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 .
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 2023/12/22. The submission is in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement is being considered by the examiner.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-15 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding independent claims 1, 6 and 11
Step 1 -- whether the claim falls within any statutory category. See MPEP 2106.03
Claim 1 is drawn to a processor implemented method claim; claim 6 is drawn to a nanofeature prediction system claim; claim 11 is drawn to a non-transitory machine-readable information storage medium claim. Therefore, each of these claims falls under one of the four categories of statutory subject matter (process/method, machine/product/apparatus, manufacture, or composition of matter). (Step 1: YES).
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding independent claim 1, the claim is directed to a processor implemented method that receives spectral response data of a plasmonic metasurface, converts the data into chromaticity coordinates, preprocesses the coordinates using a filter, and trains a network on a combined loss function, the method comprising:
The claim recites the limitations of "converting, by the NPS via the one or more hardware processors, the spectral response data into a coordinate space, the coordinate space further comprising a plurality of two-dimensional chromaticity coordinates, wherein each 2D chromaticity coordinate is associated with a dimension information of a metaunit of the plurality of metaunits," "preprocessing, by the NPS via the one or more hardware processors, the plurality of 2D chromaticity coordinates using a Euclidean filter to obtain a plurality of preprocessed chromaticity coordinates," and "wherein the combined loss function comprises a Kullback-Leibler (KL) divergence loss, a dimension prediction loss and a coordinate reconstruction loss."
These limitations are directed towards the abstract idea of a mathematical concept, specifically mathematical calculations and mathematical relationships (see MPEP § 2106.04(a)(2), subsections I.A and I.C). The recited "converting" of spectral response data into two-dimensional chromaticity coordinates is the calculation of CIE 1931 chromaticity values from a reflectance spectrum, which is performed by numerically integrating the spectrum against the color matching functions and normalizing the resulting tristimulus values, as described in the specification at ¶¶ [045]-[046]. The recited "Euclidean filter" is the computation of a Euclidean distance between coordinate points and the comparison of that computed distance against a threshold value, as described in the specification at ¶¶ [048]-[050]. The recited "Kullback-Leibler (KL) divergence loss," "dimension prediction loss" and "coordinate reconstruction loss" are mathematical formulas, which the specification sets out in equation form at ¶¶ [060]-[064], including the total loss equation at ¶ [064].
The claim further recites the limitation of "training, by the NPS via the one or more hardware processors, a remodeled variational autoencoder (VAE) network based, at least in part, on the plurality of preprocessed chromaticity coordinates and a combined loss function to obtain a trained VAE network."
This limitation is directed towards the abstract idea of a mathematical concept, specifically a mathematical calculation. The recited training consists of iteratively computing the value of the combined loss function and adjusting network weight values by gradient descent to minimize that computed value, as described in the specification at ¶¶ [054]-[055], which is a series of mathematical calculations.
As explained in the MPEP, when a claim recites multiple abstract ideas that fall in the same or different groupings, examiners should consider the limitations together as a single abstract idea, rather than as a plurality of separate abstract ideas to be analyzed individually. See MPEP § 2106.04, subsection II.B. As the above limitations fall within the same grouping of abstract ideas, namely mathematical concepts, these limitations are considered together as a single abstract idea for further analysis. (Step 2A Prong One: YES).
Independent claim 6 is a nanofeature prediction system claim reciting similar limitations to claim 1 and is directed towards the abstract idea for similar reasons.
Independent claim 11 is a non-transitory machine-readable information storage medium claim reciting similar limitations to claim 1 and is directed towards the abstract idea for similar reasons.
Step 2A Prong 2 -- whether the claim as a whole integrates the recited judicial exception into a practical application of the exception or whether the claim is "directed to" the judicial exception. This evaluation is performed by (1) identifying whether there are any additional elements recited in the claim beyond the judicial exception, and (2) evaluating those additional elements individually and in combination to determine whether the claim as a whole integrates the exception into a practical application. See MPEP 2106.04(d).
Regarding independent claim 1, this claim recites additional elements of:
"receiving, by a nanofeature prediction system (NPS) via one or more hardware processors, a spectral response data associated with a plasmonic metasurface, the plasmonic metasurface comprising a plurality of metaunits," "a nanofeature prediction system (NPS)," "one or more hardware processors," and "a remodeled variational autoencoder (VAE) network."
The recited "receiving" limitation amounts to mere data gathering. It is necessary to acquire the spectral response data in order to perform the recited mathematical calculations upon it. The claim places no limits on how the spectral response data is obtained, and the specification confirms at ¶ [044] that "the spectral response data is the publicly available data." The receiving limitation does not impose any other meaningful limits on the claim and is therefore insignificant extra-solution activity. See MPEP § 2106.05(g).
The recited "nanofeature prediction system (NPS)" and "one or more hardware processors" are recited at a high level of generality and amount to no more than generic computer components that merely act as a tool on which the abstract idea is performed. This amounts to mere instructions to apply the exception using a generic computer component. See MPEP § 2106.05(f).
The recited "a remodeled variational autoencoder (VAE) network" is invoked merely as a tool to perform the recited mathematical calculations. The claim recites only the idea of training a network on a combined loss function without reciting any detail as to how the network is structured or how it operates to solve a technical problem. As explained in MPEP § 2106.05(f), examiners may consider whether the claim recites only the idea of a solution or outcome, whether the claim invokes computers or other machinery merely as a tool to perform an existing process, and the particularity or generality of the application of the judicial exception. Here, the claim omits any details as to how the VAE network is remodeled or how the network operates, and instead recites only the idea of an outcome, namely obtaining a trained VAE network. Therefore, this limitation represents no more than mere instructions to apply the judicial exception on a computer. See MPEP § 2106.05(f).
Further, the recitation of a "plasmonic metasurface" and "metaunits" merely confines the mathematical calculations to the field of metasurface design and amounts to no more than generally linking the use of the judicial exception to a particular technological environment or field of use. See MPEP § 2106.05(h).
The claim does not recite any improvement to the functioning of a computer or to any other technology or technical field. See MPEP § 2106.05(a). Although the specification describes at ¶¶ [006] and [047] a difficulty arising from the clustering of chromaticity coordinates near the achromatic point of the CIE chart, and at ¶ [068] asserts improved prediction accuracy, claim 1 recites only the training of a network and does not recite any use of the trained network to design, fabricate, or control any structure or device. The claim therefore does not reflect the asserted improvement, and the recited generic computer components merely perform the mathematical calculations. See MPEP § 2106.05(a).
Accordingly, the additional elements, individually and in combination, do not integrate the judicial exception into a practical application. (Step 2A Prong Two: NO). Claim 1 is therefore directed to the judicial exception. (Step 2A: YES).
Regarding independent claim 6, this claim is drawn to a nanofeature prediction system claim reciting similar limitations to claim 1 and is rejected under the same rationale. Claim 6 also recites additional elements of:
"a memory storing instructions," "one or more communication interfaces," and "one or more hardware processors coupled to the memory via the one or more communication interfaces, wherein the one or more hardware processors are configured by the instructions to."
These limitations amount to no more than generic computer components that merely act as a tool on which the abstract idea is performed, and further amount to generally linking the use of the judicial exception to a particular technological environment or field of use. See MPEP §§ 2106.05(f) and 2106.05(h). The claim does not recite any particular structure or configuration of the memory, interfaces, or processors beyond that of a generic computer, and thus fails to integrate the exception into a practical application.
Regarding independent claim 11, this claim is drawn to a non-transitory machine-readable information storage medium claim reciting similar limitations to claim 1 and is rejected under the same rationale. Claim 11 also recites additional elements of:
"One or more non-transitory machine-readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause."
These limitations amount to no more than generic computer components that merely act as a tool on which the abstract idea is performed, and further amount to generally linking the use of the judicial exception to a particular technological environment or field of use. See MPEP §§ 2106.05(f) and 2106.05(h), and thus fail to integrate the exception into a practical application.
Step 2B -- whether the claim amounts to significantly more than the judicial exception. See MPEP § 2106.05.
Regarding independent claim 1, the additional elements beyond the judicial exception are the recited "receiving" step, the "nanofeature prediction system (NPS)," the "one or more hardware processors," and the "remodeled variational autoencoder (VAE) network."
As explained in the Step 2A Prong Two analysis, the nanofeature prediction system, the hardware processors, and the VAE network are recited at a high level of generality and amount to no more than mere instructions to apply the exception using generic computer components. The analysis under Step 2A Prong Two is carried through to Step 2B. See MPEP § 2106.05(f). Generic computer implementation is well-understood, routine, and conventional activity in the field. See MPEP §§ 2106.05(d) and 2106.05(d)(II).
Further, a conclusion that an additional element is insignificant extra-solution activity in Step 2A should be re-evaluated in Step 2B. See MPEP § 2106.05, subsection I.A. At Step 2B, the re-evaluation takes into account whether the extra-solution activity is well-understood, routine, and conventional in the field. See MPEP § 2106.05(g). Here, the recited "receiving" of spectral response data is mere data gathering recited at a high level of generality, and the specification confirms at ¶ [044] that the spectral response data is publicly available data. Therefore, this limitation remains insignificant extra-solution activity upon reconsideration and does not amount to significantly more.
Even when considered in combination, these additional elements represent mere instructions to apply an exception using generic computer components together with insignificant extra-solution activity, and therefore do not provide an inventive concept. (Step 2B: NO). Claim 1 is not eligible.
Regarding independent claims 6 and 11, the additional elements of "a memory storing instructions," "one or more communication interfaces," "one or more hardware processors," and "one or more non-transitory machine-readable information storage mediums" constitute generic computer implementation, which is well-understood, routine, and conventional in the field. See MPEP § 2106.05(d). These elements, individually and in combination, do not amount to significantly more than the judicial exception for the same reasons set forth above with respect to claim 1. (Step 2B: NO). Claims 6 and 11 are not eligible.
Regarding dependent claims 2-5, 7-10 and 12-15
Claims 2-5, 7-10 and 12-15 merely narrow the previously cited abstract idea limitations. For the reasons described above with respect to independent claims 1, 6 and 11, these judicial exceptions are not meaningfully integrated into a practical application, nor do they amount to significantly more than the abstract idea. The claims recite limitations similar to those described for the independent claims above and do not provide anything more than the mathematical concepts and mental processes that are practically capable of being performed in the human mind with the assistance of pen and paper. Therefore, claims 2-5, 7-10 and 12-15 also recite abstract ideas that do not integrate into a practical application or amount to significantly more than the judicial exception, and are rejected under 35 U.S.C. § 101.
Step 1 -- whether the claim falls within any statutory category. See MPEP 2106.03
Claims 2-5 are drawn to a method claim; claims 7-10 are drawn to a system claim; claims 12-15 are drawn to a non-transitory machine-readable information storage medium claim. Therefore, each of these claims falls under one of the four categories of statutory subject matter (process/method, machine/product/apparatus, manufacture, or composition of matter). (Step 1: YES).
Step 2A Prong 1 – whether the claim recites a judicial exception. See MPEP 2106.04, subsection II.
Regarding claims 2, 7 and 12, these claims recite the limitations of "creating, by the NPS via the one or more hardware processors, the Euclidean filter based on a predefined threshold value," "identifying, by the NPS via the one or more hardware processors, a 2D chromaticity coordinate as a first point," "calculating, by the NPS via the one or more hardware processors, a Euclidean distance between the first point and a second point present in the coordinate space," "selecting, by the NPS via the one or more hardware processors, the second point among the plurality of 2D chromaticity coordinates based on a comparison of the calculated Euclidean distance and the Euclidean filter," "adding, by the NPS via the one or more hardware processors, the second point in a point selection queue based on the selection," "identifying, by the NPS via the one or more hardware processors, a next point available after the second point as the second point in the plurality of 2D chromaticity coordinates," and "until the Euclidean distance between the first point and each 2D chromaticity coordinate of the plurality of 2D chromaticity coordinates is calculated in the coordinate space."
These limitations are directed towards the abstract idea of a mathematical concept and a mental process. The recited "calculating" of a Euclidean distance between two points is a mathematical calculation. See MPEP § 2106.04(a)(2), subsection I.C. The recited "identifying," "selecting" based on a comparison, and "adding" to a queue are concepts performed in the human mind, including observation, evaluation, and judgment. See MPEP § 2106.04(a)(2), subsection III. A human being can plot chromaticity coordinates on a chart, measure with a ruler the distance between a first point and each successive point, compare each measured distance against a threshold, and write the retained points onto a list using pen and paper.
Regarding claims 3, 8 and 13, these claims recite the limitations of "accessing, by the NPS via the one or more hardware processors, a VAE network, the VAE network further comprising a probabilistic encoder and a probabilistic decoder," and "remodeling, by the NPS via the one or more hardware processors, the VAE by adding a multi-layer perceptron (MLP) to a latent space of the VAE to obtain the remodeled VAE, the remodeled VAE further comprising the probabilistic encoder, the probabilistic decoder, and a MLP predictor."
These limitations are directed towards the abstract idea of a mathematical concept. The recited probabilistic encoder and probabilistic decoder are defined by probability distributions, and the recited remodeling consists of specifying an additional mathematical mapping from the latent variable to a predicted output vector, as described in the specification at ¶¶ [052] and [059]. Selecting and arranging the constituent mathematical mappings of a model does not remove the limitations from the mathematical concepts grouping.
Regarding claims 4, 9 and 14, these claims recite the limitations of "passing, by the NPS via the one or more hardware processors, one or more 2D chromaticity coordinates as an input to the probabilistic encoder of the remodeled VAE network," "passing, by the NPS via the one or more hardware processors, the dimension information of the metaunit corresponding to each 2D chromaticity coordinate of the one or more 2D chromaticity coordinates as an output to the MLP predictor of the remodeled VAE network," and "training, by the NPS via the one or more hardware processors, the remodeled VAE network based, at least in part, on the input of the probabilistic encoder, the output to the MLP predictor and the combined loss function to obtain the trained VAE network."
These limitations are directed towards the abstract idea of a mathematical concept. The recited passing of an input and an output merely designates the variables upon which the recited mathematical calculations operate, and the recited training consists of computing the combined loss function and adjusting weight values to minimize it, as described in the specification at ¶¶ [053]-[054].
Regarding claims 5, 10 and 15, these claims recite the limitation of "obtaining, by the NPS via the one or more hardware processors, a dimension information of a new metaunit corresponding to the new 2D chromaticity coordinate using the trained VAE network."
This limitation is directed towards the abstract idea of a mathematical concept. The recited obtaining consists of evaluating the trained mathematical model at an input value to compute an output value, which is a mathematical calculation. See MPEP § 2106.04(a)(2), subsection I.C.
As the above limitations fall within the same groupings of abstract ideas as those recited in the independent claims, these limitations are considered together with the previously identified abstract idea for further analysis. See MPEP § 2106.04, subsection II.B. (Step 2A Prong One: YES).
Step 2A Prong 2 -- whether the claim as a whole integrates the recited judicial exception into a practical application of the exception. See MPEP 2106.04(d).
Regarding claims 2-4, 7-9 and 12-14, these claims recite additional elements of "the NPS," "the one or more hardware processors," "a VAE network," and "a multi-layer perceptron (MLP)."
These limitations amount to no more than mere instructions to apply the judicial exception using generic computer components, and further amount to generally linking the use of the judicial exception to a particular technological environment or field of use. See MPEP §§ 2106.05(f) and 2106.05(h). The claims recite a generic computer or generic computer components that merely act as a tool on which the mathematical calculations are performed, and thus fail to integrate the exception into a practical application.
Regarding claims 5, 10 and 15, these claims recite additional elements of "receiving, by the NPS via the one or more hardware processors, a 2D chromaticity coordinate from a user device," "a user device," and "displaying, by the NPS via the one or more hardware processors, the dimension information of the new metaunit on the user device."
The recited "receiving" of a 2D chromaticity coordinate from a user device amounts to mere data gathering necessary to perform the recited mathematical calculation, and is therefore insignificant extra-solution activity. See MPEP § 2106.05(g). The recited "displaying" of the dimension information amounts to necessary data outputting that merely presents the result of the mathematical calculation, and is likewise insignificant extra-solution activity. See MPEP § 2106.05(g). The recited "user device" is claimed at a high level of generality and amounts to a generic computer component acting as a tool. See MPEP § 2106.05(f).
Notably, claims 5, 10 and 15 do not recite fabricating, manufacturing, or otherwise physically forming any metaunit or metasurface having the obtained dimension information, nor do they recite controlling any apparatus based on the obtained dimension information. The claims terminate at the display of a computed result. Accordingly, the additional elements do not apply the judicial exception in a manner that imposes a meaningful limit on the exception, and do not effect a particular treatment or transformation. See MPEP §§ 2106.05(c) and 2106.05(e).
Accordingly, the additional elements of claims 2-5, 7-10 and 12-15, individually and in combination, do not integrate the judicial exception into a practical application. (Step 2A Prong Two: NO).
Step 2B -- whether the claim amounts to significantly more than the judicial exception. See MPEP § 2106.05.
Regarding claims 2-4, 7-9 and 12-14, the additional elements identified above amount to generic computer implementation, which is well-understood, routine, and conventional activity in the field. See MPEP § 2106.05(d). The analysis under Step 2A Prong Two is carried through to Step 2B, and these elements amount to no more than mere instructions to apply the exception using a generic computer component. See MPEP § 2106.05(f).
Regarding claims 5, 10 and 15, the additional elements found to be insignificant extra-solution activity at Step 2A Prong Two have been re-evaluated at Step 2B in accordance with MPEP § 2106.05, subsection I.A. Receiving user input at a computing device and displaying a computed result on that device are well-understood, routine, and conventional computer functions. See MPEP § 2106.05(d)(II), which recognizes receiving or transmitting data over a network and presenting offers and gathering statistics as computer functions recognized as well-understood, routine, and conventional. Therefore, these limitations remain insignificant extra-solution activity upon reconsideration and do not amount to significantly more.
Accordingly, dependent claims 2-5, 7-10 and 12-15 do not include additional elements, either individually or in combination, that amount to significantly more than the recited judicial exception. Therefore, these claims fail to provide an inventive concept under Step 2B. (Step 2B: NO). Claims 2-5, 7-10 and 12-15 are not eligible.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The text of those sections of Title 35, U.S. Code not included in this action can be found in a prior Office action.
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.
Claims 1–4, 6–9, and 11–14 are rejected under 35 U.S.C. 103 as being unpatentable over Roberts et al. (Roberts), Non-Patent Literature, "A Deep Learning Approach to the Forward Prediction and Inverse Design of Plasmonic Metasurface Structural Color," Applied Physics Letters, Vol. 119, No. 6, Article 061101, published August 9, 2021, in view of Cook (Cook), Non-Patent Literature, "Stochastic Sampling in Computer Graphics," ACM Transactions on Graphics, Vol. 5, No. 1, pages 51–72, published January 1986, in view of Ma et al. (Ma), Non-Patent Literature, "Probabilistic Representation and Inverse Design of Metamaterials Based on a Deep Generative Model with Semi-Supervised Learning Strategy," Advanced Materials, Vol. 31, Issue 35, Article 1901111, pages 1901111 (1 of 9) through 1901111 (9 of 9), published July 7, 2019, together with its Supporting Information, separately paginated pages 1 of 16 through 16 of 16, and further in view of Parekh et al. (Parekh), Non-Patent Literature, "Variational Autoencoder based Metamodeling for Multi-Objective Topology Optimization of Electrical Machines," arXiv:2201.08877, published January 21, 2022.
As to independent Claim 1, Roberts teaches:
"A processor implemented method, comprising:" — Roberts teaches a computer-implemented deep learning method executed on processing hardware (Roberts, page 061101-2, column 2, paragraph 4: "LUMERICAL FDTD solver was used to simulate the reflectance spectra of 4620 metamaterial configurations"; page 061101-3, column 1, paragraph 4: "MSE loss function and Keras's Adam optimizer were used"; page 061101-3, column 1, paragraph 5: "The FDNN model was trained for a total of 4795 epochs").
"receiving, by a nanofeature prediction system (NPS) via one or more hardware processors, a spectral response data associated with a plasmonic metasurface," — Roberts teaches receiving reflectance spectra collected for a PDMS-Al nanorod plasmonic metasurface (Roberts, page 061101-1, column 1, paragraph 3: "Such plasmonic metamaterials are said to exhibit structural color (SC)"; page 061101-2, column 1, paragraph 3: "In this work, polydimethylsiloxane (PDMS) nanorod metasurfaces with aluminum (Al) coatings are used as a proof of the concept for the development of DL models"; page 061101-2, column 2, paragraph 4: "50-point discretized reflectance spectra were collected for each material between 400 and 750 nm"; page 061101-3, column 1, paragraph 3: "Encoded simulation data were shuffled and randomly split into training and test sets with a ratio of 80:20"). The deep learning system of Roberts that predicts the nanoscale dimension parameters reads on the recited nanofeature prediction system (Roberts, page 061101-4, column 1, paragraph 1: "the inverse network learns to design metamaterial parameters d, h, and t for target colors").
"the plasmonic metasurface comprising a plurality of metaunits;" — Roberts teaches a periodic array of subwavelength unit cells (Roberts, page 061101-1, column 1, paragraph 1: "Metamaterials can be defined as artificial materials, for which their electromagnetic (EM) response is dependent on periodic subwavelength structures as opposed to intrinsic material properties"; page 061101-2, column 1, paragraph 4: "As seen in Fig. 1(a), the structure topology consists of metal disk arrays, suspended by dielectric pillars with a metal back reflector"; page 061101-2, column 2, paragraph 4: "a unit cell periodicity of 200 nm was fixed for all materials. This allowed for use of the unit cell approximation, simulating 1/4 of one nanorod with symmetrical boundaries"; FIG. 1(a), page 061101-2: "Metasurface composed of polymeric pillar (PDMS) with metal coating (Al). Pitch, metal thickness, height, and pillar diameter are P, t, h, and d, respectively").
"converting, by the NPS via the one or more hardware processors, the spectral response data into a coordinate space," — Roberts teaches converting the reflectance spectra into the CIE 1931 chromaticity gamut (Roberts, page 061101-3, column 1, paragraph 2: "Direct interpretation of the metamaterial reflectance spectra cannot provide an intuitive indication of perceived color. For use as a practical design tool, dimensionality reduction to the CIE 1931 chromaticity gamut was used to encode SC chromaticity").
"the coordinate space further comprising a plurality of two-dimensional chromaticity coordinates," — Roberts teaches that the coordinate space is two-dimensional and contains the chromaticity values encoded from all simulated configurations (Roberts, page 061101-4, column 2, paragraph 2: "Due to the dimensionality reduction of color to the 2D CIE domain, here, arbitrary large sets of random colors without corresponding structures can be used in training"; FIG. 1(c), page 061101-2: "Range SCs from simulated metamaterials plotted on CIE 1931 gamut"; FIG. 2(b), page 061101-3: "CIE 1931 scatterplot showing FDNN model predictions on the test set").
"wherein each 2D chromaticity coordinate is associated with a dimension information of a metaunit of the plurality of metaunits;" — Roberts teaches that each encoded training sample pairs a chromaticity coordinate with the corresponding metaunit dimensions (Roberts, page 061101-2, column 2, paragraph 3: "Training data consist of tensors encoding the metamaterial structure and color, where color is calculated from reflectance spanning the visible EM spectrum"; page 061101-3, column 1, paragraph 1: "A method of parameter vectorization was used to encode the metamaterial structures, where pillar diameter, height, and Al thickness take numerical values ranging between 5 and 200 nm"; Table I, page 061101-2, column 2: "Al thickness (t) 10–30 [nm]," "Pillar diameter (d) 38–200 [nm]," "Pillar height (h) 30–100 [nm]"; FIG. 5, page 061101-5: "CIE 1931 x,y coordinates of both target colors and designed colors are displayed").
Roberts teaches removing the chromaticity coordinates clustered about the achromatic CIE point by applying a square mask, reducing the data set from 4620 to 2394 samples (Roberts, page 061101-3, column 2, paragraph 1).
However, Roberts does not teach "preprocessing, by the NPS via the one or more hardware processors, the plurality of 2D chromaticity coordinates using a Euclidean filter to obtain a plurality of preprocessed chromaticity coordinates;"
In the same field of endeavor, Cook teaches preprocessing a plurality of points in a two-dimensional coordinate space using a Euclidean distance threshold to obtain a retained set of points (Cook, page 55, column 2, paragraph 3: "In effect, the samples are randomly placed with the restriction that no two samples are closer together than a certain distance"; page 57, column 1, paragraph 4: "A lookup table is created by generating random sample locations and discarding any locations that are closer than a certain distance to any of the locations already chosen. Locations are generated until the sampling region is full… The locations and filter values are stored in a table"). The minimum-distance restriction of Cook is a Poisson disk criterion, the disk being the set of points within a fixed Euclidean radius, such that the certain distance of Cook reads on the recited Euclidean filter and the resulting stored set of retained locations reads on the recited plurality of preprocessed chromaticity coordinates. Cook further teaches the effect of applying the restriction (Cook, page 57, column 1, paragraph 3: "With a purely random distribution, the samples tend to bunch up in some places and leave large gaps in other places").
Roberts and Cook are analogous to the claimed invention, Roberts being from the field of endeavor of machine-learning inverse design of metamaterial geometry and Cook being reasonably pertinent to the problem of removing clustered points from a set of two-dimensional coordinate points (MPEP § 2141.01(a)). Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to preprocess the plurality of 2D chromaticity coordinates of Roberts using the Euclidean minimum-distance criterion of Cook. The motivation to combine Roberts and Cook is as recited by Cook (page 57, column 1, paragraph 3: "the samples tend to bunch up in some places and leave large gaps in other places"), since Roberts identifies the same clustering problem in the CIE 1931 coordinate space (Roberts, page 061101-3, column 2, paragraph 1: "trainable parameters are skewed toward data clusters when using an averaged loss metric").
Roberts teaches training a tandem autoencoder whose joining hidden layer encodes the geometry parameters d, h, and t and whose loss is the mean squared error between the input and output CIE chromaticity vectors (Roberts, page 061101-4, column 1, paragraph 1; FIG. 3, page 061101-4).
The combination of Roberts and Cook, however, does not teach "training, by the NPS via the one or more hardware processors, a remodeled variational autoencoder (VAE) network based, at least in part, on the plurality of preprocessed chromaticity coordinates and a combined loss function to obtain a trained VAE network, wherein the combined loss function comprises a Kullback-Leibler (KL) divergence loss, a dimension prediction loss and a coordinate reconstruction loss."
In the same field of endeavor, Ma teaches:
"a variational autoencoder (VAE) network" — Ma teaches incorporating a variational autoencoder structure comprising a recognition model and a generation model (Ma, page 1901111 (2 of 9), column 1: "we introduce latent variables as a probabilistic representation of metamaterial design by incorporating a variational auto-encoder (VAE) structure in our model, which encodes the designed pattern together with the corresponding optical response into a latent space"; Ma, Supporting Information, Section 3, page 5 of 16: "the unlabeled data contribute to the recognition model and generation model, which in combination form a conditional variational auto-encoder (VAE)"; Ma, Supporting Information, Section 3, page 4 of 16: "the recognition model and generation model together form an encoder-decoder structure that compresses the geometric pattern x of the metamaterial into a low-dimensional latent space given its corresponding spectra y").
"a combined loss function" — Ma teaches a total loss combining a generation loss and a prediction loss (Ma, Supporting Information, Section 3, page 5 of 16, Equation S8: "Finally, the total loss for the deep generative model is: ℒ = Σ L_l(x,y) + Σ L_l(x,y) + α Σ L_r(x,y)"; Ma, Supporting Information, Section 4, page 7 of 16: "The total loss includes generation loss (Equation S1 and Equation S6) that accounts for latent space configuration and input image reconstruction, and prediction loss (Equation S7) to optimize forward prediction accuracy").
"a Kullback-Leibler (KL) divergence loss" — Ma teaches the KL divergence term of the loss objective (Ma, Supporting Information, Section 2, page 3 of 16, Equation S1: "log p_θ(x|y) ≥ −KL[q_∅(z|x,y)||p_θ(z)] + 𝔼_{q_∅(z|x,y)}[log p_θ(x|y,z)] = −L_l(x,y)," and "Assuming Gaussian latent variables, the first KL-divergence term of Equation 1 can be marginalized"; Ma, Supporting Information, Section 3, page 6 of 16: "the KL divergence term on the right side of equation S5 has a close-form expression and is differentiable").
"a coordinate reconstruction loss" — Ma teaches the reconstruction component of the generation loss (Ma, Supporting Information, Section 4, page 7 of 16: the generation loss "accounts for latent space configuration and input image reconstruction"; Ma, page 1901111 (3 of 9), column 1: "our model gradually produces more and more accurate reconstruction of the inputs as the training proceeds").
"a dimension prediction loss" — Ma teaches a mean square error regression loss on the predicted output of the prediction model (Ma, Supporting Information, Section 3, page 5 of 16, Equation S7: "For the predictive distribution P_p(y|x), we use the mean square error (MSE) as a regression loss: L_r(x,y) = (ŷ − y)², ŷ~P_p(y|x)").
"to obtain a trained VAE network" — Ma teaches end-to-end training on the combined loss (Ma, Supporting Information, Section 4, page 7 of 16: "The entire deep generative model is trained in an end-to-end manner with both labeled and unlabeled data… We use Adam optimizer and the hyper-parameter α in Equation S8 is set to 2000").
Therefore, Ma teaches the above limitation.
Roberts, Cook, and Ma are analogous to the claimed invention, Roberts and Ma being from the same field of endeavor of machine-learning inverse design of metamaterial geometry from a target optical response. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the autoencoder training of Roberts with the variational autoencoder and combined loss function of Ma. The motivation to combine Roberts, Cook, and Ma is as recited by Ma (Supporting Information, Section 1, page 2 of 16: the tandem method "eliminates many candidates and is unable to obtain the 'complete' metamaterial designs," whereas "by sampling from the latent space, we can make stable, diverse and interpretable inverse retrievals"), since Roberts acknowledges the same deficiency (Roberts, page 061101-4, column 2, paragraph 3: "Evaluation of the inverse model is challenging due to the many-to-one relationship").
The combination of Roberts, Cook, and Ma, however, does not teach "a remodeled variational autoencoder (VAE) network."
In the same field of endeavor, Parekh teaches a variational autoencoder remodeled by the addition of a multi-layer perceptron to the latent space, trained on a loss comprising the three recited terms (Parekh, page 1, column 2, Abstract: "The VAE is trained to find a latent space while training a multi-layer perceptron (MLP) for KPIs prediction. After training, via a latent space, the decoder and multi-layer neural network will function as meta-models for sampling new designs and predicting associated KPIs, respectively"; Parekh, page 2, column 1, paragraph 2: "In this work, we introduced MLP along with usual VAE structure in the latent space for KPIs prediction"; Parekh, page 2, column 2, paragraph 2: "The network is split into three parts: encoder (e_θ), decoder (d_φ) and KPIs predictor (k_ψ)"; Parekh, page 2, column 1, Equation 5: "ŷ := k_ψ(z)," and "The MLP is being trained alongside the continuous latent space to predict the KPIs"; Parekh, page 2, column 1, paragraph following Equation 5: "The training loss includes three terms: parameter reconstruction loss (ℓ₂-norm), Kullback–Leibler (KL) divergence (regularization) and loss (ℓ₂-norm) for the KPIs prediction (MLP)"). The predicted quantities of Parekh are physical dimensional parameters (Parekh, page 2, Table II: "Air gap 0.8–1.8 mm," "Height of magnet 4.5–6.5 mm," "Stator tooth height 12–20 mm," "Iron length 120–160 mm," "Rotor outer diameter 150–180 mm").
Roberts, Cook, Ma, and Parekh are analogous to the claimed invention, Parekh being reasonably pertinent to the problem of predicting physical design dimensions from a variational autoencoder latent representation (MPEP § 2141.01(a); Sanofi-Aventis Deutschland GmbH v. Mylan Pharms. Inc., 66 F.4th 1373, 1380 (Fed. Cir. 2023)). Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to combine the variational autoencoder of Roberts, Cook, and Ma with the multi-layer perceptron predictor of Parekh. The motivation to combine Roberts, Cook, Ma, and Parekh is as recited by Parekh (page 1, column 2, Abstract: "the decoder and multi-layer neural network will function as meta-models for sampling new designs and predicting associated KPIs, respectively"), such that one would attach a multi-layer perceptron predictor to the latent space to directly predict the metaunit dimension information d, h, and t that Roberts encodes in that layer (Roberts, page 061101-4, column 1, paragraph 1: "The hidden layer joining the two networks is three neurons deep, representing the latent encoding of geometry parameters d, h, and t").
As to dependent Claim 2, the limitations of Claim 1 from which Claim 2 depends are rejected under the same rationale set forth above with respect to Claim 1.
Regarding the additional limitations of Claim 2, Roberts teaches:
"The processor implemented method of claim 1, wherein the step of preprocessing, by the NPS via the one or more hardware processors, the plurality of 2D chromaticity coordinates using the Euclidean filter to obtain the plurality of preprocessed chromaticity coordinates comprises:" — Roberts teaches preprocessing the plurality of 2D chromaticity coordinates by applying a distance-based geometric criterion to obtain a reduced set of coordinates used for training (Roberts, page 061101-3, column 2, paragraph 1: "As observed in Fig. 1(c), a large quantity of the simulated metamaterials are clustered around the achromatic CIE point… To overcome this, data centered around the achromatic point were removed using a square mask. The dataset was reduced from 4620 to 2394 samples and used for all subsequent model development").
"creating, by the NPS via the one or more hardware processors, the plurality of preprocessed chromaticity coordinates using one or more points available in the point selection queue, the plurality of preprocessed chromaticity coordinates further comprising one or more 2D chromaticity coordinates, and dimension information of the metaunit corresponding to each 2D chromaticity coordinate of the one or more 2D chromaticity coordinates." — Roberts teaches creating the reduced training set from the retained points, each retained point comprising a CIE 1931 chromaticity coordinate together with the corresponding metaunit dimensions (Roberts, page 061101-3, column 2, paragraph 1: "The dataset was reduced from 4620 to 2394 samples and used for all subsequent model development"; page 061101-2, column 2, paragraph 3: "Training data consist of tensors encoding the metamaterial structure and color, where color is calculated from reflectance spanning the visible EM spectrum"; page 061101-3, column 1, paragraph 1: "A method of parameter vectorization was used to encode the metamaterial structures, where pillar diameter, height, and Al thickness take numerical values ranging between 5 and 200 nm"; page 061101-3, column 1, paragraph 2: "dimensionality reduction to the CIE 1931 chromaticity gamut was used to encode SC chromaticity"). The retained 2394 samples of Roberts read on the recited one or more points available in the point selection queue, and the pillar diameter d, pillar height h, and Al thickness t paired with each retained CIE 1931 x,y coordinate read on the recited dimension information of the metaunit corresponding to each 2D chromaticity coordinate.
Roberts teaches applying a square mask centered on the achromatic CIE point to remove clustered coordinates from the data set (Roberts, page 061101-3, column 2, paragraph 1).
However, Roberts does not teach:
"creating, by the NPS via the one or more hardware processors, the Euclidean filter based on a predefined threshold value;"
"for each 2D chromaticity coordinate of the plurality of 2D chromaticity coordinate present in the coordinate space, performing: identifying, by the NPS via the one or more hardware processors, a 2D chromaticity coordinate as a first point;"
"calculating, by the NPS via the one or more hardware processors, a Euclidean distance between the first point and a second point present in the coordinate space, wherein the second point is a next 2D chromaticity coordinate available among the plurality of 2D chromaticity coordinates;"
"selecting, by the NPS via the one or more hardware processors, the second point among the plurality of 2D chromaticity coordinates based on a comparison of the calculated Euclidean distance and the Euclidean filter;"
"adding, by the NPS via the one or more hardware processors, the second point in a point selection queue based on the selection; and"
"identifying, by the NPS via the one or more hardware processors, a next point available after the second point as the second point in the plurality of 2D chromaticity coordinates,"
"until the Euclidean distance between the first point and each 2D chromaticity coordinate of the plurality of 2D chromaticity coordinates is calculated in the coordinate space; and"
In the same field of endeavor, Cook teaches:
Regarding "creating, by the NPS via the one or more hardware processors, the Euclidean filter based on a predefined threshold value": Cook teaches establishing a fixed minimum separation distance that governs which point locations are retained (Cook, page 55, column 2, paragraph 3: "In effect, the samples are randomly placed with the restriction that no two samples are closer together than a certain distance"; page 57, column 1, paragraph 3: "The minimum distance restriction decreases the magnitude of the noise"). The certain distance of Cook is the radius of the Poisson disk, which is the set of points lying within a fixed Euclidean radius of a given point, and therefore reads on the recited Euclidean filter created based on a predefined threshold value.
Regarding "for each 2D chromaticity coordinate of the plurality of 2D chromaticity coordinate present in the coordinate space, performing: identifying, by the NPS via the one or more hardware processors, a 2D chromaticity coordinate as a first point": Cook teaches designating the point locations already chosen as the reference points against which each further location is evaluated (Cook, page 57, column 1, paragraph 4: "discarding any locations that are closer than a certain distance to any of the locations already chosen"). Each of the locations already chosen of Cook reads on the recited first point.
Regarding "calculating, by the NPS via the one or more hardware processors, a Euclidean distance between the first point and a second point present in the coordinate space, wherein the second point is a next 2D chromaticity coordinate available among the plurality of 2D chromaticity coordinates": Cook teaches computing the separation between each further location and the locations already chosen (Cook, page 57, column 1, paragraph 4: "discarding any locations that are closer than a certain distance to any of the locations already chosen"). The determination whether a further location is closer than the certain distance requires computing the separation between that location and each location already chosen, and each such further location under evaluation reads on the recited second point.
Regarding "selecting, by the NPS via the one or more hardware processors, the second point among the plurality of 2D chromaticity coordinates based on a comparison of the calculated Euclidean distance and the Euclidean filter": Cook teaches retaining a location when its computed separation satisfies the minimum distance and discarding it otherwise (Cook, page 57, column 1, paragraph 4: "discarding any locations that are closer than a certain distance to any of the locations already chosen"). The retention of locations not closer than the certain distance reads on the recited selection of the second point based on a comparison of the calculated Euclidean distance and the Euclidean filter.
Regarding "adding, by the NPS via the one or more hardware processors, the second point in a point selection queue based on the selection": Cook teaches accumulating each retained location into a stored table (Cook, page 57, column 1, paragraph 4: "A lookup table is created by generating random sample locations and discarding any locations that are closer than a certain distance to any of the locations already chosen… The locations and filter values are stored in a table"). The lookup table of Cook, into which each retained location is entered, reads on the recited point selection queue.
Regarding "identifying, by the NPS via the one or more hardware processors, a next point available after the second point as the second point in the plurality of 2D chromaticity coordinates" and "until the Euclidean distance between the first point and each 2D chromaticity coordinate of the plurality of 2D chromaticity coordinates is calculated in the coordinate space": Cook teaches successively advancing to each further location and continuing until every location has been evaluated (Cook, page 57, column 1, paragraph 4: "Locations are generated until the sampling region is full"). The successive evaluation of each further location against the locations already chosen, continuing until the region is full, reads on the recited identification of a next point available after the second point as the second point and on the recited iteration until the Euclidean distance between the first point and each 2D chromaticity coordinate is calculated.
Cook further teaches the consequence of omitting the minimum distance restriction (Cook, page 57, column 1, paragraph 3: "With a purely random distribution, the samples tend to bunch up in some places and leave large gaps in other places").
Roberts and Cook are analogous to the claimed invention, Roberts being from the field of endeavor of machine-learning inverse design of metamaterial geometry and Cook being reasonably pertinent to the problem of removing clustered points from a set of two-dimensional coordinate points (MPEP § 2141.01(a)). Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to preprocess the plurality of 2D chromaticity coordinates of Roberts by the iterative minimum-distance comparison and retention procedure of Cook. The motivation to combine Roberts and Cook is as recited by Cook (page 57, column 1, paragraph 3: "the samples tend to bunch up in some places and leave large gaps in other places"), since Roberts identifies the same clustering condition in the CIE 1931 coordinate space and its effect on training (Roberts, page 061101-3, column 2, paragraph 1: "a large quantity of the simulated metamaterials are clustered around the achromatic CIE point… trainable parameters are skewed toward data clusters when using an averaged loss metric").
As to dependent Claim 3, the limitations of Claims 1 and 2 from which Claim 3 depends are rejected under the same rationale set forth above with respect to Claims 1 and 2.
Regarding the additional limitations of Claim 3, Ma teaches:
"The processor implemented method of claim 2, wherein the step of training, by the NPS via the one or more hardware processors, the remodeled variational autoencoder (VAE) network based, at least in part, on the plurality of preprocessed chromaticity coordinates and the combined loss function to obtain the trained VAE network is preceded by: accessing, by the NPS via the one or more hardware processors, a VAE network," — Ma teaches accessing a variational autoencoder network prior to training it on the combined loss (Ma, page 1901111 (2 of 9), column 1, paragraph 2: "we introduce latent variables as a probabilistic representation of metamaterial design by incorporating a variational auto-encoder (VAE) structure in our model, which encodes the designed pattern together with the corresponding optical response into a latent space"; Ma, Supporting Information, Section 3, page 6 of 16: "the unlabeled data contribute to the recognition model and generation model, which in combination form a conditional variational auto-encoder (VAE)").
"the VAE network further comprising a probabilistic encoder and a probabilistic decoder; and" — Ma teaches that the VAE network comprises a recognition model defining the posterior distribution of the latent variable and a generation model defining the generative distribution of the design, which together constitute an encoder-decoder pair (Ma, Supporting Information, Section 2, page 3 of 16: "Recognition model: The recognition model defines the posterior distribution of latent variable z given input variable x and the associated spectra y, P_r(z|x,y), which models the encoding process of metamaterial structures into latent space," and "Generation model: The generation model defines the generative distribution of geometric pattern x given reflection spectra y and latent variable z, P_g(x|y,z), which models the inverse design process of metamaterial given required spectra"; Ma, Supporting Information, Section 3, page 4 of 16: "the recognition model and generation model together form an encoder-decoder structure that compresses the geometric pattern x of the metamaterial into a low-dimensional latent space given its corresponding spectra y"; Ma, page 1901111 (3 of 9), FIG. 1(b) caption: "The recognition model encodes the metamaterial pattern with its optical response into a low-dimensional latent space… The generation model accepts the optical response and the sampled latent variable to produce feasible metamaterial designs according to specific requirements"). The recognition model of Ma, being defined as a probability distribution, reads on the recited probabilistic encoder, and the generation model of Ma, likewise defined as a probability distribution, reads on the recited probabilistic decoder.
Ma teaches that the deep generative model further comprises a prediction model implemented as a separate neural network built on the extracted feature vector (Ma, Supporting Information, Section 4, page 6 of 16).
However, the combination of Roberts, Cook, and Ma does not teach:
"remodeling, by the NPS via the one or more hardware processors, the VAE by adding a multi-layer perceptron (MLP) to a latent space of the VAE to obtain the remodeled VAE,"
"the remodeled VAE further comprising the probabilistic encoder, the probabilistic decoder, and a MLP predictor."
In the same field of endeavor, Parekh teaches:
Regarding "remodeling, by the NPS via the one or more hardware processors, the VAE by adding a multi-layer perceptron (MLP) to a latent space of the VAE to obtain the remodeled VAE": Parekh teaches modifying a conventional variational autoencoder structure by introducing a multi-layer perceptron into its latent space (Parekh, page 2, column 1, paragraph 2: "In this work, we introduced MLP along with usual VAE structure in the latent space for KPIs prediction"; Parekh, page 1, column 2, Abstract: "The VAE is trained to find a latent space while training a multi-layer perceptron (MLP) for KPIs prediction"; Parekh, page 2, column 1, following Equation 5: "The MLP is being trained alongside the continuous latent space to predict the KPIs"). The addition by Parekh of an MLP to the latent space of an otherwise usual VAE structure reads on the recited remodeling of the VAE by adding a multi-layer perceptron to a latent space of the VAE to obtain the remodeled VAE.
Regarding "the remodeled VAE further comprising the probabilistic encoder, the probabilistic decoder, and a MLP predictor": Parekh teaches that the resulting network consists of three constituent parts, namely an encoder, a decoder, and a predictor implemented by the multi-layer perceptron (Parekh, page 2, column 2, paragraph 2: "The network is split into three parts: encoder (e_θ), decoder (d_φ) and KPIs predictor (k_ψ)"; Parekh, page 2, column 1, Equations 3–5: the encoder produces the latent vector z, "p̂ := d_φ(z)" at the decoder, and "ŷ := k_ψ(z), where ψ are the trainable network parameters and ŷ is the predicted vector of KPIs"; Parekh, page 1, column 2, Abstract: "After training, via a latent space, the decoder and multi-layer neural network will function as meta-models for sampling new designs and predicting associated KPIs, respectively"). The encoder and decoder of Parekh correspond to the recited probabilistic encoder and probabilistic decoder taught by Ma above, and the KPIs predictor k_ψ of Parekh, implemented by the multi-layer perceptron, reads on the recited MLP predictor.
Roberts, Cook, Ma, and Parekh are analogous to the claimed invention, Roberts and Ma being from the same field of endeavor of machine-learning inverse design of metamaterial geometry from a target optical response, and Cook and Parekh being reasonably pertinent to the particular problems with which the inventor was concerned, namely removing clustered points from a set of two-dimensional coordinate points and predicting physical design dimensions from a variational autoencoder latent representation (MPEP § 2141.01(a); Sanofi-Aventis Deutschland GmbH v. Mylan Pharms. Inc., 66 F.4th 1373, 1380 (Fed. Cir. 2023)). Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to remodel the variational autoencoder of Ma by adding the multi-layer perceptron predictor of Parekh to its latent space. The motivation to combine Roberts, Cook, Ma, and Parekh is as recited by Parekh (page 1, column 2, Abstract: "the decoder and multi-layer neural network will function as meta-models for sampling new designs and predicting associated KPIs, respectively"), such that the remodeled network provides both generative design capability at the decoder and direct prediction of the metaunit dimension information d, h, and t at the MLP predictor, which Roberts encodes in the latent layer of its network (Roberts, page 061101-4, column 1, paragraph 1: "The hidden layer joining the two networks is three neurons deep, representing the latent encoding of geometry parameters d, h, and t").
As to dependent Claim 4, the limitations of Claims 1, 2 and 3 from which Claim 4 depends are rejected under the same rationale set forth above with respect to Claims 1, 2 and 3.
Regarding the additional limitations of Claim 4, Roberts teaches:
"The processor implemented method of claim 3, wherein the step of training, by the NPS via the one or more hardware processors, the remodeled VAE network based, at least in part, on the plurality of preprocessed chromaticity coordinates and the combined loss function to obtain the trained VAE network comprises:" — Roberts teaches training the network on the reduced set of chromaticity coordinates by backpropagation and gradient descent (Roberts, page 061101-3, column 2, paragraph 1: "The dataset was reduced from 4620 to 2394 samples and used for all subsequent model development"; page 061101-4, column 1, paragraph 1: "The tandem model is trained using backpropagation and gradient decent (Keras Adam optimizer) to minimize the loss between the desired color and the model prediction").
"passing, by the NPS via the one or more hardware processors, one or more 2D chromaticity coordinates as an input to" — Roberts teaches supplying the CIE 1931 chromaticity coordinates as the input tensor to the encoding side of the network during training (Roberts, FIG. 3, page 061101-4: "Autoencoder architecture used for IDNN training, where CIE input and output tensors encode desired and predicted color respectively"; page 061101-4, column 1, paragraph 1: "The loss is defined as the MSE between the desired color (input vector) and predicted color (output vector)").
"passing, by the NPS via the one or more hardware processors, the dimension information of the metaunit corresponding to each 2D chromaticity coordinate of the one or more 2D chromaticity coordinates as an output to" — Roberts teaches that the metaunit dimension information constitutes the output of the network that maps from the latent representation (Roberts, page 061101-4, column 1, paragraph 1: "The hidden layer joining the two networks is three neurons deep, representing the latent encoding of geometry parameters d, h, and t," and "the inverse network learns to design metamaterial parameters d, h, and t for target colors"; page 061101-4, column 1, paragraph 2: "The use of sigmoid activation functions [Eq. (1)] in the final IDNN layer of 3 neurons bounds the output between 0 and 1. This restricts the predicted geometries within the range of simulated features in the training set due to feature normalization").
Roberts teaches training a network in which the input color vector and the geometry output are related through a joining latent layer, the forward and inverse models being trained in separate stages with the forward weights frozen (Roberts, page 061101-4, column 1, paragraph 1).
However, the combination of Roberts and Cook does not teach:
"the probabilistic encoder of the remodeled VAE network;"
"training, by the NPS via the one or more hardware processors, the remodeled VAE network based, at least in part, on the input of the probabilistic encoder, the output to the MLP predictor and the combined loss function to obtain the trained VAE network."
In the same field of endeavor, Ma teaches:
Regarding "the probabilistic encoder of the remodeled VAE network": Ma teaches that the input variable is supplied to the recognition model, which defines the posterior distribution of the latent variable and performs the encoding operation of the variational autoencoder (Ma, Supporting Information, Section 2, page 3 of 16: "Recognition model: The recognition model defines the posterior distribution of latent variable z given input variable x and the associated spectra y, P_r(z|x,y), which models the encoding process of metamaterial structures into latent space"; Ma, page 1901111 (2 of 9), FIG. 1(b) caption: "The recognition model encodes the metamaterial pattern with its optical response into a low-dimensional latent space"). The recognition model of Ma, being defined as a probability distribution that performs the encoding operation, reads on the recited probabilistic encoder.
Regarding "training, by the NPS via the one or more hardware processors, the remodeled VAE network based, at least in part, on the input of the probabilistic encoder, the output to the MLP predictor and the combined loss function to obtain the trained VAE network": Ma teaches training the network end-to-end in a single training process on the combined loss, in which each labeled input-output pair contributes to both the prediction loss and the generation loss (Ma, Supporting Information, Section 4, page 7 of 16: "The entire deep generative model is trained in an end-to-end manner with both labeled and unlabeled data… In each step, the labeled data are fed as complete pairs to the model, which contribute to both prediction loss and generation loss"; Ma, Supporting Information, Section 3, page 6 of 16: "all models are implemented by deep neural networks, and the entire model is optimized end-to-end in a single training process using Stochastic Gradient Descent (SGD) method").
Roberts, Cook, and Ma are analogous to the claimed invention, Roberts and Ma being from the same field of endeavor of machine-learning inverse design of metamaterial geometry from a target optical response. Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to pass the 2D chromaticity coordinates of Roberts as the input to the probabilistic encoder of Ma and to train the network end-to-end on the combined loss of Ma. The motivation to combine Roberts, Cook, and Ma is as recited by Ma (page 1901111 (2 of 9), column 1, paragraph 2: "our model solves both the forward and inverse problems at the same time, and can be trained in an end-to-end manner"), since Roberts requires the forward and inverse models to be trained in separate stages (Roberts, page 061101-4, column 1, paragraph 1: "The weights of the FDNN are frozen and used to predict color from the geometry vector output of the IDNN").
The combination of Roberts, Cook, and Ma, however, does not teach "the MLP predictor of the remodeled VAE network; and"
In the same field of endeavor, Parekh teaches that the physical design parameters are routed as the output of the multi-layer perceptron predictor attached to the latent space, and that the encoder, decoder, and predictor are optimized concurrently under the three-term training loss (Parekh, page 2, column 1, Equation 5 and adjacent text: "ŷ := k_ψ(z), where ψ are the trainable network parameters and ŷ is the predicted vector of KPIs," and "The MLP is being trained alongside the continuous latent space to predict the KPIs"; Parekh, page 2, column 2, paragraph 2: "The network is split into three parts: encoder (e_θ), decoder (d_φ) and KPIs predictor (k_ψ)"; Parekh, page 2, column 1, following Equation 5: "The objective of the training process is to improve encoding, reconstruction and prediction process by concurrently optimizing the model parameters θ, φ, ψ. The training loss includes three terms: parameter reconstruction loss (ℓ₂-norm), Kullback–Leibler (KL) divergence (regularization) and loss (ℓ₂-norm) for the KPIs prediction (MLP)"; Parekh, page 2, Table II: "Air gap 0.8–1.8 mm," "Height of magnet 4.5–6.5 mm," "Stator tooth height 12–20 mm," "Iron length 120–160 mm," "Rotor outer diameter 150–180 mm"). The predicted vector ŷ output by the KPIs predictor k_ψ of Parekh, whose components are physical dimensional parameters, reads on the recited output to the MLP predictor of the remodeled VAE network.
Roberts, Cook, Ma, and Parekh are analogous to the claimed invention, Parekh being reasonably pertinent to the problem of predicting physical design dimensions from a variational autoencoder latent representation (MPEP § 2141.01(a); Sanofi-Aventis Deutschland GmbH v. Mylan Pharms. Inc., 66 F.4th 1373, 1380 (Fed. Cir. 2023)). Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to pass the metaunit dimension information of Roberts as the output of the MLP predictor of Parekh. The motivation to combine Roberts, Cook, Ma, and Parekh is as recited by Parekh (page 2, column 1: "The objective of the training process is to improve encoding, reconstruction and prediction process by concurrently optimizing the model parameters θ, φ, ψ"), such that the encoding of the chromaticity coordinates, the reconstruction of those coordinates, and the prediction of the metaunit dimensions d, h, and t are optimized concurrently in a single training process.
Claim 6 recites a nanofeature prediction system (NPS) comprising a memory storing instructions, one or more communication interfaces, and one or more hardware processors configured by the instructions to perform the same operations recited in Claim 1. Claim 6 is therefore rejected under the same rationale set forth above with respect to Claim 1.
Claim 7 recites the same limitations as Claim 2 and is therefore rejected under the same rationale set forth above with respect to Claim 2.
Claim 8 recites the same limitations as Claim 3 and is therefore rejected under the same rationale set forth above with respect to Claim 3.
Claim 9 recites the same limitations as Claim 4 and is therefore rejected under the same rationale set forth above with respect to Claim 4.
Claim 11 recites one or more non-transitory machine-readable information storage mediums comprising one or more instructions which when executed by one or more hardware processors cause the same operations recited in Claim 1. Claim 11 is therefore rejected under the same rationale set forth above with respect to Claim 1.
Claim 12 recites the same limitations as Claim 2 and is therefore rejected under the same rationale set forth above with respect to Claim 2.
Claim 13 recites the same limitations as Claim 3 and is therefore rejected under the same rationale set forth above with respect to Claim 3.
Claim 14 recites the same limitations as Claim 4 and is therefore rejected under the same rationale set forth above with respect to Claim 4.
Claims 5, 10, and 15 are rejected under 35 U.S.C. 103 as being unpatentable Roberts, in view of Cook, Ma together with its Supporting Information, in view of Parekh, and further in view of Park et al. (Park), US 2022/0293225 A1, published September 15, 2022.
As to dependent Claim 5, the limitations of Claim 1 from which Claim 5 depends are rejected under the same rationale set forth above with respect to Claim 1.
Regarding the additional limitations of Claim 5, Roberts teaches:
"The processor implemented method of claim 1, further comprising:" — Roberts teaches deploying the trained inverse model to design metasurfaces from a desired color (Roberts, page 061101-1, Abstract: "The tightly constrained inverse model allows for the instantaneous design of metasurfaces, given a desired color, with an accuracy of >86%, making it suitable for commercial use as well as the acceleration of photonics research").
"obtaining, by the NPS via the one or more hardware processors, a dimension information of a new metaunit corresponding to the new 2D chromaticity coordinate using the trained VAE network; and" — Roberts teaches obtaining the metaunit dimensions corresponding to a target chromaticity coordinate not present in the training data by applying the trained inverse model (Roberts, page 061101-4, column 2, paragraph 3: "Instead, 5000 randomly sampled 'unseen' colors were used as a test set. The corresponding IDNN-designed metamaterials were validated using the FDNN model in the tandem architecture depicted in Fig. 3"; page 061101-4, column 1, paragraph 1: "the inverse network learns to design metamaterial parameters d, h, and t for target colors"; page 061101-5, column 1, paragraph 1: "As seen in Fig. 5, 10 metamaterials were designed using the IDNN, given random target colors"; page 061101-5, column 1, paragraph 2: "The model is able to accurately design SCs different to those in the original training data and thus interpolating the complex relationship between color and structure"). The metamaterial designed by Roberts for a previously unseen target color reads on the recited new metaunit, and the parameters d, h, and t obtained for that metamaterial read on the recited dimension information of a new metaunit.
Roberts teaches supplying a target CIE 1931 chromaticity coordinate to the trained inverse model and presenting the resulting design parameters together with the corresponding chromaticity coordinates (Roberts, page 061101-5, FIG. 5: "Evaluation of 10 IDNN-designed metamaterials where random target SCs are compared to the true SC. CIE 1931 x,y coordinates of both target colors and designed colors are displayed").
However, the combination of Roberts, Cook, Ma and Parekh does not teach the following limitations, the untaught portions of which are shown in bold. In the same field of endeavor, Park teaches:
"receiving, by the NPS via the one or more hardware processors, a 2D chromaticity coordinate from a user device;" — Park teaches that the targeted properties from which the material information is to be derived are supplied as input to the trained encoder by way of a computing device operated by a user (Park, paragraph [0057]: "Subsequently, the trained encoder is used to acquire material information which satisfies targeted wave properties (S300). In other words, since the encoder of the autoencoder may be trained to derive information on a material from given wave properties according to the above-described training of the autoencoder, targeted wave properties are input to the trained encoder, and the result is acquired to derive information on a material from the given wave properties"; Park, paragraph [0038]: "The encoder 100, the decoder 200, and the data converter 300 may be implemented as a computing device. For example, the computing device may include a processor and a memory coupled to the processor, and the memory may include instructions configured to cause the processor to perform operations according to a method of designing a material using deep learning"; Park, paragraph [0028]: the terms of the disclosure "may vary depending on an intention of a user or an operator"). The computing device of Park through which a user supplies the targeted properties to the trained encoder reads on the recited user device from which the 2D chromaticity coordinate is received.
"displaying, by the NPS via the one or more hardware processors, the dimension information of the new metaunit on the user device." — Park teaches that the material information acquired from the trained encoder is returned as the result to the same computing device by which the targeted properties were supplied, and is made available there for the design and implementation of a resulting element (Park, paragraph [0057]: "targeted wave properties are input to the trained encoder, and the result is acquired to derive information on a material from the given wave properties"; Park, paragraph [0058]: "A wave-based active signal processing element, a waveguide, an antenna, a display, a solar cell, an electronic circuit signal processing element, etc. may be designed and implemented on the basis of the derived material information"; Park, paragraph [0039]: "The encoder 100, the decoder 200, and the data converter 300 may be configured as separate computing devices or may be configured such that functions thereof may be implemented in one computing device"). The return of the acquired result to the computing device from which the targeted properties were supplied reads on the recited displaying of the dimension information of the new metaunit on the user device.
Roberts, Cook, Ma, Parekh and Park are analogous to the claimed invention, Roberts, Ma and Park being from the same field of endeavor of machine-learning inverse design of a material structure from a target optical response, and Cook and Parekh being reasonably pertinent to the particular problems with which the inventor was concerned (MPEP § 2141.01(a)). Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to receive the target 2D chromaticity coordinate of Roberts from a user device and to display the resulting metaunit dimension information on that user device, as taught by Park. The motivation to combine Roberts, Cook, Ma, Parekh and Park is as recited by Park (paragraph [0057]: "targeted wave properties are input to the trained encoder, and the result is acquired to derive information on a material from the given wave properties"), such that a user supplies the desired color coordinate at a computing device and receives the corresponding metaunit dimensions in return at that same device, since Roberts expressly contemplates deployment of the trained inverse model in this manner (Roberts, page 061101-1, Abstract: "The tightly constrained inverse model allows for the instantaneous design of metasurfaces, given a desired color… making it suitable for commercial use"; page 061101-2, column 1, paragraph 3: "the development of a practical tool for the forward and inverse designs of the aforementioned structure").
Claim 10 recites the same limitations as Claim 5 and is therefore rejected under the same rationale set forth above with respect to Claim 5.
Claim 15 recites the same limitations as Claim 5 and is therefore rejected under the same rationale set forth above with respect to Claim 5.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to HUNG VAN LE whose telephone number is (571)270-0164. The examiner can normally be reached 8 a.m. - 5 p.m..
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, Cesar Paula can be reached at (571) 272-4128. 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.
/HUNG VAN LE/Examiner, Art Unit 2145
/CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights