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 .
Claims 1-5, 7-14, 16-22 are presented for examination based on the amended claims in the application filed on May 11, 2026. Claims 6 and 15 have been cancelled by the applicant.
Claims 1, 3-5, 7, 9-10, 12-14, 16, and 18-22 are rejected under 35 U.S.C. § 102(a)(1) as being anticipated by Guo, Tinghao et al. “Circuit synthesis using generative adversarial networks (GANs).” In AIA A Scitech 2019 Forum, p. 2350. 2019 [herein “Guo”].
Claims 2, 8, 11, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Guo as applied to claims 1 and 10 above, and further in view of US Patent 11,804,050 Milletari, Fausto et al. [herein “Milletari”].
This action is made Final.
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Response to Amendment
The amendment filed May 11, 2026 has been entered. Claims 1-5, 7-14, 16-22 remain pending in the application.
Claim Objections
Claims 1-5, 7-14, 16-18 are objected to because of the following informality: Claim 1, which cites “the physical system” in Ln. 4-5, is improper because there has been no previous recitation of “the physical system”. For the purpose of examination, “the physical system” will be interpreted as “the to-be-designed physical system”. Claim 10, having similar limitations of claim 1, is also objected. Claims 2-5 and 7-9 in addition to claims 11-14 and 16-18 are also objected to for incorporating the deficiency of its dependent claims 1 and 10, respectively.
Appropriate correction is required.
Claim Rejections - 35 U.S.C. § 102
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 the appropriate paragraphs of 35 U.S.C. § 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1, 3-5, 7, 9-10, 12-14, 16, and 18-22 are rejected under 35 U.S.C. § 102(a)(1) as being anticipated by Guo, Tinghao et al. “Circuit synthesis using generative adversarial networks (GANs).” In AIA A Scitech 2019 Forum, p. 2350. 2019 [herein “Guo”].
As per claim 1, Guo teaches “A method for determining system parameters, the method comprising: obtaining, at a design interface, a user input indicating a set of components of a to-be-designed physical system, wherein at least one component of the physical system comprises circuitry”. (Pg. 1, “Here we describe a circuit synthesis problem as one where a set of available electronic components is provided, and from this set components are selected and their connectivities are specified to define a circuit topology” [obtaining a input indicating a set of components of a to-be-designed physical system, wherein at least one component of the physical system comprises circuitry]. “Once a circuit topology is determined, component sizing and other continuous parameters can be optimized to determine the best possible performance of a given circuit topology” [A method for determining system parameters]. Pg. 1, “Circuit synthesis is a challenging design task that can be solved in certain cases by a human designer based on technical understanding, intuition, and knowledge of previous circuit design topologies” [e.g., a user input]. Pg. 10, “One strategy for performance prediction that combines active learning (using a predictive model) with the GAN-based topology generation is depicted in Fig. 7. After circuit generation, NSCs are used to alter out any remaining infeasible topologies. The predictive model is used to approximate the performance of feasible circuits. This strategy may be particularly appropriate in cases where circuit evaluation is computationally expensive, otherwise the performance can be evaluated using the original design optimization problem for each candidate circuit topology” [through the use of a computer, e.g., at a design interface]. Further see Sect. I and III. The examiner has interpreted that providing a set of available electronic components to define a circuit topology and optimizing the component parameters to determine the best possible performance of a given circuit topology which can be solved by a human desired using a strategy for performance prediction for cases where circuit evaluation is computationally expensive as a method for determining system parameters, the method comprising: obtaining, at a design interface, a user input indicating a set of components of a to-be-designed physical system, wherein at least one component of the physical system comprises circuitry.)
Guo also teaches “providing, to an enhanced generative adversarial network (GAN), a predetermined distribution for generating a distribution of feasible parameters needed for designing the physical system comprising the set of components”. (Pg. 5, “A generative adversarial network (GAN) is a class of unsupervised learning models introduced by Goodfellow et al.” [a generative adversarial network (GAN)]. “The GAN is composed of two artificial neural networks, a generator and discriminator. Figure 3 illustrates a basic GAN framework. Let Xr = {x}Ni for i = 1, 2,…, N and xi ∈ Rd denote the real data samples drawn from a probability density distribution Pr. A latent vector z ∈ Rm is pre-defined with a prior density distribution pz(z); here pz(z) is often chosen as a multivariate normal or uniform distribution. The role of the generator G(z; θG) is to produce samples xg with a probability density Pg that approximates Pr.” [a predetermined distribution for generating a distribution of feasible parameters]. Pg. 5, “Given a random sample z, the generator maps the latent space Z to the original data space X” [e.g., providing, to a GAN] . Pg. 11, “Since the improved WGAN demonstrated efficient generation of feasible topologies, we conducted a parametric study regarding two important parameters: the number of latent variables and λ. The purpose of the parametric study was to provide insights about the improved WGAN's capabilities for circuit synthesis” [an enhanced generative adversarial network (GAN)]. Pg. 1, “Here we describe a circuit synthesis problem as one where a set of available electronic components is provided, and from this set components are selected and their connectivities are specified to define a circuit topology. “Once a circuit topology is determined, component sizing and other continuous parameters can be optimized to determine the best possible performance of a given circuit topology” [the physical system comprising the set of components]. Further see Sect. III and VI. The examiner has interpreted using an improved generative adversarial network that using real data samples to produce samples that approximate the probability density distribution of the real data to generate feasible circuit topologies and parameters for components to determine the best possible performance of a circuit topology as providing, to an enhanced generative adversarial network (GAN), a predetermined distribution for generating a distribution of feasible parameters needed for designing the physical system comprising the set of components.)
Guo teaches “wherein the enhanced GAN comprises a hybrid generator and a discriminator, and wherein the hybrid generator comprises a generator of the enhanced GAN and a physics-based model modeling behaviors of the physical system”. (Pg. 5, “A generative adversarial network (GAN) is a class of unsupervised learning models introduced by Goodfellow et al. The GAN is composed of two artificial neural networks, a generator and discriminator” [wherein the enhanced GAN comprises a generator and a discriminator]. Pg. 10, “Predictive modeling can be incorporated into the GAN framework to enable performance prediction for the generated circuits” [a model modeling behaviors of the physical system]. “One strategy for performance prediction that combines active learning (using a predictive model) with the GAN-based topology generation is depicted in Fig. 7” [combining the model with GAN generation, e.g., the hybrid generator comprises a generator of the enhanced GAN and a model]. Furthermore, Pg. 5, “where G(z) is the sample produced by the generator”. Furthermore, Figure 7 shows the predictive model is used in conjunction with the GAN to evaluate the performance of the generator, G(z), e.g. wherein the hybrid generator comprises a generator of the enhanced GAN and a model. Pg. 7, “In the frequency response problem, a complete circuit IS constructed using a transfer function between input I and output O. More desirable circuits will better satisfy the following target frequency response
F
j
ω
=
=
2
π
/
10
ω
where the frequency (Hz) range of interest is 4π ≤ ω ≤ 10 evaluated over 500 logarithmically-spaced points. All circuits containing {R, C} with up to 6 impedance elements were explored previously via the efficient enumeration strategy developed by Herber. A collection of 43,249 unique circuit topologies were identified with unique transfer functions” [physics-based modeling behaviors for all topologies of the physical system]. Pg. 11, “Figures 10a and 10b illustrate how the GAN iteratively works toward feasible generated circuit topologies. A randomly selected M = 10,000 new data samples were obtained using the generator. Each generated circuit topology was verified to be in the complete design data set (sizes 43,249 and 123,156 for the two problems, respectively). We recorded the number of feasible topologies T for different GAN architectures and then computed the metric P. Figure 10a reports the result for the frequency response problem” [verifying physical properties and responses of the system, e.g., a physics-based model]. Further see Sect. III-V. The examiner has interpreted that using a GAN composed of a generator and discriminator and integrating predictive modeling to enable performance prediction for the generated circuits combined with the GAN generation and to verify the generated circuit topology is in the completed design data set to have a unique transfer function between the input and output computed to satisfy a target frequence response as wherein the enhanced GAN comprises a hybrid generator and a discriminator, and wherein the hybrid generator comprises a generator of the enhanced GAN and a physics-based model modeling behaviors of the physical system.)
Guo teaches “mapping, using the generator of the enhanced GAN, input samples from the predetermined distribution to a set of sample parameters”. (Pg. 5, “Given a random sample z, the generator maps the latent space Z to the original data space X”. Further see Sect. III. The examiner has interpreted that the generator that maps samples to original data space as mapping, using the generator of the enhanced GAN, input samples from the predetermined distribution to a set of sample parameters.)
Guo teaches “providing the set of sample parameters to the physics-based model to generate a set of outputs based on physical properties of the set of components within the physical system”. (Pg. 10, “Predictive modeling can be incorporated into the GAN framework to enable performance prediction for the generated circuits. “One strategy for performance prediction that combines active learning (using a predictive model) with the GAN-based topology generation is depicted in Fig. 7” [the model]. Pg. 11, “Figures 10a and 10b illustrate how the GAN iteratively works toward feasible generated circuit topologies. A randomly selected M = 10,000 new data samples were obtained using the generator. Each generated circuit topology was verified to be in the complete design data set (sizes 43,249 and 123,156 for the two problems, respectively). We recorded the number of feasible topologies T for different GAN architectures and then computed the metric P. Figure 10a reports the result for the frequency response problem” [verifying physical properties and responses of the system, e.g., physics-based model]. Pg. 7, “In the frequency response problem, a complete circuit IS constructed using a transfer function between input I and output O. More desirable circuits will better satisfy the following target frequency response
F
j
ω
=
=
2
π
/
10
ω
where the frequency (Hz) range of interest is 4π ≤ ω ≤ 10 evaluated over 500 logarithmically-spaced points. All circuits containing {R, C} with up to 6 impedance elements were explored previously via the efficient enumeration strategy developed by Herber. A collection of 43,249 unique circuit topologies were identified with unique transfer functions” [generate a set of outputs based on physical properties of the set of components within the physical system]. Pg. 14, “Following the procedures in Fig. 7, the numerical experiment was implemented as follows: 1) solved Case 0 problem to obtain a statistical circuit topology sample X; 2) trained an improved WGAN using sample X; 3) applied the active learning strategy to sample X to construct a predictive model; 4) used the generator to output a collection of 10,000 circuit topologies and used NSCs to filter out remaining infeasible circuit topologies; 5) the final predictive model given by the active learning strategy predicts feasible circuit topology performance” [providing the set of sample parameters to the physics-based model to generate a set of outputs based on physical properties of the set of components within the physical system]. Figure 7 shows using inputs to determine the performance, e.g. a set of outputs of the system induced by the set of sample parameters. Further see Sect. III-V. The examiner has interpreted that using inputs of the system to predict the performance of generated circuits using a predictive model that is combined with the generation of the GAN by verifying the generated circuit topology is in the completed design data set to have a unique transfer function between the input and output computed to satisfy a target frequence response as providing the set of sample parameters to the physics-based model to generate a set of outputs based on physical properties of the set of components within the physical system.)
Guo teaches “learning, by the discriminator of the enhanced GAN, to distinguish whether the set of sample parameters follows a response of the physical system within a tolerance range based on the set of outputs generated by the physics-based model.” (Pg. 5, “Discriminator D(x; θD) takes a sample x ∈ X as the input and outputs the probability of x being real. Both D(x; θD) and G(z; θG) can update iteratively in such a way that the generator produces “fake” samples capable of “fooling” the discriminator, while the discriminator aims to distinguish the “fake” samples given by the generator from the real” [learning, by the discriminator of the enhanced GAN, to distinguish whether the set of sample parameters follows a response of the system]. Pg. 10, “Predictive modeling can be incorporated into the GAN framework to enable performance prediction for the generated circuits, a typically costly operation” [the set of outputs generated by the physics-based model]. Pg. 10, “After circuit generation, NSCs [network structure constraints] are used to filter out any remaining infeasible topologies. The predictive model is used to approximate the performance of feasible circuits” [e.g., a response of the system within a tolerance range]. Furthermore, Fig. 7 shows the NSC are input into evaluating G(z), e.g., learning, by the discriminator. Further see Sect. III and V. The examiner has interpreted that having the discriminator aiming to distinguish fake samples from the real sample inputs to predict a performance prediction for generated circuits to determine a feasible circuit topology using network structure constraints and iterative learning to filter out infeasible topologies as learning, by the discriminator of the enhanced GAN, to distinguish whether the set of sample parameters follows a response of the physical system within a tolerance range based on the set of outputs generated by the physics-based model.)
Guo teaches “iteratively updating the hybrid generator and the discriminator of the enhanced GAN until outputs generated by the updated generator correspond to an expected output of the physical system, thereby ensuring feasibility for the set of sample parameters.” (Pg. 5, “The goal is to find parameter values θG for the generator G(z; θG) such that Pg is as close to Pr as possible. Discriminator D(x; θD) takes a sample x ∈ X as the input and outputs the probability of x being real. Both D(x; θD) and G(z; θG) can update iteratively” [iteratively updating the hybrid generator and the discriminator of the enhanced GAN] “in such a way that the generator produces “fake” samples capable of “fooling” the discriminator, while the discriminator aims to distinguish the “fake” samples given by the generator from the real” [until outputs generated by the updated generator]. Pg. 10, “Predictive modeling can be incorporated into the GAN framework to enable performance prediction for the generated circuits, a typically costly operation” [outputs generated by the updated generator]. Pg. 10, “After circuit generation, NSCs [network structure constraints] are used to filter out any remaining infeasible topologies. The predictive model is used to approximate the performance of feasible circuits” [e.g., correspond to expected output of the physical system, thereby ensuring feasibility for the set of sample parameters]. Further see Sect. III and V. The examiner has interpreted that updating the generator and discriminator iteratively while the discriminator aims to distinguish the fake samples given by the generator from the real samples and the generator seeks to produce fake samples that fool the discriminator such that the probability density of the fake is close to the real as possible and uses predictive modeling to enable performance prediction for the generated circuits which are filtered out to determine feasible circuit topologies as iteratively updating the hybrid generator and the discriminator of the enhanced GAN until outputs generated by the updated generator correspond to an expected output of the physical system, thereby ensuring feasibility for the set of sample parameters.)
As per claim 3, Guo teaches “classifying, using the discriminator of the enhanced GAN, whether the set of sample parameters is generated from the predetermined distribution or a data distribution of the physical system”. (Pg. 5, “The GAN is composed of two artificial neural networks, a generator and discriminator” [discriminator]. Pg. 5, “The goal is to find parameter values θG for the generator G(z; θG) such that Pg is as close to Pr as possible. Discriminator D(x; θD) takes a sample x ∈ X as the input and outputs the probability of x being real. Both D(x; θD) and G(z; θG) can update iteratively in such a way that the generator produces “fake” samples capable of “fooling” the discriminator, while the discriminator aims to distinguish the “fake” samples given by the generator from the real” [classifying, using the discriminator of the enhanced GAN, whether the set of sample parameters is generated from the predetermined distribution or a data distribution of the physical system]. Further see Sect. III The examiner has interpreted that having the discriminator of a GAN which aims to distinguish the fake samples given by the generator from the real samples as classifying, using the discriminator of the enhanced GAN, whether the set of sample parameters is generated from the predetermined distribution or a data distribution of the physical system.)
Guo also teaches “wherein iteratively updating the discriminator of the enhanced GAN comprises iteratively updating the discriminator until the discriminator correctly classifies the set of sample parameters”. (Pg. 5, “The goal is to find parameter values θG for the generator G(z; θG) such that Pg is as close to Pr as possible. Discriminator D(x; θD) takes a sample x ∈ X as the input and outputs the probability of x being real. Both D(x; θD) and G(z; θG) can update iteratively” [wherein iteratively updating the discriminator of the enhanced GAN comprises iteratively updating the discriminator iteratively updating the discriminator] “in such a way that the generator produces “fake” samples capable of “fooling” the discriminator, while the discriminator aims to distinguish the “fake” samples given by the generator from the real” [until the discriminator correctly classifies the set of parameter samples]. Further see Sect. III The examiner has interpreted that updating the discriminator iteratively while the discriminator aims to distinguish the fake samples given by the generator from the real samples as wherein iteratively updating the discriminator of the enhanced GAN comprises iteratively updating the discriminator until the discriminator correctly classifies the set of parameter samples.)
As per claim 4, Guo teaches “determining, using the discriminator, a distribution of parameters, wherein samples from the distribution of parameters produce an output from the physics-based model within a predetermined margin of the expected output of the physical system”. (Pg. 5, “Discriminator D(x; θD) takes a sample x ∈ X as the input and outputs the probability of x being real. Both D(x; θD) and G(z; θG) can update iteratively in such a way that the generator produces “fake” samples capable of “fooling” the discriminator, while the discriminator aims to distinguish the “fake” samples given by the generator from the real” [determining, using the discriminator, a distribution of parameters]. Pg. 10, “Predictive modeling can be incorporated into the GAN framework to enable performance prediction for the generated circuits, a typically costly operation” [wherein samples from the distribution of parameters produce an output from the physics-based model]. “One strategy for performance prediction that combines active learning (using a predictive model) with the GAN-based topology generation is depicted in Fig. 7”. Figure 7 demonstrates the determining samples from the distribution of parameters to produce an output. Pg. 10, “After circuit generation, NSCs (network structure constraints) are used to filter out any remaining infeasible topologies” [within a predetermined margin of the expected output of the physical system]. “The predictive model is used to approximate the performance of feasible circuits” [produce an output from the physical model]. Further see Sect. III and V. The examiner has interpreted that combining predictive modeling into the GAN framework to enable performance prediction for the generated circuits through the generation of parameters to create and filter out any remaining infeasible topologies when the discriminator distinguishes fake samples from real samples generated by the generator as determining, using the discriminator, a distribution of parameters, wherein samples from the distribution of parameters produce an output from the physics-based model within a predetermined margin of the expected output of the physical system.)
As per claim 5, Guo teaches “wherein the data distribution of the physical system includes a combination of a distribution of the expected output of the physical system and a noise distribution representing the predetermined margin”. (Pg. 5, “Both D(x; θD) and G(z; θG) can update iteratively in such a way that the generator produces “fake” samples capable of “fooling” the discriminator, while the discriminator aims to distinguish the “fake” samples given by the generator from the real” [the data distribution of the physical system]. Pg. 10, “The predictive model is used to approximate the performance of feasible circuits.” [distribution of the expected output of the system representing the predetermined margin]. Pg. 7, “Here we consider two canonical circuit synthesis problems: 1) a frequency response matching problem, and 2) a low-pass filter realizability problem” [noise baring outputs (e.g., a noise distribution)]. Further see Sect. III-IV. The examiner has interpreted that distinguishing the fake samples given by the generator from the real samples to approximate the feasible circuits performance in frequency response matching and low pass filter examples as wherein the data distribution of the physical system includes a combination of a distribution of the expected output of the physical system and a noise distribution representing the predetermined margin.)
As per claim 7, Guo teaches “wherein iteratively updating the hybrid generator further comprises applying a gradient update scheme to the mapping”. (Pg. 6, “To overcome these issues, a gradient penalty has been introduced. A soft version of a penalty on the gradient is included in the loss function to enforce the K-Lipschitz constraint: (Equation (5))” [iteratively updating the hybrid generator further comprises applying a gradient update scheme to the mapping]. Further see Sect. III. The examiner has interpreted that including a gradient in the loss function to enforce the K-Lipschitz constraint as wherein iteratively updating the hybrid generator further comprises applying a gradient update scheme to the mapping.)
As per claim 9, Guo teaches “determining the set of sample parameters based on a design architecture of the system”. (Pg. 1, “Once a circuit topology is determined, component sizing and other continuous parameters can be optimized to determine the best possible performance of a given circuit topology.” Further see Sect. I. The examiner has interpreted that optimizing the parameters for the best performance of a circuit topology as determining the set of sample parameters based on a design architecture of the system.)
Re Claim 10, it is an articles of manufacture claim, having similar limitations of claim 1. Thus, claim 10 is also rejected under the similar rationale as cited in the rejection of claim 1.
Furthermore, regarding claim 10, Guo teaches “A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for determining system parameters”. (Pg. 10, “One strategy for performance prediction that combines active learning (using a predictive model) with the GAN-based topology generation is depicted in Fig. 7. After circuit generation, NSCs are used to alter out any remaining infeasible topologies. The predictive model is used to approximate the performance of feasible circuits. This strategy may be particularly appropriate in cases where circuit evaluation is computationally expensive, otherwise the performance can be evaluated using the original design optimization problem for each candidate circuit topology” [through the use of a computer, e.g., A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for determining system parameters]. Further see Sect. III and V. The examiner has interpreted that using a strategy for performance prediction for cases where circuit evaluation is computationally expensive as a non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for determining system parameters.)
Re Claim 12, it is an articles of manufacture claim, having similar limitations of claim 3. Thus, claim 12 is also rejected under the similar rationale as cited in the rejection of claim 3.
Re Claim 13, it is an articles of manufacture claim, having similar limitations of claim 4. Thus, claim 13 is also rejected under the similar rationale as cited in the rejection of claim 4.
Re Claim 14, it is an articles of manufacture claim, having similar limitations of claim 5. Thus, claim 14 is also rejected under the similar rationale as cited in the rejection of claim 5.
Re Claim 16, it is an articles of manufacture claim, having similar limitations of claim 7. Thus, claim 16 is also rejected under the similar rationale as cited in the rejection of claim 7.
Re Claim 18, it is an articles of manufacture claim, having similar limitations of claim 9. Thus, claim 18 is also rejected under the similar rationale as cited in the rejection of claim 9.
Re Claim 19, it is a system claim, having similar limitations of claim 1. Thus, claim 19 is also rejected under the similar rationale as cited in the rejection of claim 1.
Furthermore, regarding claim 19, Guo teaches “A computer system, comprising: a storage device; a processor; a non-transitory computer-readable storage medium storing instructions, which when executed by the processor causes the processor to perform a method for determining system parameters”. (Pg. 10, “One strategy for performance prediction that combines active learning (using a predictive model) with the GAN-based topology generation is depicted in Fig. 7. After circuit generation, NSCs are used to alter out any remaining infeasible topologies. The predictive model is used to approximate the performance of feasible circuits. This strategy may be particularly appropriate in cases where circuit evaluation is computationally expensive, otherwise the performance can be evaluated using the original design optimization problem for each candidate circuit topology” [through the use of a computer, e.g., A computer system, comprising: a storage device; a processor; a non-transitory computer-readable storage medium storing instructions, which when executed by the processor causes the processor to perform a method for determining system parameters]. Further see Sect. III and V. The examiner has interpreted that using a strategy for performance prediction for cases where circuit evaluation is computationally expensive as a computer system, comprising: a storage device; a processor; a non-transitory computer-readable storage medium storing instructions, which when executed by the processor causes the processor to perform a method for determining system parameters.)
Re Claim 20, it is a system claim, having similar limitations of claim 3. Thus, claim 20 is also rejected under the similar rationale as cited in the rejection of claim 3.
Re Claim 21, it is an articles of manufacture claim, having similar limitations of claim 4. Thus, claim 21 is also rejected under the similar rationale as cited in the rejection of claim 4.
Re Claim 22, it is an articles of manufacture claim, having similar limitations of claim 5. Thus, claim 22 is also rejected under the similar rationale as cited in the rejection of claim 5.
Claims 2, 8, 11, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Guo as applied to claims 1 and 10 above, and further in view of US Patent 11,804,050 Milletari, Fausto et al. [herein “Milletari”].
As per claim 2, Guo does not specifically teach “determining a set of approximation points for the hybrid generator and generating the set of sample parameters based on the set of approximation points”.
However, in the same field of endeavor namely using machine learning model to generate parameters for optimized designs, Milletari teaches “determining a set of approximation points for the hybrid generator and generating the set of parameter samples based on the set of approximation points”. (Col. 4 Ln. 43-45, “Weight determiner 126 may determine one or more weights for one or more values for one or more parameters” [determining a set of approximation points for the hybrid generator]. Col. 4 Ln. 49-54, “Parameter determiner 128 may determine one or more values of one or more parameters of machine learning model(s) 108 by aggregating one or more values of one or more corresponding parameters from one or more of training nodes 102 using one or more corresponding weights determined by weight determiner 126” [generating the set of sample parameters based on the set of approximation points]. Further see Col. 4. The examiner has interpreted that determining one or more weights for one or more values for one or more parameters and determining one or more values of one or more parameters using one or more corresponding weights as determining a set of approximation points for the hybrid generator and generating the set of sample parameters based on the set of approximation points.)
Therefore, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to add “determining a set of approximation points for the hybrid generator and generating the set of sample parameter based on the set of approximation points” as conceptually seen from the teaching of Milletari into that of Guo because this modification of the generation of and creation of samples from the approximation points for the advantageous purpose of increasing the efficiency and accuracy of the model (Milletari, Col. 1 Ln. 11-22). Further motivation to combine be that Guo and Milletari are analogous art to the current claim are direct to using machine learning model to generate parameters for optimized designs.
As per claim 8, Guo does not specifically teach “determining a set of infeasible parameters from the predetermined distribution” and “excluding the set of infeasible parameters from the mapping”.
However, Milletari teaches “determining a set of parameters from the predetermined distribution”. ( Col. 12 Ln 58-63, “for each training node 102, discrepancy determiner 124 may compute a discrepancy value representative of an amount of discrepancy between updates, contributions and/ or values for one or more parameters from a training node 102 and corresponding consensus from training nodes 102 (e.g., computed by consensus determiner 122)”. Col. 13 Ln. 18-25, “an amount of discrepancy between a contribution, update, and/or value of at least one parameter and consensus may be based at least in part on an amount of dispersion in contributions, updates, and/or values amongst training nodes 102. In at least one embodiment, discrepancy determiner 124 computes an amount of dispersion using an estimator (e.g., a statistical estimator) of scale of discrepancies between updates, values, and/or parameters from training nodes 102” [determining dispersion between parameter values from the training nodes, e.g., determining a set of infeasible parameters from the predetermined distribution]. Further, Col. 14 Ln. 10-18, “training nodes 102 having a larger amount of discrepancy (e.g., a greater distance from a consensus value) are more likely to be outliers than those having a smaller amount of discrepancy. In at least one embodiment, an outlier may indicate mistakes in data preprocessing, bugs, wrong hyper-parameter choices, deliberate adversarial actions, or other characteristics associated with a training node 102, which may negatively influence collaborative training of machine learning model(s) 108” [e.g., infeasible parameters]. Further see Col. 12-14. The examiner has interpreted that computing a discrepancy based on the amount of dispersion from the parameter values in the training nodes as determining a set of parameters from the predetermined distribution.)
Milletari also teaches “excluding the set of infeasible parameters from the mapping”. (Col. 14, “weight determiner 126 maps a value(s) corresponding to one or more parameters from a training node 102 to a weight value using a model, such as a function and/or a machine learning model(s). In at least one embodiment, a value(s) corresponding to one or more parameters from a training node 102 is representative of an amount of discrepancy between a training node 102 (e.g., computed by discrepancy determiner 124) and a consensus of training nodes” [weight corresponds to discrepancy, e.g., infeasibility] Col. 15 Ln 39-45, “parameter determiner 128 may determine one or more values of one or more parameters of machine learning model(s) 108 by aggregating one or more values of one or more corresponding parameters from one or more of training nodes 102 using one or more corresponding weights determined by weight determiner 126” [parameters are determined based on discrepancy]. Col. 26 Ln. 27-31, “review analyzer 1114 may determine to exclude or include a portion of MLM parameter information 1120 (e.g., all parameters from a training node and/or a particular subset thereof) based on such votes (e.g., a vote count)” [excluding parameters based on discrepancy, e.g., excluding the subset of infeasible parameters]. Further, Col. 14 Ln. 10-18, “training nodes 102 having a larger amount of discrepancy (e.g., a greater distance from a consensus value) are more likely to be outliers than those having a smaller amount of discrepancy. In at least one embodiment, an outlier may indicate mistakes in data preprocessing, bugs, wrong hyper-parameter choices, deliberate adversarial actions, or other characteristics associated with a training node 102, which may negatively influence collaborative training of machine learning model(s) 108” [e.g., infeasible parameters]. Further, Col. 16, “values corresponding to an amount of discrepancy greater than a threshold value may be excluded” [excluding the set of infeasible parameters]. Further see Col. 14 and 26. The examiner has interpreted that excluding a portion of the parameter such as the particular subset that includes a discrepancy in the parameter values that leads to outlines in the values to determine a weight based on the discrepancy as excluding the set of infeasible parameters from the mapping.)
Therefore, it would have been obvious to a person having ordinary skill in the art before the effective filing date of the claimed invention to add “determining a set of infeasible parameters from the predetermined distribution” and “excluding the set of infeasible parameters from the mapping” as conceptually seen from the teaching of Milletari into that of Guo because this modification of generating and segregating parameter subsets that based on the feasibility for the advantageous purpose of removing parameters will create outliners in the training data (Milletari, Col. 14 Ln. 9-17). Further motivation to combine be that Guo and Milletari are analogous art to the current claim are direct to using machine learning model to generate parameters for optimized designs.
Re Claim 11, it is an article of manufacture claim, having similar limitations of claim 2. Thus, Claim 11 is also rejected under the similar rationale as cited in the rejection of claim 2.
Re Claim 17, it is machine claim, having similar limitations of claim 8. Thus, Claim 17 is also rejected under the similar rationale as cited in the rejection of claim 8.
Response to Arguments
Applicant's arguments filed on May 11, 2026, have been fully considered, but they are not persuasive.
Applicant argues that reference does not teach each and every limitation in the amend claims because cited reference fails to teach “the physics-based model to generate a set of outputs based on physical properties of the set of components within the physical system” (See Applicant’s response, Pg. 10-11).
MPEP § 2143.03 recites “All words in a claim must be considered in judging the patentability of that claim against the prior art” and “Examiners must consider all claim limitations when determining patentability of an invention over the prior art.”
As provided above in the rejection to claim 1, Guo discloses “the physics-based model to generate a set of outputs based on physical properties of the set of components within the physical system” as predictive model that is combined with the generation of the GAN by verifying the generated circuit topology is in the completed design data set to have a unique transfer function between the input and output computed to satisfy a target frequence response. A target frequency response is determined through a transfer function between the input and output for each of the circuit topologies containing resistors and capacitors, resulting in 43,249 circuit topologies. During the generation steps, the predictive model uses the samples generated by the generator to verify that the generator has generated a circuit topology in the completed design data set of 43,249 circuit topologies. Thus, thus when the generator circuit topology is in the completed design data set, the output given by the computed transfer function for the generator circuit topology is known. Therefore, the predictive model of Guo is a physics-based predictive model. Additional emphasis and continued citation from the primary reference has been added to this mapping in the rejection above to the amended limitation. Furthermore, the full background of invention of Guo as provided in a Ph.D. dissertation of the same author which the primary reference Guo is based on discloses that the predictive model a physics-based (Pg. 11, “One unique element is that data for these methods is obtained through systematic numerical experiments using design optimization tools and predictive physics-based models”, see below for citation).
Therefore, all of the limitations of the amended claim 1 are disclosed in Guo. Therefore, applicant’s arguments are not persuasive and the rejection of claim 1 as anticipated by Guo is maintained.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure.
Guo, Tinghao. "On the use of machine learning with design optimization data for system topology design." PhD diss., University of Illinois at Urbana-Champaign, 2018 teaches using a predictive physics-based model integrated in a GAN for circuit topology generation.
Ma, Wei, Zhaocheng Liu, Zhaxylyk A. Kudyshev, Alexandra Boltasseva, Wenshan Cai, and Yongmin Liu. "Deep learning for the design of photonic structures." Nature photonics 15, no. 2 (2021): 77-90 teaches underlying physics-based model to training a GAN in the design of photonic structures.
Wiecha, Peter R., Arnaud Arbouet, Christian Girard, and Otto L. Muskens. "Deep learning in nano-photonics: inverse design and beyond." Photonics research 9, no. 5 (2021): B182-B200 teaches an physics simulations of an ANN coupled with a GAN for inverse design of photonic devices or nano-structures
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Examiner’s Note: The examiner has cited particular columns and line numbers in the reference that applied to the claims above for the convenience of the applicant. Although the specified citations are representative of the art and are applied to specific limitations within the individual claim, other passages and figures may apply as well. It is respectfully requested from the applicant, to fully consider the references in their entirety as potentially teaching all or part of the claimed invention, as well as the context of the passage as taught by the prior art or disclosed by the examiner. In the case of amending the claimed invention, the applicant is respectfully requested to indicate the portion(s) of the specification which dictate(s) the structure relied on for the proper interpretation and also to verify and ascertain the metes and bound of the claimed invention.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Simeon P Drapeau whose telephone number is (571)-272-1173. The examiner can normally be reached Monday - Friday, 8 a.m. - 5 p.m. ET.
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, Ryan Pitaro can be reached on (571) 272-4071. 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.
/SIMEON P DRAPEAU/Examiner, Art Unit 2188
/RYAN F PITARO/Supervisory Patent Examiner, Art Unit 2188
Prosecution Insights