DETAILED ACTION
Applicant's response, filed 20 July, has been fully considered. The following rejections and/or objections are either reiterated or newly applied. They constitute the complete set presently being applied to the instant application.
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 .
Claim Status
Claims 1-20 are pending.
Claims 1-20 are rejected.
Priority
This application has the effective filing date of 01 June 2022, with no claims to priority.
Information Disclosure Statement
The information disclosure statement (IDS) submitted on 31 August 2022 in compliance with the provisions of 37 CFR 1.97. Accordingly, the information disclosure statement has been considered by the examiner.
Drawings
The drawings, submitted 01 June 2022, have been accepted by the examiner.
Claim Rejections - 35 USC § 112
Applicant’s arguments, see page 1, para. 2, filed 20, with respect to inadequate written description are persuasive, in view of the claim amendments.
As such, the rejection to claims 9-16 under 35 U.S.C 112(a) is withdrawn.
Claim Rejections - 35 USC § 101
The previous rejection to claims 1-20 under 35 U.S.C 101 is maintained.
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 therefore, subject to the conditions and requirements of this title.
Claims 1-20 are rejected under 35 U.S.C. 101 because the claimed invention is directed to abstract ideas without significantly more for the reasons detailed below.
Eligibility Step 1: Subject matter eligibility evaluation in accordance with MPEP § 2106:
Claims 1-8 are directed to a statutory category (method).
Claims 9-16 is directed to a statutory category (product).
Claims 17-20 is directed to a statutory category (system).
Therefore, claims 1-20 have patent eligible subject matter.
[Eligibility Step 1: YES]
Eligibility Step 2A: This step determines whether a claim is directed to a judicial exception in accordance with MPEP § 2106.
Eligibility Step 2A -- Prong One: The limitations below are analyzed to determine if the claims recite any concepts that could equate to a judicial exception (i.e. abstract idea, law of nature, or natural phenomenon).
Claims 1, 9, and 17:
encoding a sequence and interrelationships among events occurring in a simulation and/or experiment in an event-sourced architecture for materials provenance (ESAMP) framework; (mathematical process)
learning an initial state of a generated material sample in the ESAMP framework; (mental process)
learning how one or more processes affect the state of the generated material sample in the ESAMP framework according to the state vector shared with the other material samples in the ESAMP framework (mental process)
identifying the generated final material sample a material discovery when a final material state is different from a predicted material state based on the ESAMP framework by detecting deviations between the predicted material state generated by the neural network model and the final material state. (mental process)
Claim 2, 10, 18: The method of claim 1, further comprising:
integrate a provenance information regarding how the material samples are created and what processes the generated material samples have undergone (mathematical process)
learning shared and different characteristics of the generated material samples based on the integrated provenance information (mental process)
Claims 3, 11, and 19: The method of claim 1, further comprising predicting a state change of the generated material sample as a selected process is applied to the generated material sample. (mental process)
Claims 7 and 15: The method of claim 1, in which encoding further comprises:
assembling an ESAMP database; (mathematical concept)
Claims 8 and 16: The method of claim 7, in which encoding further comprises:
analyzing of the raw process data from the ESAMP database to derive a state information of the raw process data from the ESAMP database; (mental process)
Step 2A – Prong One Analysis:
The limitations that recite encoding data equate to transforming data using mathematical functions. Assembling a database equates to further manipulating the secondary data by organizing the information into a new form. Integrating provenance information, given broadest reasonable interpretation, is directed to transforming the information and adding it to the database in the same method as the encoded events/relational data. As such, these claims describe processes that organize information through mathematical correlations and fall under the mathematical concepts grouping of abstract ideas.
Furthermore, the disclosure provides mathematical relationships and correlations to predict state changes with a neural network, such as a weighted sum of the feature vectors and classifying data based on patterns. Generating secondary data of this nature similarly falls under mathematical concepts grouping of abstract ideas.
Limitations that equate to making observations, connections, or conclusions based on the observable [0079] data properties stored within the table represent analysis techniques in the form of mental determinations of data.
These processes can be completed using nothing more than the human mind or pen/paper, and as such fall into the mental process grouping of abstract ideas.
The limitation of “identifying the generated final material sample a material discovery when a final material state is different from a predicted material state based on the ESAMP framework by detecting deviations between the predicted material state generated by the neural network model and the final material state” falls under mental processes as it reads on using the human mind to draw a conclusion based on the comparison of two observable data entries.
Additional elements include:
Claim 1, 9, and 17: sharing a state vector representing the initial state of the generated material sample with other generated material samples in the ESAMP framework to enable transfer learning across the generated material samples having a shared initial state;
training a neural network to predict state changes based on the shared state vectors.
applying the one or more processing to the generated material sample to generate a final material sample;
Claim 4, 12, and 20: The method of claim 1, further comprising training a neural network to predict a state change of the generated material sample after a selected process is applied to the generated material sample.
Claims 5 and 13: The method of claim 4, in which the neural network is trained to predict the state change of each of the generated material samples having a shared initial state vector.
Claims 7 and 15: storing, in the ESAMP database, provenance information regarding the creation of the generated material samples and processes undergone by each of the generated material samples.
Claims 8 and 16:
storing, in the ESAMP database, raw process data from processes run on the generated material samples;
storing, in the ESAMP database, the state information regarding the processes run on the generated material samples.
Step 2A – Prong Two Analysis:
The disclosure states that training the neural network model may occur through using observables and sharing state vectors with other material samples [0076]. It further states that there are output prediction functions that can be learned from the vectors [0077]. Therefore, in light of the specification, training the neural network to predict state changes (claims 1, 4, 9, 12, 17, and 20) requires accessing/retrieving data in the form of vectors within the database and transforming the vectors and observables into a prediction via a mathematical function.
Sharing vectors, given the broadest reasonable interpretation, allows entities stored within a database to access and retrieve data stored under different classifications within the same structure. This process equates to a mere data gathering activity that does not integrate the judicial exception into practical application per MPEP 2106.05(g).
The limitation of “applying one or more processing to the generated material sample to generate a final material sample” is evaluated for significantly more below:
As the limitation reads on a transformation of a generated material sample into a generated final sample, the examiner considers if the transformation is particular using the 5 relevant analysis factors provided by MPEP 2106.05 (c):
(1) The particularity or generality of the transformation: The transformation in this limitation is a “process” in which the specification discloses to be embodied by an event that occurs to one or more samples; for example, the process is associated with an experiment in a laboratory, such as annealing in a sample furnace or performing spectroscopic characterization [0053].
Since an event occurring to the sample includes those “occurring in a simulation and/or experiment” (claims 1, 9, and 17) without other meaningful limits, the transformation encompasses in silico processes and merely moving locations of the material. Therefore, the transformation is not particular enough to equate to significantly more.
(2) The degree to which the recited article is particular: The recited article in the limitation is a “generated material,” which does not place meaningful limits on the article that could be have the transformation applied; and is thus not particular.
(3) The nature of the transformation in terms of the type or extent of change in state or thing: The specification discloses the state is defined by two entities of the sample process that share the sample and do not have an entity of the sample process chronologically between the two entities of the sample process; and by managing the state under the most conservative assumption that every process alters the sample's state, any state equivalency rules (e.g., whether a certain type of process alters the state or not) may be applied in a transparent manner [0070].
As such, a “state”, as generically recited in the claims and in light of the specification, does not limit the process to result in the transformed article having a different function or use, needed to integrate the judicial exception or equate to significantly more.
(4) The nature of the article transformed: Though a “sample” can be embodied as a tangible/physical object or substance, the specification discloses: Some aspects of the present disclosure learn the initial state of a particular sample that is stored as experimental materials science data in a database [0049]. Therefore, the claims do not limit “material sample” enough to not provide significantly more.
(5) Whether the transformation is extra-solution activity or a field-of-use: Since the “process” and “sample” are generically recited, they do not place limits on the judicial exceptions to particular machines, manufactures, transformations, articles, or states and merely recite equivalents to “apply it” regarding the judicial exceptions (2106.04(d)).
Additional elements that may be categorized differently include:
Claims 6 and 14: sharing the state vector representing the initial state of the material sample with the other material samples in the ESAMP framework is performed for each of the material samples having the initial state.
This limitation specifies what type of data undergoes manipulation via analysis techniques. Activities of this nature are classified as insignificant extra solution activity and do not integrate the judicial exception into practical application per MPEP 210.05(g).
Additional elements that may be categorized differently include:
Claim 1: A method for neural network material state prediction, comprising:
Claim 9: non-transitory computer-readable medium having program code recorded thereon for neural network material state prediction, the program code being executed by a processor to cause the processor
Claim 17:
a neural processing unit (NPU);
a memory coupled to the NPU, and instructions stored in the memory and operable, when executed by the NPU, cause the system:
A memory and non-transitory computer readable medium qualify as components of generic computing environments/implementations of a method onto a generic computer environment. Though a neural processing unit can be seen as a specialized type of hardware, it does not contain the specificity required to qualify as a component of a particular machine per MPEP 2106.05(b).
Furthermore, these components, when viewed separately and in the context of a whole claimed invention merely act as tools to apply the judicial exceptions. As such, they do not integrate the judicial exception into practical application, as exemplified by Genetic Techs. Ltd., 818 F.3d at 1377; 118 USPQ2d at 1546.
As such, the additional elements, when viewed separately and in the context of whole do not integrate the judicial exceptions into practical application (Step 2A, Prong Two: NO) and therefore are directed to judicial exceptions.
[Eligibility Step 2A: YES]
Step 2B Analysis:
Wan et al. (Nano Letters; Vol. 19; 2019), reviews materials discovery and properties prediction, and provides evidence that the following techniques are well-understood, routine, and conventional: First, a large amount of experimental data is needed; second, based on data mining, second machine learning models could be built to search for or predict new materials with desired properties; and finally, if material properties meet the necessary requirements, the materials can be synthesized in terms of machine learning predictions (page 3393, column 2).
Wang et al. (IEEE Trans on Knowl Data Eng; Vol. 29: 12; 2017) reviews knowledge graphs; and provides evidence sharing features represented as vectors, is well-understood, routine, and conventional; and Schleder et al. (J. Phys. Mater; Vol. 2: 032001; 2019), which reviews recent machine learning approaches in materials science, establishes transfer learning as further well-understood, routine, and conventional in the field of materials science repositories that include deep learning.
Furthermore, the courts have found storing and retrieving information in memory, as enected by the generic computer components and vector sharing to be routine, well-understood, and conventional in the art as exemplified by Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015).
Selecting information, based on types of information and availability of information, for analysis is also well known, routine, and conventional within the art, as exemplified by Electric Power Group, LLC v. Alstom S.A., 830 F.3d 1350, 1354-55, 119 USPQ2d 1739, 1742 (Fed. Cir. 2016).
Though the neural processing unit represents a piece of specialized hardware that does improve the efficiency of the claimed invention, NPUs are well-known and conventional for increasing general computer efficiency as evidenced by Chen et al. (Engineering; Vol. 6 (3); 2020). Furthermore, including the NPU itself does not provide an inventive solution to the problem presented by materials state prediction and thus does not provide evidence of inventive concept per TLI Communications, 823 F.3d at 611-12, 118 USPQ2d at 1747.
[Eligibility Step 2B: NO]
As such claims 1-20 are directed to judicial exceptions and rejected under 35 U.S.C 101, in accordance with Alice/Mayo, MPEP 2143 evaluation.
Response to Arguments
Applicant's arguments, filed 20 July 2026, with respect to 35 U.S.C 101, have been fully considered but they are not persuasive.
Applicant argues “sharing a state vector representing the initial state of the generated material sample with other generated material samples in the ESAMP framework to enable transfer learning across the generated material samples having a shared initial state," "learning how one or more processes affect the state of the generated material sample in the ESAMP framework according to the state vector shared with the other generated material samples in the ESAMP framework by training a neural network model to predict state changes based on the shared state vectors" (page 1, para. 5), and "identifying the generated final material sample as a material discovery when a final material state is different from a predicted material state based on the ESAMP framework by detecting deviations between the predicted material state generated by the neural network model and the final material state" describe computerized operations involving neural network model training, state vector sharing across material samples for transfer learning, and automated deviation detection that are not practically performable in the human mind and do not fall within any of the enumerated groupings of abstract ideas (page 2, para. 1).
Examiner responds, ”sharing a state vector representing the initial state of the generated material sample with other generated material samples in the ESAMP framework to enable transfer learning across the generated material samples having a shared initial state” is evaluated as an additional element.
However, the remaining limitations read on comparisons and conclusions that can be practically performed in the human mind; and though such judicial exceptions are computer implemented, the courts do not distinguish between claims that recite mental processes performed by humans and claims that recite mental processes performed on a computer per MPEP 2106.04(a)(2).
Applicant argues the claims are not merely directed to data gathering activities as claim 1 recites a specific technical process in which a neural network model is trained using shared state vectors within an ESAMP framework to predict how processes change material states, and material discoveries are identified by detecting deviations between predicted and actual final material states, which constitutes a specific, practical application in the field of materials science that transforms raw experimental materials science data into actionable material discovery determinations (page 2, para. 3).
Examiner responds material discovery determinations, enacted by detecting deviations between the predicted and actual final material states are a judicial exception in the form of an abstract idea (mental process); and training a neural network model does not integrate such judicial exceptions into practical application via any consideration of MPEP 2106.05(d).
Applicant argues the specification discloses meaningful limits on how the neural network operates and thus the claims also impose such meaningful limits, rather than merely reciting a result; and are therefore not directed to a judicial exception under Step 2a, prong two (page 2, para. 4).
Examiner responds the specification does not provide limits meaningful enough for one of ordinary skill in the art to consider “training a neural network to predict state changes based on shared state vectors” an improvement to technology or the functioning of a computer and is therefore not eligible via the considerations set forth in MPEP 2106.04(d), regarding Step 2a, prong two.
Applicant argues The ordered combination of encoding event sequences in an ESAMP framework, learning initial material states, sharing state vectors to enable transfer learning across material samples having a shared initial state, training a neural network model to predict state changes based on shared state vectors, and detecting deviations between predicted and actual final material states to identify material discoveries is not well-understood, routine, or conventional in the field of art (page 3, para. 1).
Examiner responds encoding event sequences, learning initial material states, and detecting deviations between predicted and actual final material states to identify material discoveries appear to recite judicial exceptions (abstract ideas in the form of mathematical concepts and/or mental processes); and therefore, are not evaluated for inventive concept under Step 2b (MPEP 2106.05).
Though sharing a state vectors and training a neural network model intended to be used for state change prediction based on shared state vectors is an additional element, the step is evaluated in light of the specification which draws training a neural network to using observables and sharing state vectors with other material samples [0076], in which sharing vectors, given the broadest reasonable interpretation, allows entities stored within a database to access and retrieve data stored under different classifications within the same structure. Such step even when considered in combination with the other additional elements is found well-understood, routine, and conventional per Versata Dev. Group, Inc. v. SAP Am., Inc., 793 F.3d 1306, 1334, 115 USPQ2d 1681, 1701 (Fed. Cir. 2015) and Wang et al. (IEEE Trans on Knowl Data Eng; Vol. 29: 12; 2017), which reviews knowledge graphs techniques.
Claim Rejections - 35 USC § 103
Applicant’s arguments, that Puchala, Liu, Ji, and Gaultois individually and combined fail to teach or suggest the amended limitations (page 4, para. 4-5), have been fully considered and are persuasive.
As such, the previous rejections to claims 1-20 are withdrawn.
The following rejections are newly recited and necessitated by claim amendments.
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 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 1-16 are rejected under 35 U.S.C. 103 as being unpatentable over Puchala et al. (JOM; Vol. 68: 8; 2016; previously cited), in view of Gupta et al. (Nature Com; Vol. 12: 6595: 2021; newly cited), Steinlin et al. (2023/0271276; newly cited), Bruneel et al. (2021/0031304; newly cited), Gaultois et al. (APL Materials; Vol. 4, 053213; 2016, previously cited), Tran et al. (Mach Learn Sci Tech; Vol. 1: 025006; 2020; newly cited), Liu et al. (JOM; Vol. 3:3; 2017; previously cited) and in further view of Ji et al. (Associ Comp Linguistics; Vol. 1, 2015; previously cited).
Puchala describes Materials Commons, a platform that organizes experimental materials data into a searchable infrastructure (page 2035, column 1).
Claims 1 and 9 are directed to assembling and storing data pertaining generated materials and relationships with events occurring in a simulation or experiment into a relational database.
Puchala teaches The Materials Commons is a new materials information repository (page 2044, column 1); and all files and measurements in the system are versioned and relationships between data model objects are stored as provenance information which allows scientists to recover and view the state of their experiments at different points in their workflow (page 2041, column 2).
Claims 1 and 9 are further directed to learning an initial state of the generated sample.
Puchala teaches illustrating, in Fig. 3, a subset of possible results and provenance information for DFT calculations, cluster expansion effective Hamiltonian fitting, grand canonical Monte Carlo calculations, and phase field calculations represented in Materials Commons (page 2038, column 1). Puchala further shows storing the initial atomic configuration in the database (page 2038, fig. 3).
Claims 1 and 9 are further directed to sharing vectors representing the initial state of the generated material sample with other generated material samples in the database, intended to be used to enable transfer learning across the generated material samples having a shared initial state.
Puchala teaches as new measurements are added, they are immediately applied to all relevant samples, and as new samples are created, they can immediately be associated with existing measurements (page 2037, column 2).
Claims 1 and 9 are further directed to using the state vector to learn how one or more processes affect the generated material sample’s state, by training a neural network model to predict state changes based on the shared state vectors.
Puchala teaches using process models, that describe how processing affects materials’ structure and properties (page 2036, column 2).
Claims 1 and 9 are further directed to applying the one or more processes to the generated material sample to create a second generated material sample.
Puchala teaches researchers receives some material, section and heat-treats it for different amounts of time, and continually interact with the Materials Commons to store each individual step in the process (page 2037, column 1). Puchala further teaches users select a process from a list of process types and existing samples as input to the process, and create or upload new or transformed samples, files, and measurements as output from the process (page 2037, column 1).
Claims 1 and 9 are further directed to identifying the generated final material sample as a material discovery when the final material state is different from a predicted material state, based on the database.
Puchala teaches that such an information infrastructure will lead to acceleration in discovering and engineering new materials by ensuring that information is readily available and shared (page 2035, column 1); can be used to identify samples of the material of interest (page 2039, column 1); and is one essential element of the capability for accelerated predictive structural materials science (page 2036, column 2).
Claims 2 and 10 are directed to integrating provenance information regarding a generated material’s origin into the relational database.
Puchala teaches representing materials’ associated provenance in the model (page 2036, column 1), in which the provenance information conveys the creation of samples, datafiles, and measurements by processes (page 2038, fig. 2).
Claims 3 and 11 are directed to gathering information about a generated material’s property after a particular process is applied.
Puchala teaches users can select a process from a list of process types which creates a representation, such as shown in Fig. 2a, details the provenance of all objects in the data framework (page 2037, column 1); and for transformational processes, such as heat treatment, the list of attributes associated with the transformed sample is updated (page 2037, column 1).
Claim 6 is directed to sharing information regarding each generated materials’ initial property with other generated materials in the database.
Puchala teaches sharing information about a sample’s property ‘as received’, with other samples in the framework (page 2038, fig. 2); and as new measurements are added, they are immediately applied to all relevant samples (page 2037, column 2).
Claims 7 and 15 are directed to creating a database regarding the provenance of generated materials samples and storing information regarding their origin and subsequent processes.
Puchala teaches building a dataset, which includes an ordered collection of samples, processes, attributes, measurements, and datafiles (page 2037, column 1) that is intended to store each individual step in the process done to the sample and that the provenance information ensures that all original sources of data get proper credit in the curated project (page 2041, column 1), which requires the provenance data include sample origin information.
Claims 8 and 16 are directed to the database storing raw process data run on the generated samples, analyzing it to determine property information, then storing the subsequent results from analysis.
Regarding claim 8, Puchala teaches that as a researcher performs an experiment, completes analyses, and draws conclusions from it (via models), they should continually interact with the Materials Commons to store each individual step in the process (page 2037, column 1).
Regarding claims 9-16, Puchala further teaches that the Materials Commons software is as open-source code (page 2036, column 1) and therefore completes all the steps described previously through program code/machine-readable instructions executed via a processor.
Therefore, Puchala teaches a database capable of gathering provenance information about materials, learning the initial state of generated materials, and generating new materials based on the state of other generated materials stored in the database.
Puchala et al. does not teach enabling transfer learning across the generated material samples having a shared initial state (claims 1, 9, and 17).
Gupta et al. describes cross-property deep transfer learning framework for enhanced predictive analytics.
Gupta et al. teaches using a model pre-trained on the source dataset using only Elemental Fractions as the vector-based-materials representation for the model input, for transfer learning (TL) models (page 8, column 1); it is relevant to note that some of the recent deep learning models for predicting materials properties use some form of embedding with materials structure and/or composition-based information as input (page 7, column 2); thus one could try to build upon the proposed cross-property TL workflow by incorporating structure-based information into the workflow via vector-based structural attributes with fully connected deep neural networks (page 7, column 2); and the trained source models could then be used in a similar way to either extract robust features for the target datasets or directly fine-tune on the target datasets (page 7, column 2).
Gupta et al. teaches the results suggest that a pre-trained source model with a rich set of features learned on an extensive source dataset can be effectively used with the proposed cross-property TL framework to build enhanced models on small datasets because they can better help the target model to learn their respective properties with more robustness and accuracy by refining the knowledge and rich set of hierarchical features learned by the source model (page 7, column 2); and the so-called deep learning (DL) algorithms have shown a remarkable capability to automatically and efficiently extract features from raw inputs and build accurate models for different properties of materials, often surpassing traditional ML techniques, which has been made possible due to the increasing availability of large materials databases (page 2, column 1).
Therefore Gupta et al. teaches a materials science database that uses deep learning algorithms trained on vectorized representations of materials’ structure is sufficient to enable transfer learning for property prediction. As such, it would be obvious to one of ordinary skill in the art to apply the transfer learning technique to the method of Puchala et al. yielding predictable results and an improved system.
Though Puchala teaches using process models, that describe how processing affects materials’ structure and properties (page 2036, column 2); Puchala et al does not teach using the state vector to learn how one or more processes affect the generated material sample’s state, by training a neural network model to predict state changes based on the shared state vectors (claims 1, 9, and 17).
Steinlin et al. describes a method, control unit, and laser cutting system for combined path and laser process planning for highly dynamic teal-time systems
Steinlin et al. teaches sensors can sometimes take the form of spectral intensity sensors measuring properties of the melt such as temperature, viscosity, plasma, material properties [0023]; the sensor data can be used to calibrate the two models: the process model and/or the machine model [0024]; the process model and/or the machine model can preferably be implemented as a neural network [0046] to predictively calculate or estimate the resulting quality of the cutting process [0047] and cut part of the material [0109], in which all relationships between the input, output and state variables are represented in the form of matrices and vectors [0138].
Bruneel et al. describes methods for determining laser machining parameters.
Bruneel et al. teaches in order to ensure that the machining process is as accurate as possible, it is known to model the interaction parameters between the laser beam and the material to be machined [0004], wherein machining includes as engraving, cutting, drilling, welding [0024]. Bruneel teaches when the machining of a new material is decided upon, to enrich a material database on the basis of the parameters, specific to the machined material and the results observed after they have been characterized and analyzed by the detection means and the central unit according to the invention [0060].
Bruneel et al. further teaches in particular, an advantage of another embodiment of the invention is to enable supervised machine learning on the basis of the physical parameters of interaction between the laser beam and the material to be machined during the machining, during previous machining experiments or from a database including information on a laser source or a material to be machined [0022]; and examples of machine learning for supervised learning algorithms are for example [0011], individually or in combination: [0112] the linear regression, [0113] k-nn, [0114] the Support Vector Machine (SVM), [0115] the neural networks [0116] and random forests.
Therefore Puchala et al. teaches learning how one or more processes affect the material samples with a process model; Steinlin et al. teaches completing the same process by training a neural network implemented process model to predict changes to a material due to a particular process; and measuring and storing material properties in a process model that uses state variables represented as vectors. Bruneel et al. provides motivation for one of ordinary skill in the art to combine such a process with a materials science database. As such, it would be obvious to one of ordinary skill in the art to apply the technique of Steinlin et al. to the method of Puchala et al. yielding predictable results and an improved system, based on the teachings of Bruneel et al.
Puchala et al. does not teach identifying the generated final material sample as a materials discovery when a final material state is different from a predicted material state based on the database (claims 1, 9, and 17).
Gaultois et al. describes web-based machine learning models for real-time screening of thermoelectric materials properties.
Gaultois et al. teaches illustrating the remarkable chemical homogeneity of most thermoelectric materials investigated to date by plotting each material from a thermoelectric database on the periodic table, based on the composition-weighted average of the positions of elements in the material, in which the tight cluster of previously investigated chemistries is, as expected, dominated by chalcogenides and p-block elements such as Sn and Sb (page 3, column 1).
Gaultois et al. teaches in contrast, we also show the positions of Gd12Co5Bi and Er12Co5Bi, materials derived from our recommendation engine, which we characterize as a new class of thermoelectrics in this work; in sharp contrast to thermoelectric compounds investigated to date (page 3, column 1).
Gaultois et al. further teaches for each range of thermoelectric property, the engine gives a confidence score between 0% and 100% that a given material’s measured value for that property at room temperature will fall within the targeted range; and classifies any material for which the answer to all these questions is likely “yes” as a potentially promising thermoelectric that may warrant further study (page 4, column 1).
Gaultois et al. teaches the objective of this recommendation engine is to directly enable experimental researchers to rapidly identify new materials, such as RE12Co5Bi, that are very distinct from known compound classes, and worthy of further study (page 3, column 1) and augment the chemical intuition of experimental researchers working on materials discovery, but not definitively identify record-setting compounds, which remains open challenges for future work (page 4, column 1).
Therefore Gaultois et al. teaches using a materials science database in combination with a predictive algorithm to compare expected properties to predicted properties for the identification of materials discovery, when they are different. As such, one of ordinary skill in the art has sufficient motivation to substitute the materials science database, taught by Gaultois et al., with the another, designed to aid synthesis, engineering, and identification of new materials, with a reasonable expectation of success, as taught by Puchala (page 2035, column 1) (page 2039, column 1).
Puchala et al. in view of Gaultois et al. do not teach completing the identification process by detecting deviations between the predicted material state generated by the neural network model and the final material state (claims 1, 9, and 17).
Gaultois et al. further teaches the purpose of our recommendation engine is thus neither to make quantitative predictions of these thermoelectric properties nor to definitively identify record-setting compounds -- these remain open challenges for future work (page 4, column 1).
Tran et al. describes methods for comparing uncertainty quantifications for material property predictions.
Tran et al. teaches the fields of catalysis and materials science are burgeoning with methods to screen, design, and understand materials (page 1, column 1); this research has spurned the creation of Machine Learning (ML) models to predict various material properties (page 1, column 1); under-sampling issues can limit the training data and therefore the predictive power of the models (page 1, column 1); therefore it would be helpful to have an uncertainty quantification (UQ) for a model so that we know when to trust the predictions and when not to; more specifically: UQ would enable various online, active frameworks for materials discovery and design (page 1, column 1).
Tran et al. teaches an overview of the various UQ methods we investigated in this study include: ΔE represents DFT-calculated adsorption energies; represents ML-predicted adsorption energies; UQ represents ML-predicted uncertainty quantifications; µ represents the mean of a sample of points; σ represents the standard deviation of a sample of points;
PNG
media_image1.png
8
9
media_image1.png
Greyscale
represents the residuals between DFT and ML; and e hat represents the residuals between ML-predicted
PNG
media_image1.png
8
9
media_image1.png
Greyscale
and the actual
PNG
media_image1.png
8
9
media_image1.png
Greyscale
(page 4, fig. 2).
Therefore Gaultois et al. teaches aiding material discovery by identifying the generated final material state as different from the predicted material state, utilizing a materials science database; and further provides motivation for one of ordinary skill in the art to quantitatively predict the properties and definitively identify new materials. Tran et al. teaches uncertainty quantification methods suitable for material discovery that include detecting deviations between the predicted and final material state. As such, it would be obvious to one of ordinary skill in the art to combine the method of Tran et al. with the method of Puchala et al. in view of Gaultois et al., based on the motivation for uncertainty quantification provided by both Gaultois et al. and Tran et al., with a reasonable expectation of success, improvement to the system, and each element merely performing the same function as they do separately.
Puchala in view of Gaultois et al. and Tran et al. teach a materials science database that learns and predicts generated materials properties and provenance during every experimentation stage and applies the processes and prediction analysis in order to generate and identify materials discoveries.
Puchala in view of Gaultois et al. and Tran et al. do not teach using a neural network for material state prediction.
Liu describes methods of modelling quantitative-structure activity relationships (QSARS) and machine learning techniques that aid the materials discovery process.
Liu teaches that the regression and clustering algorithms (page 164, column 2), in neural networks (page 164, fig. 3), are the most suitable machine learning techniques for material property prediction tasks on the macro and micro levels (page 164, column 2).
Therefore, Liu provides sufficient motivation for one of ordinary skill in the art to apply a neural network for the prediction of a materials property (state) to the ‘big data’ framework – Materials Commons – taught by Puchala, with a reasonable expectation of success and improvement to the system.
Puchala in view of Gaultois et al. and Liu teach using a materials science framework and neural network that learns and predicts generated materials properties and provenance during every experimentation stage and applies the processes and prediction analysis in order to generate and identify materials discoveries.
Puchala in view of Gaultois et al. and Liu do not teach explicit use of vectors to store and share state, property, or process information within the database.
Ji describes knowledge graphs modelling via neural networks.
Ji teaches that knowledge graphs typically contain large amounts of structured data in the form of triplets (head entity, relation, tail entity), where relation models the relationship between the two entities (page 687, column 1).
Ji further teaches a task of the model is to encode every element, entity, and relation of a knowledge graph into a low-dimensional embedding vector space (page 687, column 2) and then learn each embedding (page 687, column 2).
Regarding claims 4, 5, and 20, Ji teaches knowledge graph completion is the ability to predict relations between entities based on existing triplets in the knowledge graph (page 687, column 2).
Ji further teaches that a neural network can capture the correlations between entities and relations via matrix operations, where parameters of the neural network are shared by all relations (page 689, column 2).
Therefore, Puchala teaches a structured data framework that can effectively store generated materials process, structure, provenance, and property relationships as an essential element within a predictive materials science data pipeline and means of engineering/generating new materials.
Gaultois et al. teaches combining such a materials database with the technique of comparing predicted material property with the expected material property, in order to identify material discoveries.
Liu teaches that Materials Commons is a dataset of materials properties that could provide a powerful impetus to accelerate materials discovery and design (page 161, column 1) and that combining such ‘big data’ frameworks with machine learning techniques can successfully resolve the difficulties of modelling the relationships between materials properties and complex physical factors (page 161, column 1).
Lastly, Ji teaches neural network techniques that can accurately represent complex entity and relation data. The method includes encoding, sharing, and learning vectors representing the data. As such, it would be obvious to one of ordinary skill in the art to automate the Materials Commons framework into a generic neural network capable of predicting the state/property changes of materials samples, according to the claimed invention, based on the teachings of Puchala, Liu, Gaultois et al., and Ji.
Claims 17-20 are rejected under U.S.C 103 as unpatentable over Puchala et al. (JOM; Vol. 68: 8; 2016; previously cited), in view of Gupta et al. (Nature Com; Vol. 12: 6595: 2021; newly cited), Steinlin et al. (2023/0271276; newly cited), Bruneel et al. (2021/0031304; newly cited), Gaultois et al. (APL Materials; Vol. 4, 053213; 2016, previously cited), Tran et al. (Mach Learn Sci Tech; Vol. 1: 025006; 2020; newly cited), Liu et al. (JOM; Vol. 3:3; 2017; previously cited) and Ji et al. (Associ Comp Linguistics; Vol. 1, 2015; previously cited), as applied to claims 1-16 above, and in further view of Fowers et al. (IEEE; 45th Annual International Symposium on Computer Architecture (ISCA); 2018; previously cited).
The combination of references, as applied to the claims above, teach a system with machine readable instructions, that when executed, cause a processor to assemble and store data pertaining generated materials and their respective experimental processes into a relational database; apply processes to the generated material sample to generate a final material sample; identify the generated material sample as a material discovery when a final material state is different from a predicted material state (claim 17); integrate provenance information (claim 18); and train a neural network (claim 20) to gather information about a material’s property after selecting a particular process it could undergo (claim 19).
Neither Puchala, which teaches the materials data infrastructure, nor Liu, which motivates one of ordinary skill in the art to convert the database to a neural network prediction model, explicitly teach use of a neural processing unit (NPU).
However, Liu teaches that there is an urgent need to develop intelligent and high-performance prediction models that can correctly predict the properties of materials at a low temporal and computational cost (page 164, column 2). Fowers describes Neural Processing Unit architecture for a large-scale deep learning platform. Fowers teaches that neural processing units (NPUs) provide execution of DNN models with low latency, high throughput, and high efficiency, and (2) flexibility to accommodate evolving state-of-the-art models (e.g., RNNs, CNNs, MLPs) without costly silicon updates (page 1, column 1).
Therefore, Fowers suggests integration of a neural processing unit to neural network architecture in order to address the temporal and computational limitations of the neural networks, explained by Liu. The change is equivalent to a simple combination of one known element (CPU) with another (NPU) to obtain predictable results (acceleration of neural prediction). One of ordinary skill in the art would have sufficient motivation to incorporate NPUs to their neural network system in order to address the known computational limitations of a general processor for the claimed neural network/deep learning task with a reasonable expectation of success.
Response to Arguments
Applicant argues the previously cited references do not teach the amended limitations (page 4, para. 4-6).
Examiner responds, the applicant’s arguments have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
Conclusion
No claims are currently allowed.
Correspondence
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Milana Thompson whose telephone number is (571)272-8740. The examiner can normally be reached Monday - Friday, 9:00-6:00 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, Karlheinz Skowronek can be reached at (571) 272-1113. 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.
/M.K.T./Examiner, Art Unit 1687
/Karlheinz R. Skowronek/Supervisory Patent Examiner, Art Unit 1687
Prosecution Insights