DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
Style
In this action unitalicized bold is used for claim language, while italicized bold is used for emphasis.
Amendments Should be Compliant with 37 CFR § 1.121(c)
Claim 17 purports to be a marked-up version of a previous version of claim 17. But claim 17, as originally filed does not appear to be the basis for at least the first half of the most recently filed version of claim 17. Note that non-compliant amendments are generally returned to Applicant. In the future, claims should be marked up as explained in 37 CFR § 1.121(c).
Information Disclosure Statement
All information disclosure statements were submitted prior to the first action and are incompliance with the provisions of 37 C.F.R. § 1.97. Accordingly, they have been considered.
Election/Restrictions
Applicant’s election without traverse of claims 1 and 15-19 in the reply filed on 09/01/2025 is acknowledged. Applicant has elected group 4, directed to “use of a neural network trained as a second forward model in part by manipulating and optimizing inverse model design parameters in a variety of ways.” See Restriction Requirement filed 03/03/2025 (hereafter Restriction) at 6. If amending, Applicant is reminded that groups 1-3 directed to aspects of generating a model, specific ways of simulating designs, and specific ways of using perturbation to improve model’s performance, respectively, were not elected.
Applicant Reply
“The claims may be amended by canceling particular claims, by presenting new claims, or by rewriting particular claims as indicated in 37 CFR 1.121(c). The requirements of 37 CFR 1.111(b) must be complied with by pointing out the specific distinctions believed to render the claims patentable over the references in presenting arguments in support of new claims and amendments. . . . The prompt development of a clear issue requires that the replies of the applicant meet the objections to and rejections of the claims. Applicant should also specifically point out the support for any amendments made to the disclosure. See MPEP § 2163.06. . . . An amendment which does not comply with the provisions of 37 CFR 1.121(b), (c), (d), and (h) may be held not fully responsive. See MPEP § 714.” MPEP § 714.02. Generic statements or listing of numerous paragraphs do not “specifically point out the support for” claim amendments. “With respect to newly added or amended claims, applicant should show support in the original disclosure for the new or amended claims. See, e.g., Hyatt v. Dudas, 492 F.3d 1365, 1370, n.4, 83 USPQ2d 1373, 1376, n.4 (Fed. Cir. 2007) (citing MPEP § 2163.04 which provides that a ‘simple statement such as ‘applicant has not pointed out where the new (or amended) claim is supported, nor does there appear to be a written description of the claim limitation ‘___’ in the application as filed’ may be sufficient where the claim is a new or amended claim, the support for the limitation is not apparent, and applicant has not pointed out where the limitation is supported.’)” MPEP § 2163(II)(A).
Claim Rejections - 35 USC § 112
The following is a quotation of the first paragraph of 35 U.S.C. 112(a):
(a) IN GENERAL.—The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor or joint inventor of carrying out the invention.
The following is a quotation of the first paragraph of pre-AIA 35 U.S.C. 112:
The specification shall contain a written description of the invention, and of the manner and process of making and using it, in such full, clear, concise, and exact terms as to enable any person skilled in the art to which it pertains, or with which it is most nearly connected, to make and use the same, and shall set forth the best mode contemplated by the inventor of carrying out his invention.
Claims 21-22 are rejected under 35 U.S.C. 112(a) or 35 U.S.C. 112 (pre-AIA ), first paragraph, as failing to comply with the written description requirement. The claim(s) contains subject matter which was not described in the specification in such a way as to reasonably convey to one skilled in the relevant art that the inventor or a joint inventor, or for applications subject to pre-AIA 35 U.S.C. 112, the inventor(s), at the time the application was filed, had possession of the claimed invention.
Claims 21 and 22 limit to the cases in which “the first performance parameter is the same as the second performance parameter” and “the first performance parameter is not the same as the second performance parameter,” respectively. Nothing was found in the Specification which explicitly supports either limitation. Applicant does not address implicit support in the Remarks. The Remarks cite paragraphs 22-23 of the Specification filed 10 August 2021. Rem. 10. Nothing in these paragraphs or any other paragraphs in the Specification expressly describe any performance parameters as being the same as or different from one another.
All dependent claims are rejected as containing the limitations of the claims from which they depend.
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, 15-19, and 21-24 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 pre-AIA the applicant regards as the invention.
Generally: separately listed claim elements are construed as distinct components, that all claim terms must be given weight, there is presumed to be a difference in meaning and scope when different words or phrases are used in separate claims, and repeated and consistent descriptions in the specification indicate the proper scope of a claimed term. “[C]laims must ‘conform to the invention as set forth in the remainder of the specification and the terms and phrases used in the claims must find clear support or antecedent basis in the description so that the meaning of the terms in the claims may be ascertainable by reference to the description.’ 37 C.F.R. § 1.75(d)(1).” Phillips v. AWH Corp., 415 F.3d 1303, 1316 (Fed. Cir. 2005) (as cited in MPEP § 2111). Therefore, use of two different terms in the claims that both rely on the description of a single structure in the Specification may render at least one term indefinite because there is no way to determine which term should be construed in view of the description of the single structure.
Claim 1 recites an “artificial intelligence processor[.]” All claim terms must be given weight. Whether or not a processor is an “artificial intelligence processor” is subjective. Further, nothing in the specification indicates any objective meaning of this term. See MPEP §2173.05b. This rejection may be overcome by deleting “artificial intelligence” before “processor” for this term throughout the claims.
Claim 1 recites “generating, by the artificial intelligence processor, one or more designs of a first object to be designed, the generating being by a first machine learned network in response to: a first input including at least one of a fixed value of a first performance parameter defining a design requirement and a maximum or minimum value of a second performance parameter defining a constraint; and a second input identifying a third performance parameter defining a goal parameter to be optimized subject to the at least one of the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter;; simulating, by the artificial intelligence processor, operation of each of the one or more designs by a second machine-learned network, the simulating providing a first optimized value for the third performance parameter for each design[.]” The Specification explains the design requirements, constrains, and goals consistent with all three terms referring to the same thing. See Spec. ¶¶22-23 (“One or more requirements and/or constraints may be identified as a goal. For example, the efficiency (e.g., turbine efficiency) is to be optimized. The goal is a performance parameter for which the design is to operate in a best way possible or a better way than other designs. . . . In the gas turbine example, the user inputs values for performance parameters such as the ambient temperature, turbine shaft power, turbine exhaust temperature, and combustor cooling fraction of intake flow. The goal is to maximize the performance parameter of turbine efficiency, subject to the above-mentioned constraints and/or requirements. Values for other performance parameters as requirements or constraints may be used, such as[.]”) See also Spec. ¶42 (“The goal may be a requirement or constraint as well[.]”) At best, the Specification describes a “goal” of optimizing efficiency subject to a set of parameters that are alternately referred to as “parameters” and “constraints.” But the Specification also indicates that all three terms are interchangeable without providing any objective measure of the scope of each term. The use of different claim terms implies separate claim elements but the scope and meaning of terms in the claims are read in light of the Specification. This results in more than one way of interpreting the claims. Specifically, there are two issues resulting from this description. First, since all three term are used interchangeably in the Specification, it is not clear whether they must be limited to different types of parameters or if all three can refer to the same types of parameters or attributes of the values. Secondly, it is not clear whether or not all three terms can refer to a single parameter because the Specification states that “[o]ne or more requirements and/or constraints may be identified as a goal,” indicating that the terms can “be identified” as one another. Since there is no way to objectively distinguish the three recited terms in view of the Specification, the claim language is indefinite. The claim language above is illustrative, but all claims using these terms are indefinite for the reasons given above. This includes claims 16, 17, and 19, 21, and 22 all of which include these indefinite terms. It is noted that claim 24 is NOT rejected based on this rationale because, despite using these terms, each term is expressly limited to a particular category. Specifically, claim 24 limits the first and second performance parameters to measurable physical parameters (e.g. inputs in a control loop) and the third performance parameter is limited to efficiency.
Claim 1 recites “the one or more design as generated based on a difference between the first optimized value of the third performance parameter and a target optimized value of the third performance parameter[.]” It is not clear whether the “target optimized value” is a target value or another optimized value. Note that a target value generally refers to the value sought by the designer, while the optimized value appears to refer to the best value created by the model. These are different concepts. It is not clear which is being claimed.
Claim 1 recites “determining an error of the one or more design as generated based on a difference between the first optimized value of the third performance parameter and a target optimized value of the third performance parameter and determining an error of the one or more designs as perturbed, the error based on a difference between the second optimized value of the third performance parameter and the target optimized value[.]” It is not clear which “error” provides antecedent basis for “the error.”
Claim 1 recites “a first input including at least one of a fixed value of a first performance parameter defining a design requirement and a maximum or minimum value of a second performance parameter defining a constraint[.]” Claim 21 recites “[t]he method of claim 1 wherein the first performance parameter is the same as the second performance parameter[.]” Generally, separately recited claim elements are interpreted as distinct components so the “design requirement” would be construed differently from the “constraint.” But the language of dependent claim 21 recites the first performance parameter being “the same as” the second performance parameter. Since the material of dependent claim 21 is implicitly within the scope of claim 1, this language implies that the “design requirement” defined by the “first performance parameter” could also be “the same as” the “constraint” defined by the “second performance parameter,” in all claims. Therefore, it is not clear whether the “design requirement” and the “constraint” must be construed as distinct components, consistent with accepted rules of claim construction, or if the two terms can refer to the same claim component as implied by their defining parameters being “the same as” one another in claim 21.
Claim 15 recites instructions causing a processor to “generate, with a first neural network trained as an inverse model by a first machine, design parameters for one or more designs, in response to a specification for each of the one or more designs; optimize, with a second neural network trained as a forward model by the first machine or a second machine, the design parameters based on simulation of the one or more designs using the design parameters generated with the first neural network[.]” This language appears to recites instructions causing a processor to “generate . . . design parameters” “in response to a specification for . . . designs” using “inverse model[.]” The claims then recite machine instructions to “optimize . . . the design parameters” with “a forward model.” There are two issues here. First, inputting a “specification for . . . designs” to an inverse model which outputs “design parameters” seems inconsistent with the plain meaning an “inverse model,” as that term is commonly used in the art. Generally, an inverse model refers to a model which infers structure from function. See e.g. Kim, Deep-learning-based inverse design model for intelligent discovery of organic molecules, description of Fig. 1; Liu, Training Deep Neural Networks for the Inverse Design of Nanophotonic Structures P. 1365 (“We refer to these NNs as forward-modeling networks because they compute EM response from the structure. In contrast, the second type of NNs, as shown in Figure 1(b), take the EM response as the input and directly output the structure. These are referred to as inverse-design networks.”); Pilozzi, Machine learning inverse problem for topological photonics P. 1; and Sanchez, Inverse molecular design using machine learning: Generative models for matter engineering P. 1. Figures 3 and 4 include a long list of design parameters that seem inconsistent with a structure being inferred from function. In fact, most of the parameters listed in figures 3 and 4 refer to functions. This results in a meaning for “inverse modeling” that is inconsistent with the claim language without any other objective meaning found in the Specification. Alternatively, interpreting the claimed “inverse modelling” consistent with its plain meaning in the art, would require that “design parameters” refer to structure. As noted above, the design parameters in the Specification described as functions. These inconsistencies leave no reasonable objective way for one of ordinary skill in the art to interpret the claim language in view of the specification. Second, the “forward model” also outputs “design parameters.” Even assuming the design parameters referred structure, there is no way to reconcile both the forward and the inverse model outputting the same “design parameters.” Based on the foregoing, there is simply no way for one of ordinary skill in the art to determine which categories of structure or function are being input or output from these models, or how “forward” and “inverse” are meant to limit two models recited as having the same outputs.
Claim 19 recites “training, by a machine, a first neural network as an inverse model to generate design parameters for a design in response to first values of one or more performance parameters requirements for the design to generate the design; training, by the machine or another machine, a second neural network as a predictive model to simulate operation of the design based on the generated design parameters to predict second values for the one or more performance parameters from the design[.]” This language is indefinite for reasons analogous to this given in the rejection of claim 15, above.
Claim 15 recites “the design parameters based on simulation of the one or more designs using the design parameters generated with the first neural network[.]” It is not clear whether the “design parameters” are output from a simulation or from the first neural network. Alternatively, it is not clear whether the “simulation” refers to the operation of the first neural network or a separate operation. The claims recite “the design parameters based on simulation” and subsequently “the design parameters generated with the first neural network[.]” This indicates two separate, unrelated techniques for generating the same claim element (i.e. the design parameters). It is not clear which is meant, or if both are meant how two independent claim elements both independently create “the design parameters.” Further, the claim recites “design parameters based on . . . using the design parameters generated with the first neural network[.]” This appears to recite a relationship between the design parameters and themselves, which would be nonsensical. Since there is no clear way to construe the claims that meets all conditions imposed by the claim language, and there are multiple similarly reasonable interpretations, the claim language is indefinite.
Claim 17 recites “wherein the optimization generates additional second possible designs by perturbation of the design parameters of the multiple first possible designs and determines the second value for the third performance parameter for the additional second possible designs[.]” It is not clear whether the determination of “the second value for the third performance parameters for the additional second possible designs” is generated “by perturbation.” While all of the language is within the same wherein clause, the language “and determines . . .” appears to be an independent clause, which implies that “by perturbation” would not necessarily apply to the determination of the second value for the third performance parameter. Alternatively, the language “optimization generates additional second possible designs by perturbation” implies that the subsequently claimed second value of a performance parameter for the “second possible designs” results from the perturbation. Since it is not clear whether or not the “second value for the third performance parameters” is determined as a result of the perturbation, the claim is indefinite.
All dependent claims are rejected as containing the limitations of the claims from which they depend.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1, 15-19, and 21-22 are rejected under 35 U.S.C. 103 as being unpatentable over Tutum (Functional Generative Design: An Evolutionary Approach to 3D-Printing; 2018.) and Sanchez (Inverse molecular design using machine learning: Generative models for matter engineering; 2018.)
1. A method for generative design by an artificial intelligence processor, (Tutum teaches “First, an effective search space is learned through a variational autoencoder (VAE); second, a surrogate model for functional designs is built; and third, a genetic algorithm is used to simultaneously update the hyperparameters of the surrogate and to optimize the designs using the updated surrogate.” Tutum Abstract. Tutum teaches using 8 training sets of 10,000 samples each with resolution of 150X150 pixles (i.e. 150x150x10,000x8, totaling 1.8 billion data points, being trained using equations 1 and 2.) See Tutum P. 3, col. 1 and description of Fig. 5. This training is carried out for multiple iterations. See Tutum P. 2. One of ordinary skill in the art would understand the description of training an autoencoder on this amount of data to reference a computer process using of a standard computer including a processor running executing instructions stored in memory. “We agree with appellant that Figure 1 of Thacker, by itself, does not disclose every limitation in the appealed claims. However, in considering the disclosure of a reference, it is proper to take into account not only specific teachings of the reference but also the inferences which one skilled in the art would reasonably be expected to draw therefrom. In re Shepard, 319 F.2d 194, 50 CCPA 1439 (1963).” In re Preda.) the method comprising: generating, by the artificial intelligence processor, one or more designs of a first object to be designed (P. 2 Col. 2: “standard Genetic Algorithm (GA) is used to tune the hyperparameters of the surrogate. Next, the VAE decoder is applied to convert the list of encoded spring designs into 3D models”; fig. 4: “Initial Sample Set.”) the generating being by a first machine-learned network (P. 2 Col. 2: “VAE.” See also figure 4) in response to a first input including at least one of a fixed value of a first performance parameter defining a design requirement and (e.g. P. 2 Col. 2: “A simple bitmap traversal filtering is employed to discretize pixels and remove unwanted part by tracking those pixels with values greater than a fixed threshold and collectively form a connected component with both bases along neighbors[.]”) a maximum or minimum value of a second performance parameter defining a constraint; (The “constraints” of Tutum include various binary constraints, where the max/minimum value would be yes or no. further, the constraints also include physical parameters, such as stiffness. See Tutum P. 3 section 2.3 (“1) Printability constraints, 2) Performance constraints. Printability constraints have two subcategories: C1) Is the design (i.e., its image form) blurry, or in other words, could VAE decoder produce a visible pattern? C2) If there is any visible pattern, is it connected all the way from the top to the bottom surface? Performance constraints, which satisfy the printability constraints, are further divided into three sub-categories: C3) Is the spring loadable (i.e., not too stiff)? C4) Did the spring work through all 10 experimental repetitions without breaking? C5) Did the car stay on track in at least 5 out of 10 experiments?”) See also Tutum Fig. 8.) and a second input identifying a third performance parameter defining a goal parameter to be optimized subject to the at least one of the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter (Equation 3. Note also, the description indicates that “[o]ne or more requirements and/or constraints may be identified as a goal,” indicating that the terms can “be identified” as one another. Spec. ¶22.) simulating, by the artificial intelligence processor, operation of each of the one or more designs by a second machine-learned network (Fig. 4 “Fitness evaluation” and “Build Surrogate” P. 5 Col. 1: “the next step is to predict a new response value, i.e., a fitness function value, at an unobserved design vector, x*, using the sample data that are used to train the Kriging mode[.]”
Tutum does not expressly teach simulating the operation of one or more designs by a second ML network.
Sanchez teaches “Here, we review methods for achieving inverse design[.] . . . Among these, deep generative models have been applied to numerous classes of materials: rational design of prospective drugs, synthetic routes to organic compounds, and optimization of photovoltaics and redox flow batteries, as well as a variety of other solid-state materials.” Sanchez P. 1, introduction. “Simulation offers one way of probing this space without experimentation.” Sanchez P. 1 col. 2. “In the context of materials, ML techniques are often used for property prediction, seeking to learn a function that maps a molecular material to the property of choice. Deep generative models are a special class of DL methods that seek to model the underlying probability distribution of both structure and property and relate them in a nonlinear way. By exploiting patterns in massive datasets, these models can distill average and salient features that characterize molecules (12, 13). . . . The process (Fig. 1) conventionally involves the following steps: (i) generate a new or improved material concept and simulate its potential suitability; (ii) synthesize the material; (iii) incorporate the material into a device or system; and (iv) characterize and measure the desired properties. This cycle generates feedback to repeat, improve, and refine future cycles of discovery. Each step can take up to several years.” Sanchez P.1 col. 3. “The ultimate aim is to concurrently propose, create, and characterize new materials, with each component transmitting and receiving data simultaneously. This process is called “closing the loop,” and inverse design is a critical facet (12, 15).” Sanchez P.2 col. 1. “Inverse design, as its name suggests, inverts this paradigm by starting with the desired functionality and searching for an ideal molecular structure. Here the input is the functionality and the output is the structure. Functionality need not necessarily map to one unique structure but to a distribution of probable structures.” Sanchez P.2 col.1. To summarize, Sanchez is teaching a system in which a generative model is used to generate a physical structure (inverse design) and explains that this closes the loop in a system where simulation of properties based on structure by neural networks was previously commonplace.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Sanchez because simulation of physical structures created using inverse modeling results in a system that fully automates the task of taking parameters as an input and finding physical structures that are optimized for the parameters while reducing physical experiments or modeling.)
the simulating providing a first optimized value for the third performance parameter for each design (“After tuning the Kriging model parameters, the next step is to predict a new response value, i.e., a fitness function value, at an unobserved design vector, x*, using the sample data that are used to train the Kriging model. Regressing Kriging predictor (yˆ) has such a form [equation 5]” Tutum p. 4.) perturbing the one or more designs (See P. 5 Col. 2: “criterion based on improving upon the best sample found so far, ybest , by searching this probabilistic model[.]” “in Experiment-2, in addition to the EGO-candidate solution, 3 more solution vectors are numerically perturbed around it with sigma = 0.05 and a total of 4 update solutions are printed and evaluated every infill iteration.” Tutum P. 6 col. 1.); repeating the simulating for the one or more designs as perturbed to provide a second optimized valued for the third performance parameter for each of the one or more designs as perturbed (See Fig. 4. The option “Update?” leads back to the fitness evaluation, indicating a repeating process.); determining an error of the one or more design as generated based on a difference between the first optimized value of the third performance parameter and a target optimized value of the third performance parameter ((See equations 9 and 10. “The search space for the expected improvement is highly multimodal as in the case of hyperparameter tuning of the Kriging model, therefore the same rGA algorithm mentioned in the previous section is applied here to find the global optimizer of the EI, Eq.10. This candidate solution is then fed back to the surrogate again to simultaneously improve the accuracy of the approximation as well as to get closer to the global optimum.” Tutum p. 5.) and determining an error of the one or more designs as perturbed (“However, in Experiment-2, in addition to the EGO-candidate solution, 3 more solution vectors are numerically perturbed around it with sigma = 0.05 and a total of 4 update solutions are printed and evaluated every infill iteration. The following two sections show detailed results about the two experiments.” Tutum p. 6.) the error based on a difference between the second optimized value of the third performance parameter and the target optimized value (As written, this claim reads on calculating the error of a generated design based on another value of the “third performance parameter.” Note that Tutum is to a continuous process of training. “The general framework of the EGO is shown in Fig.4 (bottom). EGO iterates until a user-defined stopping criterion is met, e.g., total number of infill points, change in the objective function, tolerance on MSE, etc.” Tutum p. 4. See also Tutum Fig. 4.) ; and selecting, based on the error, at least one of the one or more designs as perturbed or as generated; ((“However, in Experiment-2, in addition to the EGO-candidate solution, 3 more solution vectors are numerically perturbed around it with sigma = 0.05 and a total of 4 update solutions are printed and evaluated every infill iteration. The following two sections show detailed results about the two experiments.” Tutum p. 6. Note that perturbing the solution vectors results in a plurality of solution vectors (i.e. design parameters.) “The search space for the expected improvement is highly multimodal as in the case of hyperparameter tuning of the Kriging model, therefore the same rGA algorithm mentioned in the previous section is applied here to find the global optimizer of the EI, Eq.10. This candidate solution is then fed back to the surrogate again to simultaneously improve the accuracy of the approximation as well as to get closer to the global optimum. The general framework of the EGO is shown in Fig.4 (bottom). EGO iterates until a user-defined stopping criterion is met, e.g., total number of infill points, change in the objective function, tolerance on MSE, etc.” Tutum p. 5 col. 2.) and storing the selected at least one of the one or more designs as perturbed or as generated (This would be understood by POSA based on figure 9. “The sole issue is whether claims 7 and 8 are anticipated[.]” In re Preda, 401 F.2d 825, 826 (C.C.P.A. 1968) “We agree with appellant that Figure 1 of Thacker, by itself, does not disclose every limitation in the appealed claims. However, in considering the disclosure of a reference, it is proper to take into account not only specific teachings of the reference but also the inferences which one skilled in the art would reasonably be expected to draw therefrom. In re Shepard, 319 F.2d 194, 50 CCPA 1439 (1963).” In re Preda.)
15. A system for machine-learning-based design, the system comprising: a processor configured by instructions stored in a memory, the instructions when executed by the processor cause the processor to: (Tutum teaches “First, an effective search space is learned through a variational autoencoder (VAE); second, a surrogate model for functional designs is built; and third, a genetic algorithm is used to simultaneously update the hyperparameters of the surrogate and to optimize the designs using the updated surrogate.” Tutum Abstract. Tutum teaches using 8 training sets of 10,000 samples each with resolution of 150X150 pixles (i.e. 150x150x10,000x8 data points being iteratively trained using equations 1 and 2. See Tutum P. 3, col. 1 and description of Fig. 5. This training is carried out for multiple iterations. See Tutum P. 2. One of ordinary skill in the art would understand the description of training an autoencoder on this amount of data to reference a computer process using of a standard computer including a processor running executing instructions stored in memory. “We agree with appellant that Figure 1 of Thacker, by itself, does not disclose every limitation in the appealed claims. However, in considering the disclosure of a reference, it is proper to take into account not only specific teachings of the reference but also the inferences which one skilled in the art would reasonably be expected to draw therefrom. In re Shepard, 319 F.2d 194, 50 CCPA 1439 (1963).” In re Preda.) generate, with a first neural network trained as an inverse model by a first machine design parameters for one or more designs, in response to a specification for each of the one or more designs,; (“Topology optimization is one way of designing structures subject to manufacturability constraints [16].” Tutum P.1 col.2. “However, designing structures which have to withstand a variety of loading conditions, and to perform certain functional purposes, as well as to comply with manufacturing constraints (tolerances, etc.) is still a challenging problem due to computational requirements in topology optimization.” Tutum P.1 col.2 - P.2 col.1. “Most known surrogates in the literature vary from simple polynomial regression models and moving least squares to neural networks, radial basis functions, Kriging, and support vector regression. Despite the variety in their mathematical construction, they all work based on the same consecutive principles: training (learning) and testing (prediction or generalization). The trained model allows the user to predict any response at an unknown design set at a negligible cost.” Tutum P. 2 col. 1.) optimize, with a second neural network trained as a forward model by the first machine or a second machine, the design parameters based on simulation of the one or more designs using the design parameters generated with the first neural network (“Overall methodology (see Fig.4) briefly involves a Variational Autoencoder (VAE), a 3D printer to produce the springs and an experimental setup for evaluating spring designs and Efficient Global Optimization (EGO) algorithm. The main driver of the pipeline is the EGO algorithm (see Section 2.5), which uses a surrogate (i.e. function approximation) tuned to have a correlation between design parameters (i.e., encodings in this case) and fitness values.” Tutum P.2 col. 2. “[T]he methodology can be generalized to other functional design problems as well.” Tutum P.2 col.1. “All 3D spring models in the initial set are 3D printed and tested in the car-launcher mechanism to get their fitness scores assigned.” Tutum P.2 col.2. “Same GA is also employed here to search for the maximizer of the Expected Improvement (EI) function, where the improvement is defined as finding a function value better than the current best one. After an infill (candidate) design vector is found at the end of an iteration, it is converted into a 3D model and printed for the fitness evaluation. This procedure is repeated for a certain number of iterations. As a result, a list of interesting, reliable and functional spring designs are evolved.” Tutum P.2 col.2. “The goal is to design a spring to propel the car to a distance of 75 cm in a consistent and reliable way. . . . The overall success of a spring design is formulated in [equation 3.]” Tutum P.4. col.1. “After tuning the Kriging model parameters, the next step is to predict a new response value, i.e., a fitness function value, at an unobserved design vector, x*, using the sample data that are used to train the Kriging model.” Tutum P.5 col. 1.
Tutum does not expressly teach using a simulation to update inverse model design parameters.
Sanchez teaches combining inverse and forward modeling as a way of automating the design and evaluation of potentially useful structures. “Here, we review methods for achieving inverse design[.] . . . Among these, deep generative models have been applied to numerous classes of materials: rational design of prospective drugs, synthetic routes to organic compounds, and optimization of photovoltaics and redox flow batteries, as well as a variety of other solid-state materials.” Sanchez P. 1, introduction. “Simulation offers one way of probing this space without experimentation.” Sanchez P. 1 col. 2. “The process (Fig. 1) conventionally involves the following steps: (i) generate a new or improved material concept and simulate its potential suitability; (ii) synthesize the material; (iii) incorporate the material into a device or system; and (iv) characterize and measure the desired properties. This cycle generates feedback to repeat, improve, and refine future cycles of discovery. Each step can take up to several years.” Sanchez P.1 col. 3. “The ultimate aim is to concurrently propose, create, and characterize new materials, with each component transmitting and receiving data simultaneously. This process is called “closing the loop,” and inverse design is a critical facet (12, 15).” Sanchez P.2 col. 1. “Inverse design, as its name suggests, inverts this paradigm by starting with the desired functionality and searching for an ideal molecular structure. Here the input is the functionality and the output is the structure. Functionality need not necessarily map to one unique structure but to a distribution of probable structures.” Sanchez P.2 col.1. Note that Sanchez also teaches forward modeling, that is modeling function based on structure: “Although theory enjoys enormous progress, now routinely modeling molecules, clusters, and perfect as well as defect-laden periodic solids, the size of chemical space is still overwhelming, and smart navigation is required. For this purpose, machine learning (ML), deep learning (DL), and artificial intelligence (AI) have a potential role to play because their computational strategies automatically improve through experience (11). In the context of materials, ML techniques are often used for property prediction, seeking to learn a function that maps a molecular material to the property of choice. Deep generative models are a special class of DL methods that seek to model the underlying probability distribution of both structure and property and relate them in a nonlinear way. By exploiting patterns in massive datasets, these models can distill average and salient features that characterize molecules (12, 13).” Sanchez p. 1 col. 3.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Sanchez because simulation of physical structures created using inverse modeling results in a system that fully automates the task of taking parameters as an input and finding physical structures that are optimized for the parameters while reducing physical experiments or modeling, thereby allowing both forward modeling (generating structure based on function) and reverse modeling (generating functions based on structure) to work together to improve a given technological area.)
and output the design parameters for the design. (Sanchez teaches “Although theory enjoys enormous progress, now routinely modeling molecules, clusters, and perfect as well as defect-laden periodic solids, the size of chemical space is still overwhelming, and smart navigation is required. For this purpose, machine learning (ML), deep learning (DL), and artificial intelligence (AI) have a potential role to play because their computational strategies automatically improve through experience (11). In the context of materials, ML techniques are often used for property prediction, seeking to learn a function that maps a molecular material to the property of choice.” Sanchez P.1 col.3.)
16. The system of claim 15 wherein the specification comprises a first input including at least one of a fixed value of a first performance parameter defining a design requirement and a maximum or minimum value of a second performance parameter defining a constraint for each of the one or more designs, (See rejection of claim 1.) wherein the optimization is for a maximization of a value of a third performance parameter for each of the one or more designs subject to the at least one of the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter. (See rejection of claim 1. See also Tutum p. 1 col. 1 (“Thus, new functional products need to be designed taking the 3D printing-related constraints into account. Topology optimization is one way of designing structures subject to manufacturability constraints [16].”))
17. The system claim 16 wherein the instructions when executed by the processor, further cause the processor to: simulate, by the second neural networked trained by the second machine, operation of each of the one or more designs, the simulation providing a first value for the third performance parameter; perturb the one or more designs; repeat the simulation for the one or more designs as perturbed to provide a second value for the third performance parameter for each of the one or more designs as perturbed; (See rejection of claim 1.) wherein the inverse model generates the design parameters for multiple first possible designs and (See Tutum Fig. 12 and accompanying description.) wherein the simulation provides the first value for the third performance value for the multiple first possible designs, (See rejection of claim 1.) wherein the optimization generates additional second possible designs by perturbation of the design parameters of the multiple first possible designs (“in experiment-2, in addition to the EGO-candidate solution, 3 more solution vectors are numerically perturbed around it with sigma = 0.05 and a total of 4 update solutions are printed and evaluated every infill iteration.” Tutum P. 6 col. 1.) and determines the second value for the third performance parameter for the additional second possible designs, (See rejection of claim 1.) and wherein the instructions further comprise an instruction to select one the first and second possible designs as the design for the output of the design parameters, (“Fig.11 (top and bottom) show the fitness and normalized fitness values as similar to Fig.9. According to the normalized fitness values, the best two designs (Design-43 and 15, respectively, Style-6 and 8) is found on the second infill iteration and the initial set of the EGO.” Tutum P. 7 col. 1.) said selection based on the first value of the third performance parameter for the multiple first possible designs and the second value of the third performance parameter for the additional second possible designs. (See rejection of claim 1. (“However, in Experiment-2, in addition to the EGO-candidate solution, 3 more solution vectors are numerically perturbed around it with sigma = 0.05 and a total of 4 update solutions are printed and evaluated every infill iteration. The following two sections show detailed results about the two experiments.” Tutum p. 6. Note that perturbing the solution vectors results in a plurality of solution vectors (i.e. design parameters.) “The search space for the expected improvement is highly multimodal as in the case of hyperparameter tuning of the Kriging model, therefore the same rGA algorithm mentioned in the previous section is applied here to find the global optimizer of the EI, Eq.10. This candidate solution is then fed back to the surrogate again to simultaneously improve the accuracy of the approximation as well as to get closer to the global optimum. The general framework of the EGO is shown in Fig.4 (bottom). EGO iterates until a user-defined stopping criterion is met, e.g., total number of infill points, change in the objective function, tolerance on MSE, etc.” Tutum p. 5 col. 2.))
18. (original) The system of claim 15 wherein the design parameters comprise settings of variables of the design and wherein the specification comprises values of operational characteristics of the design. (See rejection of claim 15.)
19. (original) A method for machine training a design system, the method comprising: training, by a machine, a first neural network as an inverse model to generate design parameters for a design in response to first values of one or more performance parameters for the design; (“Topology optimization is one way of designing structures subject to manufacturability constraints [16].” Tutum P.1 col.2. “However, designing structures which have to withstand a variety of loading conditions, and to perform certain functional purposes, as well as to comply with manufacturing constraints (tolerances, etc.) is still a challenging problem due to computational requirements in topology optimization.” Tutum P.1 col.2 - P.2 col.1. “Most known surrogates in the literature vary from simple polynomial regression models and moving least squares to neural networks, radial basis functions, Kriging, and support vector regression. Despite the variety in their mathematical construction, they all work based on the same consecutive principles: training (learning) and testing (prediction or generalization). The trained model allows the user to predict any response at an unknown design set at a negligible cost.” Tutum P. 2 col. 1. “In recent years, unsupervised learning, in particular generative design algorithms such as Variational Autoencoders (VAE) and Generative Adversarial Networks (GAN) have become more popular in computer graphics and 3D modeling.” Tutum P. 2. “Overall methodology (see Fig.4) briefly involves a Variational Autoencoder (VAE), a 3D printer to produce the springs and an experimental setup for evaluating spring designs and Efficient Global Optimization (EGO) algorithm. The main driver of the pipeline is the EGO algorithm (see Section 2.5), which uses a surrogate (i.e. function approximation) tuned to have a correlation between design parameters (i.e., encodings in this case) and fitness values.” Tutum P.2 col. 2. “[T]he methodology can be generalized to other functional design problems as well.” Tutum P.2 col.1. “All 3D spring models in the initial set are 3D printed and tested in the car-launcher mechanism to get their fitness scores assigned.” Tutum P.2 col.2.) training, by the machine or another machine, a second neural network as a predictive model to simulate operation of the design based on the generated design parameters to predict second values for the one or more performance parameters from the design; and programming for optimization of the design using the second neural network. (Tutum teaches “All 3D spring models in the initial set are 3D printed and tested in the car-launcher mechanism to get their fitness scores assigned.” Tutum P.2 col.2.
This teaches optimization but fails to teach using a second neural network to predict the properties of the design.
Sanchez teaches “Here, we review methods for achieving inverse design[.] . . . Among these, deep generative models have been applied to numerous classes of materials: rational design of prospective drugs, synthetic routes to organic compounds, and optimization of photovoltaics and redox flow batteries, as well as a variety of other solid-state materials.” Sanchez P. 1, introduction. “Simulation offers one way of probing this space without experimentation.” Sanchez P. 1 col. 2. “In the context of materials, ML techniques are often used for property prediction, seeking to learn a function that maps a molecular material to the property of choice. Deep generative models are a special class of DL methods that seek to model the underlying probability distribution of both structure and property and relate them in a nonlinear way. By exploiting patterns in massive datasets, these models can distill average and salient features that characterize molecules (12, 13). . . . The process (Fig. 1) conventionally involves the following steps: (i) generate a new or improved material concept and simulate its potential suitability; (ii) synthesize the material; (iii) incorporate the material into a device or system; and (iv) characterize and measure the desired properties. This cycle generates feedback to repeat, improve, and refine future cycles of discovery. Each step can take up to several years.” Sanchez P.1 col. 3. “The ultimate aim is to concurrently propose, create, and characterize new materials, with each component transmitting and receiving data simultaneously. This process is called “closing the loop,” and inverse design is a critical facet (12, 15).” Sanchez P.2 col. 1. “Inverse design, as its name suggests, inverts this paradigm by starting with the desired functionality and searching for an ideal molecular structure. Here the input is the functionality and the output is the structure. Functionality need not necessarily map to one unique structure but to a distribution of probable structures.” Sanchez P.2 col.1. To summarize, Sanchez is teaching a system in which a generative model is used to generate a physical structure (inverse design) and explains that this closes the loop in a system where simulation of properties based on structure by neural networks was previously commonplace.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of Sanchez because simulation of physical structures created using inverse modeling results in a system that fully automates the task of taking parameters as an input and finding physical structures that are optimized for the parameters while reducing physical experiments or modeling.)
21. (New) The method of claim 1 wherein the first input comprises both the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter; wherein the first performance parameter is the same as the second performance parameter; and (The “constraints” of Tutum include various binary constraints, where the a yes or no teaches both a “fixed value” and also a max/minimum value. Further, the constraints also include physical parameters associated with numerical values, such as stiffness. See Tutum P. 3 section 2.3 (“1) Printability constraints, 2) Performance constraints. Printability constraints have two subcategories: C1) Is the design (i.e., its image form) blurry, or in other words, could VAE decoder produce a visible pattern? C2) If there is any visible pattern, is it connected all the way from the top to the bottom surface? Performance constraints, which satisfy the printability constraints, are further divided into three sub-categories: C3) Is the spring loadable (i.e., not too stiff)? C4) Did the spring work through all 10 experimental repetitions without breaking? C5) Did the car stay on track in at least 5 out of 10 experiments?”) See also Tutum Fig. 8. See also rejection of claim 1.)) wherein the third performance parameter is optimized subject to both the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter. (See rejection of claim 1.)
22. (New) The method of claim 1, wherein the first input comprises both the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter; (See rejection of claim 1.) wherein the first performance parameter is the not same as the second performance parameter; (See rejection of claim 1.) and wherein the third performance parameter is optimized subject to both the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter. (See rejection of claim 1.)
Claims 23-24 are rejected under 35 U.S.C. 103 as being unpatentable over Tutum, Sanchez, and AAPA in the originally filed Specification.
23. (New) The method of claim 1, wherein the first object is a gas turbine or a component of a gas turbine engine. (The previously cited art does not teach application of the machine learning techniques in the references to parameter selection for maximizing turbine efficiency. However, from the description of the problem in the background section, it is clear that selecting the above listed parameters to improve turbine efficiency manually were known before the effective filing date. “Such designs account for properties such as temperature, turbine shaft power, exhaust pressure of the turbine, and other characteristics. In the turbine use-case, given the restrictions for a given design, the efficiency is to be maximized while constraining exhaust temperatures to being below a maximum value and keeping ambient temperature, turbine shaft power fixed.” Spec. ¶1. “Design engineers work with these parameters, constraints, and evaluation criteria (e.g. the shaft power must be within 0.05% of the design criteria and the overall errors must be within 5% of the design parameters) to design efficient gas turbines and engines. Design engineers spend a lot of time generating new design iterations using a number of time-consuming simulation-based tools.” Spec. ¶2.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of the AAPA such that the machine learning techniques of the previously cited art are used to design turbines with optimized parameters because this may result in more efficient turbines, which ultimately cost less to operate, while potentially shortening the design process.)
24. (New) The method of claim 23, wherein the first input comprises both the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter; wherein the third performance parameter is optimized subject to both the fixed value of the first performance parameter and the maximum or minimum value of the second performance parameter; (See rejection of claim 22.) wherein the first performance parameter and the second performance parameter are each selected from a group comprising one or more of ambient temperature, turbine shaft power, turbine exhaust temperature, and combustor cooling fraction of intake flow, fuel lower heating value, external cooler power, gas turbine exhaust total pressure, combustor cooling fraction of intake flow, combustor cooling efficiency, combustor cooling power, absence or presence of external cooling, ambient air cooling fraction of intake flow, delta verses arbitrary, compressor exit temperature, fuel flow, CO2 exhaust gas composition, H2O exhaust gas composition, N2 exhaust gas composition, Ar exhaust gas composition, O2 exhaust gas composition, SO2 exhaust gas composition, compressor intake flow, turbine inlet temperature, compressor exit pressure, and enthalpy at compressor exit; and wherein the third performance parameter is turbine efficiency. (The previously cited art does not teach application of the machine learning techniques in the references to parameter selection for maximizing turbine efficiency. However, from the description of the problem in the background section, it is clear that selecting the above listed parameters to improve turbine efficiency manually were known before the effective filing date. “Such designs account for properties such as temperature, turbine shaft power, exhaust pressure of the turbine, and other characteristics. In the turbine use-case, given the restrictions for a given design, the efficiency is to be maximized while constraining exhaust temperatures to being below a maximum value and keeping ambient temperature, turbine shaft power fixed.” Spec. ¶1. “Design engineers work with these parameters, constraints, and evaluation criteria (e.g. the shaft power must be within 0.05% of the design criteria and the overall errors must be within 5% of the design parameters) to design efficient gas turbines and engines. Design engineers spend a lot of time generating new design iterations using a number of time-consuming simulation-based tools.” Spec. ¶2.
It would have been obvious to one of ordinary skill in the art before the effective filing date to combine the teaching of the AAPA such that the machine learning techniques of the previously cited art are used to design turbines with optimized parameters because this may result in more efficient turbines, which ultimately cost less to operate, while potentially shortening the design process.)
Response to Arguments
Applicant's arguments filed 12/29/2025 have been fully considered but they are not persuasive.
General:
Based on searching this invention, it is noted that the field of inverse and forward modeling of turbines using neural networks is sparsely populated. It is submitted that clearly written claims directed to a turbine specific way of configuring this type of model (as opposed to merely applying a generic model of this type to turbine specific parameters) may be a way forward, assuming support.
Rejections under § 112b:
The Remarks submit that the Specification provides an objective meaning for “artificial intelligence processor” “such as at paragraph [0038], for example.” Rem. 12. Notably absent from the remarks, is the referenced example. Paragraph 38 merely lists various generic computing structures which can perform processing. If anything, this description tends to show a lack of any plausible structural distinction between an “artificial intelligence processor,” further supporting the position that it is unclear how “artificial intelligence” should be given weight when interpreting the claim.
Applicant asserts that deletion of language forming the basis for each rejection under this section overcomes all other rejections under this section. In some cases, this was sufficient. In other cases, similarly indefinite language was simply added back into the claims in other locations. It is submitted that addressing the underlying reasons for the ambiguity in the claim scope may advance prosecution. See rejections above.
Rejections under § 103:
Applicant asserts that the Non-Final Action “argued that Tatum [sic] discloses . . . simulating . . . operation of each of the one or more designs by a second machine-learned network[.]” Rem. 12, emphasis original. Contrary to Applicant’s assertion, the Non-Final Action clearly articulated the position that “Tutum does not expressly teach simulating the operation of one or more designs by a second ML network.” Non-Final Office Action of 10/01/2025 at 12 (hereafter Non-Fin. Act.), emphasis added. Presumably, Applicant’s position is based on a misunderstanding. Note however, that a reply “must appear throughout to be a bona fide attempt to advance the application . . . to final action[.]” 37 CFR § 1.111. Incorrectly characterizing the position taken in an Office Action on a single occasion is consistent with a simple oversight. It is submitted, however, that repeatedly inventing and then attributing an easily refutable position to the Office would be inconsistent with a reply that appears throughout to be bona-fide attempt to advance prosecution. See 37 CFR § 1.111. See also MPEP §§ 714.02 - 714.04.
The Office Does Not Permit Shift:
Applicant is reminded that the Office does not permit shift. The inventive concept involving aspects of simulation was not elected. While small aspects of simulation that do not require significant search and/or consideration may be permitted, claiming significant details relating to simulation or other non-elected inventions would be inconsistent with the election without traverse filed 09/01/2025.
Conclusion
THIS ACTION IS MADE FINAL. 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.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PAUL M KNIGHT whose telephone number is (571) 272-8646. The examiner can normally be reached Monday - Friday 9-5 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, Michelle Bechtold can be reached on (571) 431-0762. 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.
PAUL M. KNIGHTExaminerArt Unit 2148
/PAUL M KNIGHT/Examiner, Art Unit 2148
Prosecution Insights