone number is (571)270-0722. The examiner can normally be reached Monday-Friday 8am-5pm.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, David Yi can be reached at (571) 270-7519. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/SAMIYAH KABIR/Examiner, Art Unit 2126
/DAVID YI/Supervisory Patent Examiner, Art Unit 2126
Prosecution Insights