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 .
DETAILED ACTION
1. Claims 1-22 are presented for examination.
Claim Objections
2. Claims 6, 9, and 13-14 are objected to because of the following informalities:
As per Claim 6, the limitation “the surrogate model” which is unclear what the limitation refers. Claim 5 recites “a surrogate model method” rather than a surrogate model. For examination purposes, it is interpreted as the surrogate model employed by the surrogate model method.
As per Claim 9, it recites the limitation “they” which is unclear what the limitation refers.
As per Claims 13 and 14, they recite the limitation “wherein the projections generated by the trained artificial neural network” which is unclear what the limitation refers. Claim 1 recites the act of projecting point positions and the production of predicted shape point positions but not a set of projections. For examination purposes, it is interpreted as the predicted shape point positions produced by the trained artificial neural network as recited in claim 1.
Appropriate correction is required.
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
3. Claims 1-22 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
As per Claims 1 and 19- 21, they recite the limitation “training an artificial neural network to project point positions when fed with at least training shape parameters into the artificial neural network onto a respective shape obtained from respective training CAD parameters to thereby produce respective predicted shape point positions as modified or unmodified as an output of the artificial neural network” which is unclear what the limitation refers. In particular, it is unclear how “respective predicted shape point positions” is defined as modified or unmodified. There is a step of “modifying” is missing if whether the recited predicted shape point positions and shape parameters is modified. Further it is unclear if the recited predicted shape point positions and shape parameters are required to be modified, are required to be unmodified, or may be either in the further steps; the metes and bounds of the claim cannot be determined.
As per Claim 9, the limitation “the modified CAD parameters” lacks proper antecedent basis, claim 1 having recited “the CAD parameters as modified or unmodified” but not having established a distinct set of modified CAD parameters. Further, Claim 1 does not recite how “a step of “modifying”. Thus the metes and bounds of the claim cannot be determined.
As per Claim 10, the term “substantially zero values” and “substantially equal to the CAD parameters” is a relative term which renders the claim indefinite. The term “substantially” is not defined by the claim, the specification does not provide a standard for ascertaining the requisite degree, and one of ordinary skill in the art would not be reasonably apprised of the scope of the invention.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
4. Claims 1-8, 11-16, and 18-20 are rejected under 35 U.S.C. 103 as being unpatentable over Hennigh (“Automated Design using Neural Networks and Gradient Descent” listed on IDS filed 07/03/2023) in view of Leblond (“Gradient-based Optimization of Parameterized CAD Geometries” listed on IDS filed 07/03/2023), and further in view of Baydin (“Automatic differentiation in machine learning: a survey” listed on IDS filed 07/03/2023).
As per Claim 1 and 19-20, Hennigh teaches a computer-implemented method/ non-transitory computer program product/ computer-assisted design (“CAD”) system for obtaining gradients of an objective with respect to computer-assisted design (“CAD”) parameters of a shape in a coordinate space (Figure 1, §4 “An automated design optimization problem can be viewed in concrete terms as maximizing some desired fitness function F(x), where F : X → R for some space X of design parameters”), comprising:
training an artificial neural network to project point positions when fed with at least training shape parameters into the artificial neural network onto a respective shape obtained from respective training CAD parameters to thereby produce respective predicted shape point positions as modified or unmodified as an output of the artificial neural network (§4 “takes in the fin parameters and generates a binary image corresponding to the geometry of the heat sink. We refer to this as the parameterization network”: a parameterization network is trained to map design parameters to a corresponding geometry. Examiner’s Note – Hennigh’s design parameters (e.g., the fin heights and airfoil polynomial coefficients that parameterize the geometry input to the parameterization network) read on the recited computer-assisted design (CAD) thus the recited CAD parameters are accordingly mapped to Hennigh’s design parameters throughout this rejection, and Hennigh’s gradient of the objective with respect to those parameters reads on the recited second gradients.).
However, Hennigh fails to teach explicitly obtaining shape from the CAD parameters using a CAD tool system, the shape being at least partially defined by shape point positions in the coordinate space,
obtaining gradients of the objective with respect to the shape point positions, referred to as first gradients,
applying the trained artificial neural network in a forward pass to at least the shape point positions and shape parameters to obtain activation values of the artificial neural network, and
applying the trained artificial neural network in a backward pass to at least the first gradients using the activation values to obtain gradients of the objective with respect to the CAD parameters, referred to as second gradients.
Leblond teaches obtaining shape from the CAD parameters using a CAD tool system, the shape being at least partially defined by shape point positions in the coordinate space (§5 “the initial values of the parameters … are given to the CAD software. The outputs are the initial geometry”), and
obtaining gradients of the objective with respect to the shape point positions, referred to as first gradients (§4, Eq.(1) “represents the shape sensitivity of the objective function J with regard to node displacements. It is given by adjoint solvers and indicates how to move the nodes of the mesh to improve the objective function”). In particular, Leblond teaches gradient-based optimization with regard to CAD parameters, in which the objective sensitivity with respect to the CAD parameters is obtained as the product of the adjoint-supplied sensitivity with respect to node positions and the sensitivity of the shape with respect to the CAD parameters.
Furthermore, Baydin teaches applying the trained artificial neural network in a forward pass to at least the shape point positions and shape parameters to obtain activation values of the artificial neural network (§3.2 “the original function code is run forward, populating intermediate variables”: a forward pass evaluates the trained network and populates its intermediate activation values), and
applying the trained artificial neural network in a backward pass to at least the first gradients using the activation values to obtain gradients of the objective with respect to the CAD parameters, referred to as second gradients (§3.2 “In the second phase, derivatives are calculated by propagating adjoints … in reverse, from the outputs to the inputs”: a backward pass propagates the objective’s first gradients from the outputs to the inputs, using the stored activation values, to obtain the derivatives of the objective with respect to the input CAD parameters). In particular, Baydin teaches reverse-mode automatic differentiation in which a first, forward phase runs the code forward and populates the intermediate activation values, and a second, backward phase propagates adjoints in reverse from the outputs to the inputs to obtain the derivatives of the objective with respect to the inputs.
Hennigh, Leblond, and Baydin are analogous art because they are all from the same field of endeavor, gradient-based optimization of a geometry with respect to its design parameters.
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of cited references. Thus, one of ordinary skill in the art before the effective filing date of the claimed invention would have been motivated to incorporate the teaching of Leblond and Baydin into Hennigh’s invention for the purpose of automated design optimization using neural networks and gradient descent to provide an efficient parameterized computer-assisted design embodiment in which a CAD tool generates the geometry from the CAD parameters and the objective’s first gradients with respect to the shape point positions are obtained by adjoint solvers (Leblond: Abstract), and to incorporate Baydin into Hennigh as modified by Leblond’s invention to provide efficient reverse-mode automatic differentiation accurately that obtains the objective’s second gradients with respect to the CAD parameters through a forward pass and a backward pass over the trained network (Baydin: Abstract).
As per Claim 2, Hennigh teaches wherein the objective is a function to be minimized or maximized (Hennigh §4 “maximizing some desired fitness function”).
As per Claim 3, Hennigh teaches wherein the method further comprises calculating the objective using a physical simulation or a surrogate model of a physical simulation (Hennigh §3 “the gradients are coming from a neural network that is emulating the simulation”, §4 “evaluating the fitness function F can be computationally demanding as is the case with our fluid simulations. The first aspect of our method is to replace F with a computationally efficient neural network Fnet”).
As per Claim 4, Hennigh teaches wherein the method further comprises at least one of generating the CAD parameters and receiving them from a user (Hennigh §4 “we simply make a set of parameter vectors and their corresponding geometries”: Hennigh generates the parameter vectors — the design/CAD parameters — used to train the parameterization network).
As per Claim 5, Hennigh fails to teach explicitly wherein the first gradients are obtained using a method selected from the group consisting of an adjoint method, a surrogate model method, and combinations thereof.
Leblond teaches wherein the first gradients are obtained using a method selected from the group consisting of an adjoint method, a surrogate model method, and combinations thereof (Leblond §4 “It is given by adjoint solvers and indicates how to move the nodes of the mesh to improve the objective function”).
As per Claim 6, Hennigh teaches wherein the surrogate model is another artificial neural network (Hennigh §4.1: a neural network predicts the steady-state flow from the geometry, serving as the surrogate).
As per Claim 7, Hennigh teaches wherein the shape parameters are equal to the CAD parameters (Hennigh §4: the design/CAD parameters are used directly as the shape parameters input to the network).
As per Claim 8, Hennigh teaches wherein the artificial neural network is trained by feeding the point positions together with the respective training CAD parameters, as modified or unmodified, into the artificial neural network (Hennigh §4: training set composed of parameter vectors and their corresponding geometries).
As per Claim 11, Hennigh fails to teach explicitly wherein the artificial neural network is trained to project the point positions onto the respective shape thereby producing projections of the point positions.
Leblond teaches wherein the artificial neural network is trained to project the point positions onto the respective shape thereby producing projections of the point positions (Leblond §4.2 “is used to project 3D geometries on a 2D parametric domain, it corresponds to a reparameterization in 2D space”).
As per Claim 12, Hennigh fails to teach explicitly wherein the artificial neural network is trained to project the point positions associated with a first set of training CAD parameters onto a respective shape obtained from a second set of training CAD parameters.
Leblond teaches wherein the artificial neural network is trained to project the point positions associated with a first set of training CAD parameters onto a respective shape obtained from a second set of training CAD parameters (Leblond §4.2: the mesh of the initial parameter set Ωα₀ is fitted onto the mesh of the perturbed parameter set Ωαᵢ).
As per Claim 13, Hennigh fails to teach explicitly wherein the projections generated by the trained artificial neural network are absolute predicted shape point positions in the coordinate space.
Leblond teaches wherein the projections generated by the trained artificial neural network are absolute predicted shape point positions in the coordinate space (Leblond §4.2: the harmonic fit yields the new absolute coordinates of each node).
As per Claim 14, Hennigh fails to teach explicitly wherein the projections generated by the trained artificial neural network are relative predicted shape point positions in the coordinate space with respect to at least one of the point positions and/or the shape point positions as fed into the artificial neural network.
Leblond teaches wherein the projections generated by the trained artificial neural network are relative predicted shape point positions in the coordinate space with respect to at least one of the point positions and/or the shape point positions as fed into the artificial neural network (Leblond §4.2 “The evaluation of the perturbation is given by the finite differences of the old and new coordinates of each node” (a relative displacement)).
As per Claim 15, Hennigh teaches wherein the method further comprises using (115) the second gradients to modify the CAD parameters by applying a gradient descent algorithm aimed to at least one of maximize and minimize the objective (Hennigh §4 “Fnet is differentiable and can be used to obtain a usable gradient in the direction of maximizing fitness”; §5.3 “scale the computed gradient by some learning rate and add it to the design parameters”: as construed above, Hennigh’s design parameters read on the CAD parameters and its objective gradient reads on the second gradients; Hennigh applies a gradient descent algorithm that modifies those parameters in the direction that optimizes the objective).
As per Claim 16, Hennigh teaches wherein the gradient descent algorithm involves at least one of adding and subtracting the second gradients multiplied by a scaling factor from the CAD parameters (Hennigh §5.3 “scale the computed gradient by some learning rate and add it to the design parameters”: Hennigh scales the computed gradient by a learning rate and adds it to its design parameters, which as construed above read on the CAD parameters, the scaled gradient reading on the second gradients).
As per Claim 18, Hennigh teaches wherein the coordinate space is at least one of a two-dimensional and a three-dimensional coordinate space (Hennigh §5.6: designing both 2D and 3D airfoils).
5. Claims 9, 10, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Hennigh (“Automated Design using Neural Networks and Gradient Descent”) in view of Leblond (“Gradient-based Optimization of Parameterized CAD Geometries”) and Baydin (“Automatic differentiation in machine learning: a survey”), further in view of Flore (EP 3 757 903 A1).
Hennigh as modified by Leblond and Baydin teaches most all the instant invention as applied to claims 1-8, 11-16, and 18-20 above.
As per Claim 9, Hennigh as modified by Leblond and Baydin fails to teach explicitly wherein the trained artificial neural network is further applied to compensation parameters to compensate for the shape parameters when they are equal to the modified CAD parameters.
Flore teaches wherein the trained artificial neural network is further applied to compensation parameters to compensate for the shape parameters when they are equal to the modified CAD parameters ([0046] “a concatenation layer CC which may concatenate the flattened output of the convolutional layers with scalar input parameters”; [0054] “The concatenating layer CC may accept scalar parameters SP as input”; [0048] “The scalar input may be used for various scalar parameters associated with the physical structure”: the recited “compensation parameters” read on the additional scalar design parameters to which Flore’s network is further applied through the concatenation layer). In particular, Flore teaches a trained neural network that is further applied, through a concatenation layer, to one or more scalar design parameters concatenated as an additional input alongside the geometry.
Hennigh, Leblond, Baydin, and Flore are analogous art because they are all from the same field of endeavor, neural-network-assisted optimization of a geometry with respect to design parameters.
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of cited references. Thus, one of ordinary skill in the art before the effective filing date of the claimed invention would have been motivated to incorporate Flore into Hennigh as modified by Leblond and Baydin for the purpose of automated design optimization using neural networks and gradient descent to provide an efficient parameterized computer-assisted design embodiment in which a CAD tool generates the geometry from the CAD parameters and the objective’s first gradients with respect to the shape point positions are obtained by adjoint solvers (Leblond: Abstract), and to incorporate Baydin into Hennigh as modified by Leblond’s invention to provide efficient reverse-mode automatic differentiation accurately that obtains the objective’s second gradients with respect to the CAD parameters through a forward pass and a backward pass over the trained network (Baydin: Abstract). Further the motivation to combine the teaching of Flore is to reduce the computational complexity of the optimization (Flore: Abstract).
As per Claim 10, Hennigh as modified by Leblond and Baydin fails to teach explicitly wherein the shape parameters take zero values or substantially zero values, and wherein the trained artificial neural network is further applied to compensation parameters to compensate for the shape parameters, and wherein the compensation parameters are equal or substantially equal to the CAD parameters.
Flore teaches wherein the shape parameters take zero values or substantially zero values, and wherein the trained artificial neural network is further applied to compensation parameters to compensate for the shape parameters, and wherein the compensation parameters are equal or substantially equal to the CAD parameter ([0046], [0048], [0053] “a set of scalar geometry parameters which each may represent a parametric geometry template”, [0054]; [0034] “The geometry template 10 may be represented by a set of scalar geometry parameters”: the scalar parameters supplied to the network are the scalar geometry parameters of the geometry template, i.e., parameters of the CAD-type parametric representation. Examiner’s Note – the recited “compensation parameters … equal or substantially equal to the CAD parameters” read on Flore’s scalar design/geometric parameters supplied to the network through the concatenation layer).
As per Claim 17, Hennigh as modified by Leblond and Baydin fails to teach explicitly wherein the second gradient values are returned to a user through an application programming interface.
Flore teaches wherein the second gradient values are returned to a user through an application programming interface ([0068] “the optimization system 400 may comprise a user interface subsystem 480 which may be configured to, during operation of the system 400, enable a user to interact with the system 400”, “may further comprise a display output interface (not separately shown in Fig. 10) for outputting display data 482 to a rendering device”: the optimization system’s interface subsystem returns the system’s outputs, including the computed values, to the user).
6. Claim 21 is rejected under 35 U.S.C. 103 as being unpatentable over Hennigh (“Automated Design using Neural Networks and Gradient Descent”) in view of Leblond (“Gradient-based Optimization of Parameterized CAD Geometries”).
As per Claim 21, Hennigh teaches a computer-implemented method for obtaining gradients of shape point positions representing a shape in a coordinate space with respect to computer-assisted design (“CAD”) parameters used for obtaining the shape (§4 “we use the differentiable nature of the trained network to get a gradient on the parameter space when performing optimization”), comprising:
training an artificial neural network to project point positions when fed with at least training shape parameters into the artificial neural network onto a respective shape obtained from respective training CAD parameters to thereby produce respective predicted shape point positions as modified or unmodified as an output of the artificial neural network, the training shape parameters being equal to the training CAD parameters as modified or unmodified (§4 “takes in the fin parameters and generates a binary image corresponding to the geometry of the heat sink. We refer to this as the parameterization network”: a parameterization network is trained to map design/CAD parameters to the corresponding geometry); and
applying the trained artificial neural network to at least the shape point positions and shape parameters to obtain gradients of the shape point positions with respect to the shape parameters, the shape parameters being equal to the CAD parameters as modified or unmodified (§4 “we use the differentiable nature of the trained network to get a gradient on the parameter space”: differentiation through the trained parameterization network yields the gradients of the produced geometry with respect to the design parameters. Examiner’s Note – under the broadest reasonable interpretation, Hennigh’s design parameters read on the recited computer-assisted design (CAD) parameters, and Hennigh’s gradient on the parameter space reads on the recited gradients of the shape point positions with respect to the CAD parameters.)
However, Hennigh fails to teach explicitly obtaining shape point positions in the coordinate space; and
providing the gradients to a user or returning the gradients to a program.
Leblond teaches obtaining shape point positions in the coordinate space (§5 “the initial values of the parameters … are given to the CAD software. The outputs are the initial geometry”: the CAD software outputs the geometries whose node coordinates are the obtained shape point positions); and
providing the gradients to a user or returning the gradients to a program (§5 “The optimization loop uses the result of the Eq. (1) into a gradient descent algorithm”: the computed gradients are returned to the gradient descent program of the optimization loop). In particular, Leblond teaches computing the shape sensitivity with regard to CAD parameters and supplying it to the optimization loop.
Hennigh and Leblond are analogous art because they are both from the same field of endeavor, gradient-based optimization of a geometry with respect to its design parameters.
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of cited references. Thus, one of ordinary skill in the art before the effective filing date of the claimed invention would have been motivated to incorporate Leblond into Hennigh’s invention for the purpose of automated design optimization using neural networks and gradient descent to provide an efficient parameterized computer-assisted design embodiment in which a CAD tool that provides the shape sensitivity with regard to CAD parameters for use in the optimization loop (Leblond: Abstract).
7. Claim 22 is rejected under 35 U.S.C. 103 as being unpatentable over Hennigh (“Automated Design using Neural Networks and Gradient Descent”) in view of Leblond (“Gradient-based Optimization of Parameterized CAD Geometries”), further in view of Flore (EP 3 757 903 A1).
Hennigh as modified by Leblond teaches most all the instant invention as applied to claim 21 above.
As per Claim 22, Hennigh as modified by Leblond fails to teach explicitly wherein the gradients are provided to the user via a user interface, via an application programming interface, stored via at least one selected from the group consisting of in a cloud computing environment, locally in a computing device, saved in a file, and combinations thereof.
Flore teaches wherein the gradients are provided to the user via a user interface, via an application programming interface, stored via at least one selected from the group consisting of in a cloud computing environment, locally in a computing device, saved in a file, and combinations thereof ([0068]; [0070] “transitory or non-transitory data 510 representing a geometry of a physical structure as obtained by an optimization described elsewhere in this specification”). In particular, Flore teaches providing the optimization output to a user through an interface and storing it on a computer-readable medium.
Hennigh, Leblond and Flore are analogous art because they are all from the same field of endeavor, neural-network-assisted optimization of a geometry with respect to design parameters.
It would have been obvious to one having ordinary skill in the art before the effective filing date of the claimed invention to combine the teachings of cited references. Thus, one of ordinary skill in the art before the effective filing date of the claimed invention would have been motivated to incorporate Flore into Hennigh as modified by Leblond’s invention for the purpose of automated design optimization using neural networks and gradient to provide an efficient parameterized computer-assisted design embodiment in which a CAD tool that provides the shape sensitivity with regard to CAD parameters for use in the optimization loop (Leblond: Abstract, §4) and to provide a user interface and storage by which the optimization output which reduced the computational complexity of the optimization is delivered to a user (Flore: Abstract, [0068], [0070]).
Conclusion
8. The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure:
Umetani, N. and Bickel, B., (“Learning three-dimensional flow for interactive aerodynamic design”) teaches projecting shape point positions onto a parameterized shape representation for machine-learned aerodynamic design.
Baqué, P. et al., (“Geodesic Convolutional Shape Optimization”) teaches a geodesic convolutional surrogate model providing objective gradients with respect to mesh vertex positions.
Hoyer, S. et al., (“Neural Reparameterization Improves Structural Optimization”) teaches optimizing the parameters of a neural network that outputs a design using forward evaluation and automatic differentiation backward passes.
Chen (US 8,065,244 B2) teaches constructing neural network surrogate models of physical simulations.
Mezghanni (US 2022/0101105 A1) teaches gradient descent optimization of the latent design parameters of a deep generative model of three-dimensional shapes.
9 Any inquiry concerning this communication or earlier communications from the examiner should be directed to EUNHEE KIM whose telephone number is (571)272-2164. The examiner can normally be reached Monday-Friday 9am-5pm 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 at (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.
EUNHEE KIM
Primary Examiner
Art Unit 2188
/EUNHEE KIM/Primary Examiner, Art Unit 2188
Prosecution Insights