DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This non-final office action is in response to the application filed 18 January 2024.
Claims 1-20 are pending. Claims 1, 9, and 15 are independent claims.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 6 September 2024 is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Drawings
The examiner accepts the drawings filed 18 January 2024.
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.
Claims 1-20 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
With respect to independent claims 1, 9, and 15, the claim recites the limitation "the optimized NN (claim 1, line 17; claim 9, line 11; claim 15, line 18)." There is insufficient antecedent basis for this limitation in the claim.
While each of these claims recites “optimizing… parameters of the NN (claim 1, lines 15-16; claim 9, line 10; claim 15, lines 16-17),” the claim fails to provide antecedent basis for “the optimized NN.” For the purpose of examination, the examiner will interpret the claim as though it recites “the optimized parameters of the NN.”
Claims 2-8, 10-14, and 16-20 fail to cure the deficiencies of independent claims 1, 9, and 15, respectively. Claims 2-8, 10-14, and 16-20 are rejected under similar rationale.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Step 1:
According to Step 1 of the two Step analysis, claims 1-20 are directed toward a method (process). Therefore, each of these claims falls within one of the four statutory categories.
Claim 1:
Step 2A, Prong 1:
The claim recites:
calculating, using standard symbolic interval propagation (SSIP), bounds for at least a portion of intermediate nodes in the NN, wherein the calculating action using SSIP is based on the dataset of perturbations (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation) using SSIP, based on the dataset of perturbations, and bounds for at least a portion of the intermediate nodes)
calculating, using reversed symbolic interval propagation (RSIP), at least one of an upper bound or a lower bound on a robust loss function, wherein the calculating action using RSIP is based on each of: (i) the calculated bounds for the at least a portion of intermediate nodes, (ii) the dataset of perturbations, and (iii) the loss function of the NN (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation) using RSIP based on the calculated lower and upper bounds for the at least a portion of intermediate nodes, the dataset of perturbations, and the loss function of the NN)
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
receiving, by a computer system, a neural network (NN), a loss function for the NN, and a dataset of perturbations
returning, by the computer system, the optimized NN
As discussed above, the additional elements of data gathering and data transmission recited at a high level of generality and amounts to extra-solution activity of receiving data i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
The claim recites the additional element:
iteratively, by the computer system
Iteratively performing actions by a computer system recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
The claim recites the additional element:
optimizing, based on the calculated bounds and the at least one of an upper bound or lower bound, parameters of the NN
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
receiving, by a computer system, a neural network (NN), a loss function for the NN, and a dataset of perturbations
returning, by the computer system, the optimized NN
As discussed above, the additional elements of data gathering and data transmission recited at a high level of generality and amounts to extra-solution activity of receiving data i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
The claim recites the additional element:
iteratively, by the computer system
Iteratively performing actions by a computer system recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
The claim recites the additional element:
optimizing, based on the calculated bounds and the at least one of an upper bound or lower bound, parameters of the NN
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 2:
With respect to claim 2, the claim depends upon claim 1. The analysis of claim 1 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites the abstract idea identified with respect to claim 1.
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
wherein the neural network training technique comprises stochastic gradient decent
The training of the neural network is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
wherein the neural network training technique comprises stochastic gradient decent
The training of the neural network is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 3:
With respect to claim 3, the claim depends upon claim 1. The analysis of claim 1 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites the abstract idea identified with respect to claim 1.
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
wherein optimizing the parameters of the NN uses NN training techniques
The training of the neural network is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
wherein optimizing the parameters of the NN uses NN training techniques
The training of the neural network is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 4:
With respect to claim 4, the claim depends upon claim 1. The analysis of claim 1 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites:
wherein the loss function for the NN comprises a user-defined loss function value (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses a judgment to define a loss function using a user-defined loss function value)
Step 2A, Prong 2:
There are no additional elements considered under Step 2A, Prong 2.
Step 2B:
There are no additional elements considered under Step 2B.
Claim 5:
With respect to claim 5, the claim depends upon claim 1. The analysis of claim 1 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites:
wherein the dataset of perturbations comprises at least one of bias field input changes, white noise input changes, brightness input changes, or contrast input changes (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses a judgment to identify perturbations for use in calculating, using RSIP, an upper bound or a lower bound on a robust loss function)
Step 2A, Prong 2:
There are no additional elements considered under Step 2A, Prong 2.
Step 2B:
There are no additional elements considered under Step 2B.
Claim 6:
With respect to claim 6, the claim depends upon claim 1. The analysis of claim 1 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites:
wherein the dataset of perturbations comprises at least one perturbation (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses a judgment to identify perturbations for use in calculating, using RSIP, an upper bound or a lower bound on a robust loss function)
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
at least one perturbation that is encoded into: (i) one or more layers of the NN or (ii) a composition of multiple layers of the NN
The training of the neural network is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
at least one perturbation that is encoded into: (i) one or more layers of the NN or (ii) a composition of multiple layers of the NN
The training of the neural network is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 7:
With respect to claim 7, the claim depends upon claim 1. The analysis of claim 1 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites the abstract idea identified with respect to claim 1.
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
wherein the parameters of the NN comprise at least one of weights or biases of layers in the NN
The training of the neural network is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
wherein the parameters of the NN comprise at least one of weights or biases of layers in the NN
The training of the neural network is recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 8:
With respect to claim 8, the claim depends upon claim 1. The analysis of claim 1 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites:
generating… the dataset of perturbations based on applying one or more rules to each data point in the normal dataset, the one or more rules specifying at least one of perturbations of neighbors of each data point (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses a an evaluation of at least one perturbation of neighbors of each data point in a data set by applying rules to generate a dataset of perturbations)
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
receiving, by the computer system, a normal dataset
As discussed above, the additional elements of data gathering and data transmission recited at a high level of generality and amounts to extra-solution activity of receiving data i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
receiving, by the computer system, a normal dataset
As discussed above, the additional elements of data gathering and data transmission recited at a high level of generality and amounts to extra-solution activity of receiving data i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 9:
Step 2A, Prong 1:
The claim recites:
calculating, using the loss function…, at least an upper bound on a robust loss function…, wherein the calculating action is based on applying a relaxation technique to the transform network (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation) using a relaxation technique and a loss function to calculate lower and upper bounds)
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
receiving, by a computer system, a neural network (NN), a loss function for the NN, and a training dataset
returning, by the computer system, the optimized NN
As discussed above, the additional elements of data gathering and data transmission recited at a high level of generality and amounts to extra-solution activity of receiving data i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
The claim recites the additional element:
iteratively, by the computer system, and for at least one datapoint in the training dataset
Iteratively performing actions by a computer system recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
The claim recites the additional element:
generating a transform network based on augmenting the NN
optimizing, based on the calculated upper bound, parameters of the NN
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
receiving, by a computer system, a neural network (NN), a loss function for the NN, and a training dataset
returning, by the computer system, the optimized NN
As discussed above, the additional elements of data gathering and data transmission recited at a high level of generality and amounts to extra-solution activity of receiving data i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
The claim recites the additional element:
iteratively, by the computer system, and for at least one datapoint in the training dataset
Iteratively performing actions by a computer system recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
The claim recites the additional element:
generating a transform network based on augmenting the NN
optimizing, based on the calculated upper bound, parameters of the NN
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 10:
With respect to claim 10, the claim depends upon claim 9. The analysis of claim 9 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites the abstract idea identified with respect to claim 9.
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
wherein the iteratively generating, calculating, and optimizing operations are performed, by the computer system, until a predefined condition ins satisfied
Iteratively performing actions by a computer system recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
wherein the iteratively generating, calculating, and optimizing operations are performed, by the computer system, until a predefined condition ins satisfied
Iteratively performing actions by a computer system recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 11:
With respect to claim 11, the claim depends upon claim 9. The analysis of claim 9 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites the abstract idea identified with respect to claim 9.
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
wherein generating the transform network based on augmenting the NN comprises encoding one or more brightness transformation into at least one layer of the NN
The use of the neural network and layers of the neural network are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
wherein generating the transform network based on augmenting the NN comprises encoding one or more brightness transformation into at least one layer of the NN
The use of the neural network and layers of the neural network are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 12:
With respect to claim 12, the claim depends upon claim 9. The analysis of claim 9 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites the abstract idea identified with respect to claim 9.
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
wherein generating the transformation network based on augmenting the NN comprises encoding one or more contrast transformations into at least one layer of the NN
The use of the neural network and layers of the neural network are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
wherein generating the transformation network based on augmenting the NN comprises encoding one or more contrast transformations into at least one layer of the NN
The use of the neural network and layers of the neural network are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 13:
With respect to claim 13, the claim depends upon claim 9. The analysis of claim 9 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites the abstract idea identified with respect to claim 9.
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
wherein generating the transform network comprises pre-pending one or more transform layers to the NN, wherein the one or more transform layers encode perturbations for which to robustify the NN against, and wherein the transform network is a composition of the NN and the one or more pre-pended transform layers
The use of the neural network and layers of the neural network are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
wherein generating the transform network comprises pre-pending one or more transform layers to the NN, wherein the one or more transform layers encode perturbations for which to robustify the NN against, and wherein the transform network is a composition of the NN and the one or more pre-pended transform layers
The use of the neural network and layers of the neural network are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 14:
With respect to claim 14, the claim depends upon claim 9. The analysis of claim 9 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites:
wherein the relaxation technique comprises at least one of RSIP, SSIP, SIP, abstract interpretation-based methods, software-defined perimeter (SDP) based methods, partial linearization of a Rectified Linear Units (ReLU) activation functions, or total linearization of the ReLU activation function (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation) using a relaxation technique and a loss function to calculate lower and upper bounds)
Step 2A, Prong 2:
There are no additional elements considered under Step 2A, Prong 2.
Step 2B:
There are no additional elements considered under Step 2B.
Claim 15:
Step 2A, Prong 1:
The claim recites:
calculating, using standard symbolic interval propagation (SSIP), bounds for at least a portion of intermediate nodes in the transform network, wherein calculating action using SSIP is based on the transform network (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation) using SSIP, based on the dataset of perturbations, and bounds for at least a portion of the intermediate nodes)
calculating, using reversed symbolic interval propagation (RSIP), at least one of an upper bound or a lower bound for a robust loss function, wherein the calculating action using RSIP is based on each of: (i) the calculated bounds for the at least a portion of intermediate nodes, (ii) the loss function for the NN, and (iii) the transform network (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation) using RSIP based on the calculated lower and upper bounds for the at least a portion of intermediate nodes, the dataset of perturbations, and the loss function of the NN)
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional elements:
receiving, by a computer system, an NN, a loss function for the NN, and a training dataset
returning, by the computer system, the optimized NN
As discussed above, the additional elements of data gathering and data transmission recited at a high level of generality and amounts to extra-solution activity of receiving data i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
The claim recites the additional element:
iteratively, by the computer system, and for at least one data point in the training dataset
Iteratively performing actions by a computer system recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
The claim recites the additional element:
generating a transform network based on augmenting the NN
optimizing, based on the calculated bounds and the at least one of an upper bound or lower bound, parameters for the NN
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional elements:
receiving, by a computer system, an NN, a loss function for the NN, and a training dataset
returning, by the computer system, the optimized NN
As discussed above, the additional elements of data gathering and data transmission recited at a high level of generality and amounts to extra-solution activity of receiving data i.e. pre-solution activity of gathering data for use in the claimed process. The courts have found limitations directed to obtaining information electronically, recited at a high level of generality, to be well-understood, routine, and conventional (see MPEP 2106.05(d)(II), “receiving or transmitting data over a network”, "electronic record keeping," and "storing and retrieving information in memory").
The claim recites the additional element:
iteratively, by the computer system, and for at least one data point in the training dataset
Iteratively performing actions by a computer system recited at a high-level of generality such that it amounts to no more than mere instructions to apply the exception using a generic computer component (See MPEP 2106.05(f)).
The claim recites the additional element:
generating a transform network based on augmenting the NN
optimizing, based on the calculated bounds and the at least one of an upper bound or lower bound, parameters for the NN
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 16:
With respect to claim 16, the claim recites the limitations substantially similar to those in claim 10. Claim 16 is rejected under similar rationale.
Claim 17:
With respect to claim 17, the claim depends upon claim 15. The analysis of claim 15 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites:
wherein calculating, using RSIP, at least one of an upper bound or a lower bound on a robust loss function comprises back-propagating gradients through the at least one of an upper bound or a lower bound (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation), using RSIP, to identify an upper/lower bound on a robust loss function by back propagating gradients)
Step 2A, Prong 2:
There are no additional elements considered under Step 2A, Prong 2.
Step 2B:
There are no additional elements considered under Step 2B.
Claim 18:
With respect to claim 18, the claim depends upon claim 16. The analysis of claim 16 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites:
wherein calculating, using RSIP, at least one of an upper bound or a lower bound on a robust loss function further comprises back-propagating gradients through the calculated bounds for the at least a portion of intermediate nodes (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation), using RSIP, to identify an upper/lower bound on a robust loss function by back propagating gradients)
Step 2A, Prong 2:
There are no additional elements considered under Step 2A, Prong 2.
Step 2B:
There are no additional elements considered under Step 2B.
Claim 19:
With respect to claim 19, the claim depends upon claim 15. The analysis of claim 15 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites:
calculating a standard training loss function based on output of a forward pass of the NN (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation), to calculate a standard training loss function based on an output from a NN)
calculating a total training loss function based on a weighted summation of the standard training loss and the robust loss function (mental process; As drafted and under its broadest reasonable interpretation, this limitation covers performance of the limitation in the mind (including an observation, evaluation, judgment, opinion) or with the aid of pencil and paper but for the recitation of generic computer components. For example, this limitation encompasses an evaluation (calculation), of a total training loss function based on a weighted summation of the standard training loss and the robust loss function)
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional element:
optimizing the parameters of the NN based on back-propagation using the total training
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional element:
optimizing the parameters of the NN based on back-propagation using the total training
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim 20:
With respect to claim 20, the claim depends upon claim 15. The analysis of claim 15 is incorporated herein by reference.
Step 2A, Prong 1:
The claim recites the abstract idea identified with respect to claim 15.
Step 2A, Prong 2:
The judicial exception is not integrated into a practical application.
The claim recites the additional element:
wherein augmenting the NN comprises pre-pending one or more transform layers to the NN and wherein each of the one or more transform layers is based on the training dataset and at least one perturbation type
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2A, prong two, the additional elements individually or in combination do no integrate the judicial exception into a practical application.
Step 2B:
In accordance with Step 2B, the claim does not include additional elements that are sufficient to amount to significantly more that the judicial exception.
The claim recites the additional element:
wherein augmenting the NN comprises pre-pending one or more transform layers to the NN and wherein each of the one or more transform layers is based on the training dataset and at least one perturbation type
The use of the neural network, and optimizing the neural network, are recited at a high-level of generality and amounts to no more than adding the words “apply it” (or an equivalent) with the judicial exception, or mere instructions to implement an abstract idea on a computer, or merely uses a computer as a tool to perform an abstract idea. Accordingly, this additional element does not integrate the abstract idea into a practical application because it does not impose any meaningful limits on practicing the abstract idea (See MPEP 2106.05(f)).
Accordingly, at Step 2B the additional elements individually or in combination do not amount to significantly more than the judicial exception.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
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.
Claims 1-3, 6-10, 13-16, and 19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Chen et al. (US 12524677, filed 25 January 2021, hereafter Chen) and further in view of Lomuscio et al. (WO 2022/117760, published 9 June 2022, hereafter Lomuscio).
As per independent claim 1, Chen discloses a method for robust training of a neural network (NN), the method comprising:
receiving, by a computer system, an NN, a loss function for the NN, and a dataset of perturbations (column 4, lines 38-59: Here, a machine learning model (deep neural network), a training dataset of perturbations, and a task-specific metrics, such as a loss function, are received)
iteratively, by the computer system (column 6, lines 21-26):
calculating bounds for at least a portion of intermediate nodes in the NN, wherein the calculating action is based on the dataset of perturbations (column 4, line 52- column 16, line 26: Here, bounds, such as MinMax optimization may be calculated to constrain the optimization problem)
calculating at least one of an upper bound or a lower bound on a robust loss function (column 6, lines 16-24: Here, a MinMax algorithm defines both an upper and lower bound), wherein the calculating action is based on each of:
(i) calculated bounds for the at least a portion of intermediate nodes (column 6, lines 9-58)
(ii) the dataset of perturbations (column 6, lines 9-58)
(iii) the loss function of the NN (column 6, lines 9-58: Here, the MinMax optimization problem calculates bounds for the portion of intermediate nodes based upon the perturbations and the loss function)
optimizing, based on the calculated bounds and the at least one of an upper bound or lower bound, parameters of the NN (column 10, lines 27-67: Here, the MinMax algorithm optimizes the parameters of the NN)
returning, by the computer system, the optimized parameters of the NN (column 10, lines 27-67)
Chen fails to specifically disclose:
using standard symbolic interval propagation (SSIP)
using reversed symbolic interval propagation (RSIP)
However, Lomuscio, which is analogous to the claimed invention because it is directed toward verifying the robustness of a neural network, discloses:
using standard symbolic interval propagation (SSIP) (page 21, lines 8-32: Here, a standard Symbolic Interval Propagation is used for establishing lower and upper bounds of pre-activations and output of nodes by progressively calculating lower and upper linear function bounds for the pre-activations and outputs of the nodes in the network starting at a first layer and progressing layer by layer)
using reversed symbolic interval propagation (RSIP) (page 25, lines 6-17: Here, a Reverse Symbolic Interval Propagation is taught for improving precision by obtaining, for the pre-activation of each node in the layer, terms of the pre-activations of the previous layer for every layer after the first)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio with Chen, with a reasonable expectation of success, as it would have allowed for improving precision of linear function bounds by progressively establishing lower and upper bounds using pre-activations of each layer (Lomuscio: page 25, lines 6-17).
As per dependent claim 2, Chen and Lomuscio disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Chen discloses wherein the neural network training techniques comprise projected gradient descent (column 10, lines 27-67). Chen fails to specifically disclose stochastic gradient descent.
However, the examiner takes official notice that it was notoriously well-known in the art at the time of the applicant’s effective filing date to use stochastic gradient descent to train neural networks using small gradients for training, instead of a full gradient, thereby making training computationally cheaper. It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined the well-known with Chen-Lomuscio, with a reasonable expectation of success, as it would have allowed for training using a computationally cheaper manner.
As per dependent claim 3, Chen and Lomuscio disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Chen discloses wherein optimizing the parameters of the NN uses NN training techniques (column 6, lines 21-26: Here, a model is trained and based upon augmenting the training dataset using perturbations, the model is trained again).
As per dependent claim 6, Chen and Lomuscio disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Chen discloses wherein the dataset of perturbations comprises at least one perturbation that is encoded into: (i) one or more layers of the NN (column 7, line 59- column 8, line 45: Here, the perturbations are used on convolutional layer) or (ii) a composition of multiple layers of the NN.
As per dependent claim 7, Chen and Lomuscio disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Lomuscio discloses wherein the parameters of the NN comprise at least one of weights or biases of layers of the NN (page 21, lines 8-32: Here, weights from each layer are obtained and a linear optimization is performed using these weighted values). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio with Chen-Lomuscio, with a reasonable expectation of success, as it would have allowed for obtaining linear function founds for pre-activation nodes in a layer (Lomuscio: page 21, lines 8-32).
As per dependent claim 8, Chen and Lomuscio disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Chen discloses:
receiving, by the computer system, a normal dataset (column 4, lines 38-59)
generating, by the computer system, the dataset of perturbations based on applying one or more rules to each data point in the normal dataset, the one or more rules specifying at least one of perturbations or neighborhoods of each data point (column 4, lines 38-59; column 10, lines 27-67)
As per independent claim 9, Chen discloses a method for robust training of neural networks (NN) with a transformation network, the method comprising:
receiving, by a computer system a NN, a loss function for the NN, and a training dataset (column 4, lines 38-59: Here, a machine learning model (deep neural network), a training dataset of perturbations, and a task-specific metrics, such as a loss function, are received)
iteratively, by the computer system, and for at least one data point in the training dataset (column 6, lines 21-26):
calculating, using a loss function of the NN, at least an upper bound on a robust loss function for the NN, wherein the calculating action is based on applying a technique to the transform network (column 6, lines 9-58: Here, a MinMax algorithm defines both an upper and lower bound. Further, the MinMax optimization problem calculates bounds for the portion of intermediate nodes based upon the perturbations and the loss function)
optimizing, based on the calculated upper bound, parameters of the NN (column 10, lines 27-67: Here, the MinMax algorithm optimizes the parameters of the NN)
returning, by the computer system, the optimized parameters of the NN (column 10, lines 27-67)
Chen fails to specifically disclose:
generating a transform network based on augmenting the NN
applying a relaxation technique
However, Lomuscio, which is analogous to the claimed invention because it is directed toward verifying the robustness of a neural network, discloses:
generating a transform network based on augmenting the NN (Figure 3; page 9, line 32- page 10, line 17: Here, a NN is augmented by decomposing the NN robustness problem into two child problems)
applying a relaxation technique (page 25, lines 6-17: Here, a Reverse Symbolic Interval Propagation is taught for improving precision by obtaining, for the pre-activation of each node in the layer, terms of the pre-activations of the previous layer for every layer after the first)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio with Chen, with a reasonable expectation of success, as it would have allowed for improving precision of linear function bounds by progressively establishing lower and upper bounds using pre-activations of each layer (Lomuscio: page 25, lines 6-17).
As per dependent claim 10, Chen and Lomuscio disclose the limitations similar to those in claim 9, and the same rejection is incorporated herein. Chen discloses wherein the iteratively generating, calculating, and optimizing operations are performed, by the computer system, until a predefined condition is met (column 5, lines 16-24: Here, these operations are performed until convergence (predefined condition)).
As per dependent claim 13, Chen and Lomuscio disclose the limitations similar to those in claim 9, and the same rejection is incorporated herein. Lomuscio discloses:
wherein generating the transform network comprises pre-pending one or more transform layers to the NN (page 25, lines 6-17: Here, a Reverse Symbolic Interval Propagation is taught for improving precision by obtaining, for the pre-activation of each node in the layer, terms of the pre-activations of the previous layer for every layer after the first. These pre-activation of layers are interpreted as pre-pending one or more transform layers)
wherein the one or more transform layers encode perturbations for which to robustify the NN against (page 25, lines 6-17)
wherein the transform network is a composition of the NN and one or more prepended transform layers (page 25, lines 6-17)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio with Chen, with a reasonable expectation of success, as it would have allowed for improving precision of linear function bounds by progressively establishing lower and upper bounds using pre-activations of each layer (Lomuscio: page 25, lines 6-17).
As per dependent claim 14, Chen and Lomuscio disclose the limitations similar to those in claim 9, and the same rejection is incorporated herein. Lomuscio discloses wherein the relaxation technique comprises at least one of RSIP (page 25, lines 6-17), SSIP, SIP, abstract interpretation-based methods, software-defined perimeter (SDP) based methods, partial linearization of a Rectified Linear Units (ReLU) activation function, or total linearization of the ReLU activation function.
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio with Chen, with a reasonable expectation of success, as it would have allowed for improving precision of linear function bounds by progressively establishing lower and upper bounds using pre-activations of each layer (Lomuscio: page 25, lines 6-17).
As per independent claim 15, Chen discloses a method for robust training of a neural network (NN), the method comprising:
receiving, by a computer system, an NN, a loss function for the NN, and a dataset (column 4, lines 38-59: Here, a machine learning model (deep neural network), a training dataset of perturbations, and a task-specific metrics, such as a loss function, are received)
iteratively, by the computer system, and for at least one data point in the training dataset (column 6, lines 21-26):
calculating bounds for at least a portion of intermediate nodes in the NN, wherein the calculating action is based on the network (column 4, line 52- column 16, line 26: Here, bounds, such as MinMax optimization may be calculated to constrain the optimization problem)
calculating at least one of an upper bound or a lower bound on a robust loss function (column 6, lines 16-24: Here, a MinMax algorithm defines both an upper and lower bound), wherein the calculating action is based on each of:
(i) calculated bounds for the at least a portion of intermediate nodes (column 6, lines 9-58)
(ii) the loss function of the NN (column 6, lines 9-58: Here, the MinMax optimization problem calculates bounds for the portion of intermediate nodes based upon the perturbations and the loss function)
optimizing, based on the calculated bounds and the at least one of an upper bound or lower bound, parameters of the NN (column 10, lines 27-67: Here, the MinMax algorithm optimizes the parameters of the NN)
returning, by the computer system, the optimized parameters of the NN (column 10, lines 27-67)
Chen fails to specifically disclose:
generating a transform network based on augmenting the NN
using standard symbolic interval propagation (SSIP)
using reversed symbolic interval propagation (RSIP)
the transform network
However, Lomuscio, which is analogous to the claimed invention because it is directed toward verifying the robustness of a neural network, discloses:
generating a transform network based on augmenting the NN (Figure 3; page 9, line 32- page 10, line 17: Here, a NN is augmented by decomposing the NN robustness problem into two child problems)
using standard symbolic interval propagation (SSIP) (page 21, lines 8-32: Here, a standard Symbolic Interval Propagation is used for establishing lower and upper bounds of pre-activations and output of nodes by progressively calculating lower and upper linear function bounds for the pre-activations and outputs of the nodes in the network starting at a first layer and progressing layer by layer)
using reversed symbolic interval propagation (RSIP) (page 25, lines 6-17: Here, a Reverse Symbolic Interval Propagation is taught for improving precision by obtaining, for the pre-activation of each node in the layer, terms of the pre-activations of the previous layer for every layer after the first)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio with Chen, with a reasonable expectation of success, as it would have allowed for improving precision of linear function bounds by progressively establishing lower and upper bounds using pre-activations of each layer (Lomuscio: page 25, lines 6-17).
With respect to claim 16, the claim recites the limitations substantially similar to those in claim 10. Claim 16 is rejected under similar rationale.
As per dependent claim 19, Chen and Lomuscio disclose the limitations similar to those in claim 15, and the same rejection is incorporated herein. Lomuscio discloses:
calculating a standard training loss function based on output of a forward pass of the NN (page 21, lines 8-32: Here, weights from each layer are obtained and a linear optimization is performed using these weighted values)
calculating a total training loss function based on a weighted summation of the standard training loss function and the robust loss function (page 21, lines 8-32: Here, the values of each weighted layer are combined and used as input for the network)
optimizing the parameters of the NN based on back-propagation using the total training loss function (page 25, lines 6-17: Here, a Reverse Symbolic Interval Propagation is taught for improving precision by obtaining, for the pre-activation of each node in the layer, terms of the pre-activations of the previous layer for every layer after the first)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio with Chen, with a reasonable expectation of success, as it would have allowed for improving precision of linear function bounds by progressively establishing lower and upper bounds using pre-activations of each layer (Lomuscio: page 25, lines 6-17).
As per dependent claim 20, Chen and Lomuscio disclose the limitations similar to those in claim 15, and the same rejection is incorporated herein. Lomuscio discloses:
wherein augmenting the NN comprises pre-pending one or more transform layers to the NN (page 25, lines 6-17: Here, a Reverse Symbolic Interval Propagation is taught for improving precision by obtaining, for the pre-activation of each node in the layer, terms of the pre-activations of the previous layer for every layer after the first. These pre-activation of layers are interpreted as pre-pending one or more transform layers)
wherein each of the one or more transform layers is based on the training dataset and at least one perturbation type (page 25, lines 6-17)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio with Chen, with a reasonable expectation of success, as it would have allowed for improving precision of linear function bounds by progressively establishing lower and upper bounds using pre-activations of each layer (Lomuscio: page 25, lines 6-17).
Claim 4 is rejected under 35 U.S.C. 103 as being unpatentable over Chen and Lomuscio and further in view of Jampani et al. (US 2020/0320401, published 8 October 2020, hereafter Jampani).
As per dependent claim 4, Chen and Lomuscio disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Chen fails to specifically disclose wherein the loss function comprises a user-defined loss function value.
However, Jampani, which is analogous to the claimed invention because it is directed toward loss functions for training neural networks, discloses wherein the loss function comprises a user-defined loss function value (paragraph 0038: Here, the loss function is specified by the user). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Jampani with Chen-Lomuscio, with a reasonable expectation of success, as it would have allowed for users to specify the loss function, and thereby specify desirable factors (Jampani: paragraph 0038).
Claim 5 is rejected under 35 U.S.C. 103 as being unpatentable over Chen and Lomuscio and further in view of Daneshjou et al. (US 2023/0107485, filed 3 October 2022, hereafter Daneshjou).
As per dependent claim 5, Chen and Lomuscio disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Chen fails to specifically disclose wherein the dataset of perturbations comprises at least one of bias field input changes, white noise input changes, brightness input changes, or contrast input changes (paragraph 0053: Here, the perturbations cause small changes to brightness and/or contrast). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Daneshjou with Chen-Lomuscio, with a reasonable expectation of success, as it would have allowed for improving model performance by augmenting image brightness and/or contrast (Daneshjou: paragraph 0053).
Claims 11-12 are rejected under 35 U.S.C. 103 as being unpatentable over Chen and Lomuscio and further in view of Lomuscio et al. (US 12547879, 371 date 20 May 2021, hereafter Lomuscio patent).
As per dependent claim 11, Chen and Lomuscio disclose the limitations similar to those in claim 9, and the same rejection is incorporated herein. Chen fails to specifically disclose wherein generating the transform network based on augmenting the NN comprises encoding one or more brightness transformations into at least one layer of the NN.
However, Lomuscio patent, which is analogous to the claimed invention because it is directed toward transforming using layers of the neural network, discloses encoding one or more brightness transformations into at least one layer of the NN (claim 1). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio patent, with Chen-Lomuscio, with a reasonable expectation of success, as it would have allowed for transforming data to improve image classification (Lomuscio patent: claim 1).
As per dependent claim 12, Chen and Lomuscio disclose the limitations similar to those in claim 9, and the same rejection is incorporated herein. Chen fails to specifically disclose wherein generating the transform network based on augmenting the NN comprises encoding one or more contrast transformations into at least one layer of the NN.
However, Lomuscio patent, which is analogous to the claimed invention because it is directed toward transforming using layers of the neural network, discloses encoding one or more contrast transformations into at least one layer of the NN (claim 1). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Lomuscio patent, with Chen-Lomuscio, with a reasonable expectation of success, as it would have allowed for transforming data to improve image classification (Lomuscio patent: claim 1).
Claims 17-18 are rejected under 35 U.S.C. 103 as being unpatentable over Chen and Lomuscio and further in view of Gupta et al. (US 2020/0372300, published 26 November 2020, hereafter Gupta).
As per dependent claim 17, Chen and Lomuscio disclose the limitations similar to those in claim 15, and the same rejection is incorporated herein. Chen fails to specifically disclose wherein calculating, using RSIP, at least one of an upper bound or a lower bound on a robust loss function comprises back-propagating gradients through the at least one of an upper bound or a lower bound.
However, Gupta, which is analogous to the claimed invention because it is directed toward training neural networks, discloses back-propagating gradients through the at least one of an upper bound or a lower bound (paragraphs 0046-0047: Here, gradients are back propagated through convolutional layers). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Gupta with Chen-Lomuscio, with a reasonable expectation of success, as it would have allowed for updating current values of parameters in the neural network (Gupta: paragraph 0046).
As per dependent claim 18, Chen and Lomuscio disclose the limitations similar to those in claim 16, and the same rejection is incorporated herein. Chen fails to specifically disclose wherein calculating, using RSIP, at least one of an upper bound or a lower bound on a robust loss function further comprises back-propagating gradients through the calculated bounds for the at least a portion of intermediate nodes.
However, Gupta, which is analogous to the claimed invention because it is directed toward training neural networks, discloses back-propagating gradients through the calculated bounds for the at least a portion of intermediate nodes (paragraphs 0046-0047: Here, gradients are back propagated through convolutional layers). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Gupta with Chen-Lomuscio, with a reasonable expectation of success, as it would have allowed for updating current values of parameters in the neural network (Gupta: paragraph 0046).
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Andreopoulos et al. (US 11445222): Discloses back propagation using gradient descent methods (column 8, line 53- column 9, line 3)
Any inquiry concerning this communication or earlier communications from the examiner should be directed to KYLE R STORK whose telephone number is (571)272-4130. The examiner can normally be reached 8am - 2pm; 4pm - 6pm.
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, Omar Fernandez Rivas can be reached at 571/272-2589. 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.
/KYLE R STORK/Primary Examiner, Art Unit 2128
Prosecution Insights