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 .
Specification
The title of the invention is not descriptive. A new title is required that is clearly indicative of the invention to which the claims are directed.
The following title is suggested: "METHOD FOR DETERMINING MANUFACTURING VARIABLE VALUES USING TRAINED MODEL BASED ON FUNCTIONAL PROPERTY SIGNAL FEATURE VALUES"
Claim Rejections - 35 USC § 112
The following is a quotation of 35 U.S.C. 112(b):
(b) CONCLUSION.—The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the inventor or a joint inventor regards as the invention.
The following is a quotation of 35 U.S.C. 112 (pre-AIA ), second paragraph:
The specification shall conclude with one or more claims particularly pointing out and distinctly claiming the subject matter which the applicant regards as his invention.
Claims 2, 3, 4, 5, 6, 7, 8, 9, 15, and 17 are rejected under 35 U.S.C. 112(b) or 35 U.S.C. 112 (pre-AIA ), second paragraph, as being indefinite for failing to particularly point out and distinctly claim the subject matter which the inventor or a joint inventor (or for applications subject to pre-AIA 35 U.S.C. 112, the applicant), regards as the invention.
Claim 2 recites the limitation "the prototype functional property signal". There is insufficient antecedent basis for this limitation in the claim. Claim 2 depends from claim 1, which recites a method comprising measuring a target functional property signal for a target sample, extracting target functional property feature values, and using a trained model to determine manufacturing variable values. Neither claim 1 nor claim 2 introduces "a prototype functional property signal" prior to the use of the definite article "the" in claim 2. The word "the" signals reference to a previously introduced element; however, no prototype functional property signal was previously introduced in the claim or its dependency chain. Therefore, a person of ordinary skill in the art cannot determine with reasonable certainty what "the prototype functional property signal" refers to. See MPEP § 2173.05(e). For purposes of examination, "the prototype functional property signal" is interpreted under BRI as any measurement data series associated with a functional property of a prototype sample.
Claim 3 recites the limitation "the prototype functional property feature values". There is insufficient antecedent basis for this limitation in the claim. Claim 3 depends from claim 2, which depends from claim 1. The dependency chain (claims 1–2) does not introduce "prototype functional property feature values." Claim 2 introduces "the prototype functional property signal" (which itself lacks antecedent basis as noted above), but does not introduce "prototype functional property feature values." Claim 1 introduces "target functional property feature values" but no corresponding prototype feature values. Because no prior recitation of "prototype functional property feature values" appears in the dependency chain, claim 3 is indefinite. See MPEP § 2173.05(e). For purposes of examination, "the prototype functional property feature values" is interpreted under BRI as any values characterizing functional property measurement signals for a prototype sample.
Claims 4, 5, and 15 each recite the limitation "the functional property features". There is insufficient antecedent basis for this limitation in any of these claims. Claim 4 depends from claim 1, which introduces "target functional property feature values" but not "functional property features" as a standalone term. "Functional property features" and "functional property feature values" are distinct terms: the former refers to feature types or signal characteristics, the latter to their measured values. The definite article "the" in "the functional property features" presupposes prior introduction of that exact term, which does not appear in claim 1. Claim 5 depends from claim 4 and inherits this deficiency. See MPEP § 2173.05(e). Claim 15 depends from claim 13, which introduces "functional property feature values" but not "functional property features." The same analysis applies: "the functional property features" lacks antecedent basis in the dependency chain of claim 15. See MPEP § 2173.05(e). For purposes of examination, "the functional property features" in claims 4, 5, and 15 is interpreted under BRI as the features of functional properties associated with the functional property feature values of the respective parent claim.
Claim 6 recites the limitations "the comparison between the prototype functional property feature values and target functional property feature values" and "the prototype functional property feature values" (appearing twice). Claim 6 depends from claim 1. First, "the comparison" lacks antecedent basis. Claim 1's method consists of measuring a target functional property signal, extracting target functional property feature values, and using a trained model to determine manufacturing variable values. No comparison is introduced in claim 1. The use of the definite article "the" in "the comparison" presupposes a prior introduction of "a comparison," which is absent from the dependency chain. See MPEP § 2173.05(e). Second, "the prototype functional property feature values" lacks antecedent basis. Neither claim 1 nor claim 6 introduces "prototype functional property feature values" or any prototype element. "The prototype functional property feature values" of claim 6 therefore has no antecedent basis in the dependency chain. See MPEP § 2173.05(e).
Claim 7 depends from claim 6 and inherits both deficiencies. While claim 7 is internally consistent (properly introducing "the weighted prototype functional property feature values" through the step of weighting), claim 7 cannot cure the antecedent basis deficiencies inherited from claim 6.
For purposes of examination: "the comparison" is interpreted under BRI as a comparison between prototype and target functional property feature values; "the prototype functional property feature values" is interpreted under BRI as values characterizing functional property measurement signals for a prototype sample.
Claim 8 recites the limitation "the prototype sample". There is insufficient antecedent basis for this limitation in the claim. Claim 8 depends from claim 1. Claim 1 introduces "a target sample" — not "a prototype sample." The specification at ¶[0028]–[0031] explicitly distinguishes prototypes from targets: a prototype is "a test sample that is intended to mimic a target," while a target serves as "an objective (e.g., gold standard, goal, etc.) for prototype evaluation." These are distinct sample types. "The prototype sample" cannot derive antecedent basis from "a target sample." No prototype sample is introduced in the dependency chain of claim 8. See MPEP § 2173.05(e). For purposes of examination, "the prototype sample" is interpreted under BRI as a test sample intended to mimic a target.
Claims 9 and 17 are each rejected under two independent grounds: (1) lack of antecedent basis for "training the model," and (2) logical inconsistency between the dependent and independent claims.
(1) Lack of Antecedent Basis: Claim 9 recites "wherein training the model comprises" and depends from claim 1. Claim 17 recites "wherein training the model comprises" and depends from claim 13. In both instances, the phrase "training the model" presupposes that a step of training the model has been introduced in the parent claim. However, claim 1 recites "a trained model" as a structural element used in the method — the model's training is presupposed as complete prior to use. Claim 1 does not introduce "training the model" as a method step. Claim 13 similarly recites "a trained model" without introducing a training step. The phrases "training the model" in claims 9 and 17 therefore lack antecedent basis. See MPEP § 2173.05(e).
(2) Logical Inconsistency: Independent claims 1 and 13 recite methods that use "a trained model" — a model already trained when the method is performed. Claims 9 and 17 respectively attempt to add model training steps as part of these methods. This creates an irreconcilable logical inconsistency: the claimed methods simultaneously use an already-trained model and purport to train it as part of the same method. A person of ordinary skill in the art cannot determine with reasonable certainty whether the training is performed as part of the claimed method or whether the method uses a pre-trained model. This internal contradiction renders claims 9 and 17 indefinite. See MPEP § 2173.05.
For purposes of examination: in claim 9, "training the model" is interpreted under BRI to encompass training the variable value model to predict functional property feature values from manufacturing variable values. In claim 17, "training the model" is interpreted under BRI to encompass training the characterization model by clustering training functional property feature values and training the model to select a cluster.
Double Patenting
The nonstatutory double patenting rejection is based on a judicially created doctrine grounded in public policy (a policy reflected in the statute) so as to prevent the unjustified or improper timewise extension of the “right to exclude” granted by a patent and to prevent possible harassment by multiple assignees. A nonstatutory double patenting rejection is appropriate where the conflicting claims are not identical, but at least one examined application claim is not patentably distinct from the reference claim(s) because the examined application claim is either anticipated by, or would have been obvious over, the reference claim(s). See, e.g., In re Berg, 140 F.3d 1428, 46 USPQ2d 1226 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046, 29 USPQ2d 2010 (Fed. Cir. 1993); In re Longi, 759 F.2d 887, 225 USPQ 645 (Fed. Cir. 1985); In re Van Ornum, 686 F.2d 937, 214 USPQ 761 (CCPA 1982); In re Vogel, 422 F.2d 438, 164 USPQ 619 (CCPA 1970); In re Thorington, 418 F.2d 528, 163 USPQ 644 (CCPA 1969).
A timely filed terminal disclaimer in compliance with 37 CFR 1.321(c) or 1.321(d) may be used to overcome an actual or provisional rejection based on nonstatutory double patenting provided the reference application or patent either is shown to be commonly owned with the examined application, or claims an invention made as a result of activities undertaken within the scope of a joint research agreement. See MPEP § 717.02 for applications subject to examination under the first inventor to file provisions of the AIA as explained in MPEP § 2159. See MPEP § 2146 et seq. for applications not subject to examination under the first inventor to file provisions of the AIA . A terminal disclaimer must be signed in compliance with 37 CFR 1.321(b).
The filing of a terminal disclaimer by itself is not a complete reply to a nonstatutory double patenting (NSDP) rejection. A complete reply requires that the terminal disclaimer be accompanied by a reply requesting reconsideration of the prior Office action. Even where the NSDP rejection is provisional the reply must be complete. See MPEP § 804, subsection I.B.1. For a reply to a non-final Office action, see 37 CFR 1.111(a). For a reply to final Office action, see 37 CFR 1.113(c). A request for reconsideration while not provided for in 37 CFR 1.113(c) may be filed after final for consideration. See MPEP §§ 706.07(e) and 714.13.
The USPTO Internet website contains terminal disclaimer forms which may be used. Please visit www.uspto.gov/patent/patents-forms. The actual filing date of the application in which the form is filed determines what form (e.g., PTO/SB/25, PTO/SB/26, PTO/AIA /25, or PTO/AIA /26) should be used. A web-based eTerminal Disclaimer may be filled out completely online using web-screens. An eTerminal Disclaimer that meets all requirements is auto-processed and approved immediately upon submission. For more information about eTerminal Disclaimers, refer to www.uspto.gov/patents/apply/applying-online/eterminal-disclaimer.
Claims 1-12 are rejected on the ground of nonstatutory double patenting as being unpatentable over claim 1-12 of U.S. Patent No. 11,823,070. Although the claims at issue are not identical, they are not patentably distinct from each other because claims 1-12 of the instant application are broadened versions of the corresponding claims of U.S. Patent No. 11,823,070. Independent Patent claim 1 recites all of the limitations of instant claim 1 plus additional limitations directed to measuring a prototype functional property signal, extracting prototype functional property feature values, and determining the manufacturing variable values based on a comparison between prototype and target functional property feature values. The entire scope of Patent claim 1 falls within the scope of instant claim 1, anticipating instant claim 1. Each of dependent Patent claims 2-12 recites a wherein/further-comprising clause that is text-identical to the corresponding wherein/further-comprising clause of instant claims 2-12; combined with the inherited anticipation of Patent claim 1 over instant claim 1, each Patent dependent claim 2-12 anticipates the corresponding instant dependent claim (see MPEP § 804 II.B.2; In re Berg, 140 F.3d 1428 (Fed. Cir. 1998); In re Goodman, 11 F.3d 1046 (Fed. Cir. 1993)).
Instant Claim 1 ↔ U.S. Pat. No. 11,823,070 Claim 1
Instant Application Claims
U.S. Pat. No. 11,823,070 Claims
A method, comprising:
A method, comprising:
measuring a target functional property signal for a target sample;
measuring a target functional property signal for a target sample;
extracting target functional property feature values from the target functional property signal; and
extracting target functional property feature values from the target functional property signal;
(no corresponding limitation — instant is broader)
measuring a prototype functional property signal for a prototype sample, wherein the prototype sample is associated with a set of initial manufacturing variable values;
(no corresponding limitation — instant is broader)
extracting prototype functional property feature values from the prototype functional property signal;
using a trained model, determining a set of manufacturing variable values based on the target functional property feature values.
using a trained model, determining a set of manufacturing variable values based on a comparison between the prototype functional property feature values and target functional property feature values.
Instant Claim 2 ↔ U.S. Pat. No. 11,823,070 Claim 2
The method of Claim 1, wherein the prototype functional property signal and the target functional property signal each comprise a data time series.
The method of claim 1, wherein the prototype functional property signal and the target functional property signal each comprise a data time series.
Instant Claim 3 ↔ U.S. Pat. No. 11,823,070 Claim 3
The method of Claim 2, wherein the prototype functional property feature values and target functional property feature values are each extracted using time series decomposition.
The method of claim 2, wherein the prototype functional property feature values and target functional property feature values are each extracted using time series decomposition.
Instant Claim 4 ↔ U.S. Pat. No. 11,823,070 Claim 4
The method of Claim 1, wherein the functional property features comprise non-semantic features.
The method of claim 1, wherein the functional property features comprise non-semantic features.
Instant Claim 5 ↔ U.S. Pat. No. 11,823,070 Claim 5
The method of Claim 4, wherein the functional property features further comprise semantic features.
The method of claim 4, wherein the functional property features further comprise semantic features.
Instant Claim 6 ↔ U.S. Pat. No. 11,823,070 Claim 6
The method of Claim 1, wherein the comparison between the prototype functional property feature values and target functional property feature values comprises a distance between the prototype functional property feature values and the target functional property feature values.
The method of claim 1, wherein the comparison between the prototype functional property feature values and target functional property feature values comprises a distance between the prototype functional property feature values and the target functional property feature values.
Instant Claim 7 ↔ U.S. Pat. No. 11,823,070 Claim 7
The method of Claim 6, further comprising weighting the prototype functional property feature values and the target functional property feature values, wherein the distance comprises a distance between the weighted prototype functional property feature values and the weighted target functional property feature values.
The method of claim 6, further comprising weighting the prototype functional property feature values and the target functional property feature values, wherein the distance comprises a distance between the weighted prototype functional property feature values and the weighted target functional property feature values.
Instant Claim 8 ↔ U.S. Pat. No. 11,823,070 Claim 8
The method of Claim 1, further comprising measuring a binary characteristic of the prototype sample, wherein the set of manufacturing variable values is determined further based on the binary characteristic.
The method of claim 1, further comprising measuring a binary characteristic of the prototype sample, wherein the set of manufacturing variable values is determined further based on the binary characteristic.
Instant Claim 9 ↔ U.S. Pat. No. 11,823,070 Claim 9
The method of Claim 1, wherein training the model comprises:
The method of claim 1, wherein training the model comprises:
measuring a training functional property signal for a training sample, wherein the training sample is associated with a set of training manufacturing variable values;
measuring a training functional property signal for a training sample, wherein the training sample is associated with a set of training manufacturing variable values;
extracting training functional property feature values from the training functional property signal; and
extracting training functional property feature values from the training functional property signal; and
training the model to predict the training functional property feature values based on the set of training manufacturing variable values.
training the model to predict the training functional property feature values based on the set of training manufacturing variable values.
Instant Claim 10 ↔ U.S. Pat. No. 11,823,070 Claim 10
The method of Claim 1, wherein the model comprises an encoder trained to encode functional property feature values and manufacturing variable values.
The method of claim 1, wherein the model comprises an encoder trained to encode functional property feature values and manufacturing variable values.
Instant Claim 11 ↔ U.S. Pat. No. 11,823,070 Claim 11
The method of Claim 1, wherein the target functional property signal comprises a measurement for at least one of: texture, melt, or flavor.
The method of claim 1, wherein the target functional property signal comprises a measurement for at least one of: texture, melt, or flavor.
Instant Claim 12 ↔ U.S. Pat. No. 11,823,070 Claim 12
The method of Claim 1, wherein the target sample comprises a dairy product.
The method of claim 1, wherein the target sample comprises a dairy product.
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 1-19 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more. The analysis of the claims will follow the 2019 Revised Patent Subject Matter Eligibility Guidance (“2019 PEG”) and the precedential decision in Ex Parte Desjardins, Appeal No. 2024-000567 (PTAB Sept. 26, 2025, Appeals Review Panel Decision).
CLAIM 1
Step 1: Claim 1 recites “A method, comprising:” and is therefore directed to the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 1 is directed to an abstract idea. Specifically, claim 1 recites the following limitations that fall within the “mental processes” and “mathematical concepts” groupings of abstract ideas (MPEP 2106.04(a)(2)(I) and (III)):
“extracting target functional property feature values from the target functional property signal” — This limitation recites a mental process using observation, evaluation, and judgment, with the aid of pen and paper, in which a person (e.g., a sensory panelist or food scientist, as the specification acknowledges at ¶[0023]) examines a measurement signal and identifies characteristic feature values. It also recites a mathematical concept, namely the computation of feature values from a signal. See MPEP 2106.04(a)(2)(I) and (III).
“using a trained model, determining a set of manufacturing variable values based on the target functional property feature values” — This limitation recites a mental process of evaluation and judgment as to what manufacturing variables (e.g., ingredients, process parameters) should be selected to mimic the target, which a person of ordinary skill in food science would perform mentally based on observed feature values; the specification at ¶[0020] confirms that the determination involves “predicting…which of the potential ingredients, process parameters, and/or quantities or values thereof would result in a similar analog to a target product, based on the functional property feature values.” It also recites a mathematical concept — a mathematical mapping (the “trained model”) from a feature-value input vector to a variable-value output. See MPEP 2106.04(a)(2)(I) and (III).
Step 2A, Prong 2: Claim 1 recites the following additional elements:
“measuring a target functional property signal for a target sample” — This limitation is mere data gathering / pre-solution activity that provides the input data on which the judicial exception operates and does not impose meaningful limits on the abstract idea. See MPEP 2106.05(g)
• “a trained model” — This limitation is recited at a high level of generality and recites use of a generic class of computer algorithms to perform the corresponding abstract idea limitations. The specification at ¶[0046] confirms that the model “can include or leverage: regression … classification, neural networks (e.g., CNN, DNN, CAN, LSTM, RNN, autoencoders, etc.), rules, heuristics, equations … and/or any other suitable method or model,” expressly identifying many interchangeable generic ML options. Mere recitation that a judicial exception is to be performed using generic computer equipment and/or a generic class of computer algorithms in their ordinary capacity cannot meaningfully integrate the judicial exception into a practical application. See MPEP 2106.05(f).
Examiner has evaluated the specification under Ex Parte Desjardins for any disclosed improvement to the functioning of a computer or to other technology. The specification at ¶[0022]–[0025] describes purported advantages of comparing non-semantic feature values for food-prototype evaluation: a more accurate representation of subjective sensory adjacency relative to conventional sensory panels (¶[0023]); a smaller training dataset / increased prediction accuracy (¶[0024]); and reduced computational complexity through feature selection (¶[0025]). These purported advantages are improvements in the methodology of food-analog evaluation — i.e., improvements in the food science field — not improvements to how the machine learning model itself operates and not improvements to any underlying computer technology. Unlike the disclosure credited in Ex Parte Desjardins (which described overcoming the technological problem of “catastrophic forgetting” in continual learning systems by training the model to learn new tasks while protecting prior-task knowledge, with corresponding reductions in storage and complexity), the present specification identifies no improvement that is intrinsic to the operation of the machine learning model itself.
Even assuming arguendo that the specification described a technological improvement, the claim itself does not reflect such an improvement. Claim 1 recites “a trained model” at the highest level of generality, without any limitation as to model architecture, training method, parameter adjustment, or technical operation. Unlike the structurally specific limitation credited in Desjardins (“adjust the first values of the plurality of parameters to optimize performance of the machine learning model on the second machine learning task while protecting performance of the machine learning model on the first machine learning task”), claim 1 recites no operational limitation tied to the specification’s purported improvements. The claim is broader than the disclosed improvement. See Intellectual Ventures I LLC v. Symantec Corp., 838 F.3d 1307, 1316 (Fed. Cir. 2016) (the claim must include the components or steps of the invention that provide the alleged improvement). Accordingly, claim 1 does not satisfy Step 2A, Prong 2.
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. The additional elements, taken alone or in combination, do not represent significantly more than the abstract idea itself. The additional element of “measuring a target functional property signal” being mere data gathering has been recognized by the courts as well-understood, routine, and conventional. See CyberSource Corp. v. Retail Decisions, Inc., 654 F.3d 1366, 1372 (Fed. Cir. 2011); OIP Techs., Inc. v. Amazon.com, Inc., 788 F.3d 1359, 1363 (Fed. Cir. 2015); MPEP 2106.05(d)(II). The specification at ¶[0045] further confirms that signal measurement is performed using conventional assay tools (differential scanning calorimeter, texture analyzer, rheometer, spectrophotometer, etc.). The additional element of “a trained model”, per specification at ¶[0046] expressly identifies a wide range of generic ML methods as interchangeable conventional options, which constitutes an express statement of the well-understood, routine, and conventional nature of the additional element per MPEP 2106.05(d)(I)(1). See also Alice Corp. v. CLS Bank Int’l, 573 U.S. 208, 225–26 (2014) (generic computer implementation of an abstract idea does not supply an inventive concept). The ordered combination of the additional elements (a measurement step followed by application of a generic trained model) does not add significantly more, but merely describes the implementation of the abstract idea using generic data-gathering and generic ML components. The claim limitations do not improve the functioning of the computer itself, nor do they improve any other technology or technical field. Accordingly, claim 1 does not satisfy Step 2B and is rejected under 35 U.S.C. 101.
CLAIM 2
Step 1: Claim 2 depends from claim 1 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 2 additionally recites: “the prototype functional property signal and the target functional property signal each comprise a data time series.” This added limitation is a descriptive characterization of the data on which the same mental-process / mathematical-concept steps of claim 1 operate; it is itself part of the abstract idea. See MPEP 2106.04(a)(2)(III). No new additional elements are introduced.
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 3
Step 1: Claim 3 depends from claim 2 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 3 additionally recites: “the prototype functional property feature values and target functional property feature values are each extracted using time series decomposition.” Time series decomposition is a mathematical operation that decomposes a signal into mathematical components (e.g., trend, seasonality, residual). This limitation is itself a further mathematical-concept refinement of the feature-extraction step and is part of the abstract idea. See MPEP 2106.04(a)(2)(I).
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 4
Step 1: Claim 4 depends from claim 1 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 4 additionally recites: “the functional property features comprise non-semantic features.” This added limitation is a descriptive characterization of the features used in the same mental-process / mathematical-concept steps of claim 1; it is itself part of the abstract idea. See MPEP 2106.04(a)(2)(III).
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 5
Step 1: Claim 5 depends from claim 4 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 5 additionally recites: “the functional property features further comprise semantic features.” This added limitation is a further descriptive characterization of the features used in the same mental-process / mathematical-concept steps and is itself part of the abstract idea. See MPEP 2106.04(a)(2)(III).
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 6
Step 1: Claim 6 depends from claim 1 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 6 additionally recites: “the comparison between the prototype functional property feature values and target functional property feature values comprises a distance between the prototype functional property feature values and the target functional property feature values.” Computing a distance between two sets of feature values is a mathematical concept (a mathematical relationship/calculation) per MPEP 2106.04(a)(2)(I), and is also a mental process — a person can compute or estimate such a distance with pen and paper. See MPEP 2106.04(a)(2)(III).
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 7
Step 1: Claim 7 depends from claim 6 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 7 additionally recites: “weighting the prototype functional property feature values and the target functional property feature values, wherein the distance comprises a distance between the weighted prototype functional property feature values and the weighted target functional property feature values.” Weighting values and computing a distance between weighted vectors are mathematical operations (mathematical calculations) and are themselves part of the mathematical-concept abstract idea. See MPEP 2106.04(a)(2)(I). No new additional elements are introduced.
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 8
Step 1: Claim 8 depends from claim 1 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 8 additionally recites: “…wherein the set of manufacturing variable values is determined further based on the binary characteristic.” This itself is part of the same mental-process / mathematical-concept determination step of claim 1 (judgment about manufacturing variables based on observed sample characteristics), expanded to include an additional input. See MPEP 2106.04(a)(2)(I) and (III).
Step 2A, Prong 2: Claim 8 recites the following additional elements: “measuring a binary characteristic of the prototype sample” — this limitation is mere data gathering / insignificant pre-solution activity per MPEP 2106.05(g). It does not impose meaningful limits on the judicial exception; it merely supplies an additional input on which the abstract idea operates. Accordingly, claim 8 does not satisfy Step 2A, Prong 2.
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. For the additional elements overlapping with claim 1, the same Step 2B reasoning as claim 1 applies. With respect to the new additional element of measuring a binary characteristic, this is well-understood, routine, and conventional data gathering. See CyberSource Corp. v. Retail Decisions, Inc., 654 F.3d 1366, 1372 (Fed. Cir. 2011); MPEP 2106.05(d)(II). The specification at ¶[0041] indicates that functional property values “can be a binary characteristic” determined “using an assay tool” — i.e., conventional measurement. The ordered combination does not amount to significantly more. Accordingly, claim 8 does not satisfy Step 2B and is rejected under 35 U.S.C. 101.
CLAIM 9
Step 1: Claim 9 depends from claim 1 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 9 additionally recites:
“extracting training functional property feature values from the training functional property signal” — This limitation recites a mental-process / mathematical-concept feature-extraction step, parallel to the corresponding step of claim 1)
“training the model to predict the training functional property feature values based on the set of training manufacturing variable values” — This is a mathematical procedure of iterative parameter adjustment to minimize prediction error. See MPEP 2106.04(a)(2)(I).
Step 2A, Prong 2: Claim 9 recites the following additional elements: “measuring a training functional property signal for a training sample, wherein the training sample is associated with a set of training manufacturing values” — this limitation is mere data gathering / insignificant pre-solution activity per MPEP 2106.05(g). It merely supplies additional training data on which the abstract mathematical training procedure operates and imposes no meaningful limit on the judicial exception. The Desjardins analysis is unchanged: while the specification (¶[0024] describes a smaller training dataset as a benefit, claim 9 recites no specific training operation, parameter adjustment, or architectural feature that would reflect any disclosed improvement to the operation of the model itself — it merely recites generic training (“training the model”) at a high level of generality. The claim does not reflect any disclosed technological improvement. Accordingly, claim 9 does not satisfy Step 2A, Prong 2.
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. For the additional elements overlapping with claim 1, the same Step 2B reasoning as claim 1 applies. With respect to the new additional element of measuring a training signal, this is well-understood, routine, and conventional data gathering. See CyberSource, 654 F.3d at 1372; MPEP 2106.05(d)(II). The training procedure itself is generic; the specification at ¶[0047] confirms that “models can be trained using self-supervised learning, semi-supervised learning, supervised learning, unsupervised learning, transfer learning, reinforcement learning, and/or any other suitable training method,” indicating that training methods are interchangeable and conventional. The ordered combination does not amount to significantly more. Accordingly, claim 9 does not satisfy Step 2B and is rejected under 35 U.S.C. 101.
CLAIM 10
Step 1: Claim 10 depends from claim 1 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 10 additionally recites: “an encoder trained to encode functional property feature values and manufacturing variable values” —The functional behavior of “encoding” feature values and variable values into a shared latent space is itself a mathematical operation — part of the mathematical-concept abstract idea. See MPEP 2106.04(a)(2)(I). The additional elements in claim 10 are those of claim 1 plus the new additional element “an encoder” (a generic ML architectural component).
Step 2A, Prong 2: Claim 10 recites the following additional elements: “an encoder” — this is recited at the same high level of generality as “a trained model” in claim 1. The claim recites no specific encoder architecture, no specific training procedure, no parameter-protection mechanism, and no other technical operation. Per MPEP 2106.05(f), mere recitation that a judicial exception is to be performed using a generic ML component does not integrate the judicial exception into a practical application. The Desjardins analysis is unchanged: the specification at ¶[0046] lists “encoders” as one of many interchangeable generic ML options without identifying any operational improvement specific to encoders, and the claim recites no limitation tied to any purported improvement. See Intellectual Ventures I v. Symantec, 838 F.3d 1307, 1316 (Fed. Cir. 2016).
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. For the additional elements overlapping with claim 1, the same Step 2B reasoning as claim 1 applies. With respect to the new additional element “an encoder,” the specification at ¶[0046] expressly lists “encoders” as one of many interchangeable generic ML techniques the model “can include or leverage,” alongside regression, classification, neural networks (CNN, DNN, etc.), Bayesian methods, support vectors, decision trees, and many others. This express specification statement of interchangeable conventional options satisfies the Berkheimer requirement of an express specification statement of WURC nature per MPEP 2106.05(d)(I)(1). The ordered combination does not amount to significantly more. Accordingly, claim 10 does not satisfy Step 2B and is rejected under 35 U.S.C. 101.
CLAIM 11
Step 1: Claim 11 depends from claim 1 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: There are no additional abstract idea limitations.
Step 2A, Prong 2: Claim 11 recites the following additional elements: “the target functional property signal comprises a measurement for at least one of: texture, melt, or flavor” — This limitation merely specifies the subject of the measurement and therefore characterizes the data on which the same mental-process / mathematical-concept steps of claim 1 operate; it does not add any non-abstract activity. See MPEP 2106.04(a)(2)(III). The additional elements in claim 11 are those of claim 1 plus the new additional element of a field-of-use limitation that the measurement be of texture, melt, or flavor. The judicial exception is not integrated into a practical application. Specifying that the measurement be of texture, melt, or flavor is a field-of-use limitation that generally links the use of the judicial exception to a particular technological environment without limiting how the abstract idea is applied. See MPEP 2106.05(h). Such a limitation does not integrate the judicial exception into a practical application. Accordingly, claim 11 does not satisfy Step 2A, Prong 2.
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. For the additional elements overlapping with claim 1, the same Step 2B reasoning as claim 1 applies. Generally linking the abstract idea to a particular technological environment or field of use does not add significantly more. “Generic computer functions to execute an abstract idea, even when limiting the use of the idea to one particular environment, does not add significantly more” (MPEP 2106.05(h)). See Bilski v. Kappos, 561 U.S. 593, 612 (2010). The specification at ¶[0039]–[0040] further describes texture, melt, and flavor as among many conventional functional property categories measured using conventional assay tools (¶[0045]). Accordingly, claim 11 does not satisfy Step 2B and is rejected under 35 U.S.C. 101.
CLAIM 12
Step 1: Claim 12 depends from claim 1 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: There are no additional abstract idea limitations.
Step 2A, Prong 2: Claim 12 recites the following additional elements: “the target sample comprises a dairy product.” This limitation specifies the subject matter of the data on which the same abstract idea operates. See MPEP 2106.04(a)(2)(III). The additional elements in claim 12 are those of claim 1 plus the new additional element of a field-of-use limitation that the target sample be a dairy product. Specifying that the target sample be a dairy product is a field-of-use limitation that generally links the use of the judicial exception to a particular field of use (dairy products) without limiting how the abstract idea is applied. See MPEP 2106.05(h). It does not integrate the judicial exception into a practical application. Accordingly, claim 12 does not satisfy Step 2A, Prong 2.
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. For the additional elements overlapping with claim 1, the same Step 2B reasoning as claim 1 applies. Generally linking the abstract idea to a particular field of use does not provide significantly more. See MPEP 2106.05(h); Bilski v. Kappos, 561 U.S. 593, 612 (2010). Accordingly, claim 12 does not satisfy Step 2B and is rejected under 35 U.S.C. 101.
CLAIM 13
Step 1: Claim 13 recites “A method, comprising:” and is therefore directed to the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 13 is directed to an abstract idea. Specifically, claim 13 recites the following limitations that fall within the “mental processes” and “mathematical concepts” groupings of abstract ideas (MPEP 2106.04(a)(2)(I) and (III)):
“extracting functional property feature values from the functional property signal” — This limitation recites a mental process (a person can observe a measurement signal and identify characteristic feature values) and a mathematical concept (computing feature values from a signal). See MPEP 2106.04(a)(2)(I) and (III).
“determining a sample classification for the sample based on the functional property feature values, using a trained model” — This limitation recites a mental process of evaluation and judgment (categorizing a sample into a class such as “dairy” vs. “non-dairy,” as the specification confirms at ¶[0052]) and a mathematical concept (a mathematical model mapping feature inputs to a class output). See MPEP 2106.04(a)(2)(I) and (III).
“determining a set of updated variable values based on the sample classification” — This limitation recites a mental process of judgment about how manufacturing variables should be updated based on the sample classification, which a person of ordinary skill in food science would perform mentally. See MPEP 2106.04(a)(2)(III). It is also a mathematical concept (a mapping from a class input to a variable-value output). See MPEP 2106.04(a)(2)(I).
Step 2A, Prong 2: Claim 13 recites the following additional elements:
The additional elements of claim 13 are:
“measuring a functional property signal for a sample, wherein the sample is manufactured according to a set of variable values” — This limitation is mere data gathering / pre-solution activity. The clause “wherein the sample is manufactured according to a set of variable values” merely describes how the sample was prepared and does not impose meaningful limits on the judicial exception. See MPEP 2106.05(g).
“a trained model” — This limitation is recited at a high level of generality and recites use of a generic class of computer algorithms to perform the abstract idea. See MPEP 2106.05(f). Accordingly, claim 13 does not satisfy Step 2A, Prong 2.
Examiner has evaluated the specification under Ex Parte Desjardins for any disclosed improvement to the functioning of a computer or to other technology. For the same reasons discussed in claim 1, the specification at ¶[0022]–[0025] (pages {5}–{6}) describes only improvements to the methodology of food-analog evaluation — i.e., improvements in the food science field — not improvements to how the machine learning model itself operates or to any underlying computer technology. Unlike the disclosure credited in Desjardins (overcoming “catastrophic forgetting” by training the model to protect prior-task knowledge, with reduced storage and complexity), the specification here identifies no improvement intrinsic to the operation of the model itself. Even assuming arguendo a disclosed technological improvement, claim 13 recites “a trained model” generically without any limitation as to architecture, training procedure, parameter adjustment, or technical operation. The claim does not reflect any disclosed technological improvement. See Intellectual Ventures I v. Symantec, 838 F.3d 1307, 1316 (Fed. Cir. 2016). See MPEP 2106.05(f). Accordingly, claim 13 does not satisfy Step 2A, Prong 2.
Step 2B: The claim does not include additional elements that amount to significantly more than the judicial exception. • “measuring a functional property signal” — Mere data gathering is recognized as WURC. See CyberSource Corp. v. Retail Decisions, Inc., 654 F.3d 1366, 1372 (Fed. Cir. 2011); OIP Techs., 788 F.3d at 1363; MPEP 2106.05(d)(II). • “a trained model” — The specification at ¶[0046] expressly identifies a wide range of generic ML methods as interchangeable conventional options, an express specification statement of WURC nature per MPEP 2106.05(d)(I)(1). See also Alice Corp. v. CLS Bank Int’l, 573 U.S. 208, 225–26 (2014). The ordered combination of these additional elements does not amount to significantly more, but merely describes the implementation of the abstract idea using generic data-gathering and generic ML. The claim limitations do not improve the functioning of the computer itself, nor do they improve any other technology or technical field. Accordingly, claim 13 does not satisfy Step 2B and is rejected under 35 U.S.C. 101.
CLAIM 14
Step 1: Claim 14 depends from claim 13 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 14 additionally recites: “the functional property signal comprises a data time series, wherein the functional property feature values are extracted using time series analysis.” Time-series analysis is a mathematical operation; this limitation is itself part of the mathematical-concept abstract idea. See MPEP 2106.04(a)(2)(I). No new additional elements are introduced.
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 15
Step 1: Claim 15 depends from claim 13 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 15 additionally recites: “the functional property features comprise non-semantic features.” This added limitation is a descriptive characterization of the features used in the same mental-process / mathematical-concept steps of claim 13 and is itself part of the abstract idea. See MPEP 2106.04(a)(2)(III). No new additional elements are introduced.
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 16
Step 1: Claim 16 depends from claim 13 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 16 additionally recites: “the model is trained using training data comprising training functional property feature values labeled with associated sample classifications.” Supervised training using labeled feature-value data is itself a mathematical training procedure and part of the mathematical-concept abstract idea. See MPEP 2106.04(a)(2)(I). No new additional elements are introduced.
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 17
Step 1: Claim 17 depends from claim 13 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 17 additionally recites: “training the model comprises clustering training functional property feature values into a set of clusters, wherein determining the sample classification for the sample is determined by using the model to select a cluster from the set of clusters based on the functional property feature values for the sample.” Clustering values and selecting a cluster based on feature inputs are mathematical operations — part of the mathematical-concept abstract idea. See MPEP 2106.04(a)(2)(I). They are also mental processes (a person can mentally group similar items and assign new items to groups). See MPEP 2106.04(a)(2)(III). No new additional elements are introduced.
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 18
Step 1: Claim 18 depends from claim 13 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 18 additionally recites: “the model is trained using adversarial machine learning methods.” Adversarial machine learning is a class of mathematical training procedures. This limitation merely identifies a category of training algorithm and is itself part of the mathematical-concept abstract idea. See MPEP 2106.04(a)(2)(I). No new additional structural elements are introduced.
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
CLAIM 19
Step 1: Claim 19 depends from claim 13 and recites a method falling within the statutory category of a process. See MPEP 2106.03.
Step 2A, Prong 1: Claim 19 additionally recites that “the set of updated variable values are determined using explainability methods applied to the trained model.” “Explainability methods” — enumerated by the specification at ¶[00125] (e.g., LIME, SHAP, DeepLift, partial dependence plots, etc.) — are mathematical/statistical analyses applied to a model. This limitation is itself a further mathematical-concept refinement of the abstract “determining updated variable values” step. See MPEP 2106.04(a)(2)(I). No new structural additional elements are introduced.
Step 2A, Prong 2 and Step 2B: There are no more additional elements, therefore the analysis from the parent claim is maintained.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1, 4, 5, 9, 12, 13 and 15-17 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by US Pat. Pub. No. 2020/0365053 to Pichara et al. (hereinafter Pichara).
Per claim 1, Pichara discloses A method (Pichara: ¶0029 and Figures…Pichara describes a computer-implemented method that uses artificial intelligence to mimic a target food item, "Systems and methods to mimic a target food item using artificial intelligence are disclosed. The system can learn from open source and proprietary databases. A prediction model can be trained using features of the source ingredients to match those of the given target food item"), comprising:
measuring a target functional property signal for a target sample (Pichara: ¶0020…Pichara discloses using Near-Infrared (NIR) spectroscopy to identify physical and chemical features of a target food item, where NIR spectroscopy produces a spectral signal from a target sample, which constitutes measuring a target functional property signal for a target sample, "Near-Infrared (NIR) spectroscopy techniques may be used to identify physical and/or chemical features of the ingredients"; pg. 10…"identifying a set of data features associated with the given animal-based food item ... the set of data features for the given animal-based food item comprising at least one of physiochemical data features, nutritional data features, and molecular data features");
extracting target functional property feature values from the target functional property signal (Pichara: ¶0025 and pg. 10…Pichara discloses identifying the set of data features (physiochemical, nutritional, molecular) of the target food item and storing them in a target ingredients database, which constitutes extracting target functional property feature values because the disclosed data features are feature values derived from the NIR/spectral characterization of the target, "identifying a set of data features associated with the given animal-based food item, wherein a target ingredients database is configured to store the set of data features for the given animal-based food item"; ¶0026…"feature compression techniques such as kernel principal component analysis (KPCA) and/or auto-encoding"); and
using a trained model, determining a set of manufacturing variable values based on the target functional property feature values (Pichara: pg. 10…Pichara discloses training a machine learning prediction model and then using the trained model to determine a formula combining source ingredients in specific proportions to match the target's features, where the ingredient identities and proportions are the manufacturing variable values, "training a machine learning prediction model by using the first training set ... to select potential candidates from the plurality of plant-based ingredients to match data features ... to the identified set of data features associated with the given animal-based food item ... determining a formula, using the trained machine learning prediction model, to combine the potential candidates from the plurality of plant-based ingredients in specific proportions"; ¶0016…"Once the prediction model is trained, the most important features used for the prediction may be selected as the potential candidates to mimic the target food item").
Per claim 4, Pichara discloses claim 1, further disclosing wherein the functional property features comprise non-semantic features (Pichara: ¶0016… Pichara teaches that the prediction algorithms apply feature compression techniques such as kernel principal component analysis (PCA) and auto-encoding which produce a compact, latent representation of the input features, and such latent/compressed features are non-human-interpretable and lack a direct physical analog, i.e., non-semantic features, "the prediction algorithms may use feature compression techniques such as kernel principal component analysis and/or auto-encoding for training the prediction model").
Per claim 5, Pichara discloses claim 4, further disclosing wherein the functional property features further comprise semantic features (Pichara: ¶0025…Pichara additionally teaches that the system uses sensorial descriptors such as flavor, color, texture, and taste, which are intrinsically human-interpretable and therefore are semantic features, "sensorial descriptors (e.g., flavor, color, texture or taste)").
Per claim 9, Pichara discloses claim 1, further disclosing wherein training the model comprises: measuring a training functional property signal for a training sample, wherein the training sample is associated with a set of training manufacturing variable values; extracting training functional property feature values from the training functional property signal; and training the model to predict the training functional property feature values based on the set of training manufacturing variable values (Pichara: fig. 9 and pg. 10…Pichara explicitly teaches that, for each of a plurality of training ingredients (each associated with a set of variable values — the ingredient identity and proportions), the system identifies the ingredient's data features via NIR spectroscopy (the training signal), stores the data features in a source ingredients database (extraction), and then trains the prediction model on this corpus by mapping the features to the ingredient identities — i.e., the model is trained on (feature value, ingredient/variable value) pairs, "identifying a respective set of data features for each of a plurality of plant-based ingredients, the respective set of data features comprising at least one of physiochemical data features, nutritional data features ... creating a first training set comprising the compact representation of the respective set of data features for each of the plurality of plant-based ingredients; training a machine learning prediction model by using the first training set ... wherein a source ingredients database is configured to store the respective set of data features for each of the plurality of plant-based ingredients").
Per claim 12, Pichara discloses claim 1, further disclosing wherein the target sample comprises a dairy product (Pichara: ¶0014…Pichara teaches that the target food item can be a dairy product (e.g., cow milk, butter, cheese, yogurt, ice cream), which constitutes the target sample comprises a dairy product, "animal-based food item can be a food product that includes any animal-based ingredient, such as cow milk and all dairy products"; ¶0019 — "animal-based ingredients may include dairy products (e.g., milk, butter, cheese, yogurt, ice cream, etc.)").
Per claim 13, Pichara discloses A method (Pichara: ¶0029 and Figures…Pichara describes a computer-implemented method that uses artificial intelligence to mimic a target food item, "Systems and methods to mimic a target food item using artificial intelligence are disclosed. The system can learn from open source and proprietary databases. A prediction model can be trained using features of the source ingredients to match those of the given target food item."), comprising:
measuring a functional property signal for a sample, wherein the sample is manufactured according to a set of variable values (Pichara: ¶14…Pichara discloses measuring physiochemical features of a sample (e.g., a candidate plant-based formulation) via NIR spectroscopy, wherein the sample is prepared from a set of source ingredients (the variable values), "The disclosed embodiments can utilize data science , food science and / or machine learning algorithms to find a combination of source ingredients that can taste , look and/or feel like a given target food item. The source ingredients and/or the target food item can be plant-based, animal-based or synthetic…In some examples, a combination of plant-based ingredients can be cooked using a certain recipe to taste, look and/or feel like cow milk"; ¶20…"Near-Infrared (NIR) spectroscopy techniques may be used to identify physical and/or chemical features of the ingredients");
extracting functional property feature values from the functional property signal (Pichara: ¶0057 and pg. 10 — "identifying a respective set of data features for each of a plurality of plant-based ingredients, the respective set of data features comprising at least one of physiochemical data features, nutritional data features");
determining a sample classification for the sample based on the functional property feature values, using a trained model (Pichara: pg. 10…Pichara trains an ML model to select potential candidates (the sample classification, e.g., matches dairy class or does not match) based on the matching of functional property feature values to target features, "training a machine learning prediction model by using the first training set ... to select potential candidates from the plurality of plant-based ingredients to match data features"); and
determining a set of updated variable values based on the sample classification (Pichara: pg. 10…Pichara teaches that based on the prediction model's output, the system determines a formula (i.e., updated variable values comprising the set of ingredients and proportions) to be applied for the next iteration of mimicking the target, "determining a formula, using the trained machine learning prediction model, to combine the potential candidates from the plurality of plant-based ingredients in specific proportions"; ¶16…"Once the prediction model is trained, the most important features used for the prediction may be selected as the potential candidates to mimic the target food item. After the potential candidates are selected, an optimization process can be executed to find a final formula comprising specific proportions of the source ingredients to mimic the target food item").
Per claim 15, Pichara discloses claim 13, further teaching wherein the functional property features comprise non-semantic features (Pichara: ¶0016… Pichara teaches that the prediction algorithms apply feature compression techniques such as kernel principal component analysis (PCA) and auto-encoding which produce a compact, latent representation of the input features, and such latent/compressed features are non-human-interpretable and lack a direct physical analog, i.e., non-semantic features, "the prediction algorithms may use feature compression techniques such as kernel principal component analysis and/or auto-encoding for training the prediction model").
Per claim 16, Pichara combined with Navon discloses claim 13, further teaching wherein the model is trained using training data comprising training functional property feature values labeled with associated sample classifications (Pichara: pg. 10…the source ingredients database labels each plant-based ingredient's feature vector with the ingredient identity (the sample classification), "a source ingredients database is configured to store the respective set of data features for each of the plurality of plant-based ingredients").
Per claim 17, Pichara combined with Navon discloses claim 13, further teaching wherein training the model comprises clustering training functional property feature values into a set of clusters, wherein determining the sample classification for the sample is determined by using the model to select a cluster from the set of clusters based on the functional property feature values for the sample (Pichara: ¶0027…Pichara expressly discloses unsupervised K-means clustering as one of the contemplated training approaches, and Pichara further discloses using clustering for determining ingredient/sample classes, "unsupervised learning (e.g., using an Apriori algorithm, using K-means clustering)").
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 2, 3, 6, 7, 8, 10, 11 and 14 are rejected under 35 USC 103 as being unpatentable over Pichara in view of US Pat. No. 10,957,424 to Navon et al. (cited in IDS, hereinafter Navon).
Per claim 2, Pichara discloses claim 1. Pichara does not expressly disclose, but with Navon does teach: the prototype functional property signal and the target functional property signal each comprise a data time series (Navon: col. 13, ln. 55-67…Navon teaches that the input acquisition computing system supports force sensors, motion sensors, heat sensors, and IMU sensors, and that the feature vectors are derived from sensor outputs of such sensors, wherein the sensor outputs of force/motion/heat sensors are by definition time-series data sampled over time, which constitutes data time series, "various types of sensors such as force sensors, motion sensors, heat sensors, accelerometers, gyroscopes, and inertial measurement unit (IMU) sensors")
Pichara and Navon are analogous art because they are from both within the same field of endeavor, specifically the field of using artificial intelligence to generate food formulations that mimic target food items. They address the same problem solving area of predicting plant-based ingredient combinations and proportions that approximate animal-based or other target foods based on a feature representation of the target. Pichara cites a prediction model trained on physiochemical, nutritional and molecular features of ingredients, which is the core subject of Navon (Navon: FIG. 3 and col. 10, ln. 53-60, "Each digitally stored feature vector associated with a combination of ingredients represents a set of features including at least one chemical feature, nutritional feature, and molecular feature of each ingredient in the combination of ingredients").
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to combine the feature-compression-plus-gradient-boosting / Lasso training pipeline of Pichara with the explicit neural-network feature/ingredient matching of Navon (Navon: FIG. 3) because both references operate on the same physiochemical/nutritional/molecular feature representation, both target the same plant-based mimicry problem, and Navon's neural-network framework provides a more general training architecture that subsumes Pichara's gradient-boosting prediction model. The combination is the simple substitution or addition of a known training architecture (Navon's NN) for the same purpose in Pichara's pipeline, with predictable results — KSR Int'l Co. v. Teleflex Inc., 550 U.S. 398, 416 (2007).
The suggestion/motivation for doing so would have been to obtain Navon's explicit neural-network framework for matching feature vectors and ingredient vectors, which directly improves Pichara's framework by supplying a bidirectional neural-network architecture suitable for the reverse application (Navon: col. 11, ln. 42-56…"the trained neural network may be applied in reverse in the second stage to generate the predicted formula"). A PHOSITA would have been motivated to extend Pichara to include the time-series sensor inputs and texture/flavor measurements described by Navon to broaden the set of properties the system can replicate.
Per claim 3, Pichara combined with Navon discloses claim 2. Pichara does not expressly disclose, but Navon teaches: the prototype functional property feature values and target functional property feature values are each extracted using time series decomposition (Navon: col. 10, ln. 53-60… that the feature vectors can be generated using neural-network-based feature extraction including convolutional layers; col. 12, ln. 45-53…time series decomposition (e.g., Fourier transform, wavelet decomposition, STL decomposition) is a well-known signal processing technique, where Navon teaches feature extraction via neural network layers over sensor inputs and a PHOSITA would have employed for routine engineering reasons when extracting features from sensor time-series outputs, "electronic signal paths ... I/O subsystem ... at least one hardware processor ... general-purpose microprocessor such as an embedded system or a graphics processing unit (GPU) or a digital signal processor").
The rationale to combine Navon with Pichara is the same as in the parent claim.
Per claim 6, Pichara combined with Navon discloses claim 1. Pichara further teaches: wherein the comparison between the prototype functional property feature values and target functional property feature values comprises a distance between the prototype functional property feature values and the target functional property feature values (Pichara: ¶0024, 0036…Pichara describes the matching process in terms of representing each food item in a D-dimensional feature space and computing proximity/matching scores between the source ingredient feature representation and the target feature representation, “In some implementations, the formula generator 106 may create a screening of the target food item and a set of source ingredients. The set of source ingredients may include some or all of the first source ingredient 102a, the second source ingredient 102b, the third source ingredient 102c, and the Pth source ingredient 102p. The screening may represent each food item in a D-dimensional space comprising the physiochemical, nutritional or molecular features associated with the food item. For example, each food item can be represented in a vector space of multiple features associated with the physiochemical, nutritional or molecular properties”.
Per claim 7, Pichara combined with Navon discloses claim 6, Pichara further teaching weighting the prototype functional property feature values and the target functional property feature values, wherein the distance comprises a distance between the weighted prototype functional property feature values and the weighted target functional property feature values (Pichara: ¶0025…Pichara explicitly applies Lasso regression to weight feature coefficients during prediction such that ingredients with negligible contribution are deleted, which constitutes weighting the prototype and target feature values, "Lasso optimization may be performed to balance the proportions of different ingredients in the set of source ingredients. For example, ingredients with almost negligible contribution ... may be deleted").
Per claim 8, Pichara combined with Navon discloses claim 1. Pichara further teaches: measuring a binary characteristic of the prototype sample, wherein the set of manufacturing variable values is determined further based on the binary characteristic (Pichara: ¶0041…Pichara teaches that human panelists or sensory evaluations may provide acceptance or rejection feedback for the prototype, which is a binary characteristic and which feeds back into the formula determination, which constitutes binary characteristic input, " The human feedback 610 may include a flavor 610a, a color 610b, and/or any other sensorial feedback. The flavor 610a may include a flavor of the cooked food and the color 610b may include a color of the cooked food. As an example, the human feedback 610 may be provided by the formula feedback panel 112").
Per claim 10, Pichara combined with Navon discloses claim 1. Pichara combined with Navon further teaches: the model comprises an encoder trained to encode functional property feature values and manufacturing variable values (Pichara: ¶0016…Pichara discloses auto-encoding for feature compression to encode functional property feature values into a latent representation, "feature compression techniques such as kernel principal component analysis and/or auto-encoding for training the prediction model"; Navon: col. 2, ln. 45-56…Navon teaches training a neural network in a first stage to match feature vectors and ingredients (proportions) vectors by modifying parameters, which is functionally equivalent to encoding both feature values and manufacturing variable values into the same learned representation, "training, using the training set, the neural network in a first stage to match the plurality of digitally stored feature vectors and the plurality of digitally stored ingredients vectors by modifying parameters of the neural network").
The rationale to combine Navon with Pichara is the same as in the parent claim.
Per claim 11, Pichara combined with Navon discloses claim 1. Pichara combined with Navon further teaches wherein the target functional property signal comprises a measurement for at least one of: texture, melt, or flavor (Pichara: ¶0025… "sensorial descriptors (e.g., flavor, color, texture or taste)"; Navon: col. 5, ln. 1-9…Navon discloses that the properties of the target food include texture descriptors and taste (salt, sweet, bitter, sour, umami) which are flavor measurements, which constitutes a measurement for at least one of: texture, melt, or flavor, "The properties may include human sensorial feedback such as taste (e.g., salt, sweet, bitter, sour, and umami), texture descriptors, acceptance, and the like").
The rationale to combine Navon with Pichara is the same as in the parent claim.
Per claim 14, Pichara combined with Navon discloses claim 13, further teaching wherein the functional property signal comprises a data time series, wherein the functional property feature values are extracted using time series analysis (Navon: col. 13, ln. 55-67…Navon teaches that the input acquisition system uses force sensors, motion sensors, and heat sensors whose outputs are time series, and that neural-network feature extraction is applied to these inputs, which under BRI is time series analysis, "various types of sensors such as force sensors, motion sensors, heat sensors, accelerometers, gyroscopes").
The rationale to combine Navon with Pichara is the same as in the parent claim.
Claim 18 is rejected under 35 USC 103 as being unpatentable over Pichara in view of Generative Adversarial Nets to Goodfellow et al. (hereinafter Goodfellow).
Pichara discloses claim 13.
Pichara does not expressly disclose, but with Goodfellow does teach: wherein the model is trained using adversarial machine learning methods (Goodfellow: Abstract…Goodfellow defines the generative adversarial network (GAN) framework, in which a generative model is trained against a discriminative adversary, and shows it is a generally applicable framework for training neural networks, which constitutes adversarial machine learning methods, "We propose a new framework for estimating generative models via an adversarial process, in which we simultaneously train two models: a generative model G that captures the data distribution, and a discriminative model D that estimates the probability that a sample came from the training data rather than G").
Pichara and Goodfellow are analogous art because they are both within the field of machine learning for training neural-network-based predictive models. Pichara discloses neural networks, restricted Boltzmann machines, and stacked autoencoders as candidate training architectures (Pichara: ¶0027), and Goodfellow's GAN framework is one of the most well-known training paradigms for neural-network-based generators that Pichara contemplates.
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to train Pichara using GAN-based adversarial training (or other adversarial methods derived therefrom) because Goodfellow's GAN was an established and broadly-applied training framework that provides a generally applicable method for training generative neural networks, and applying GANs to food-formula generation is a routine choice yielding predictable improvements in sample quality.
The suggestion/motivation for doing so would have been to obtain Goodfellow's demonstrated training framework for generating realistic synthetic samples (here, candidate plant-based formulations) without explicit probability density estimation, which directly improves Pichara's prediction-and-generation pipeline by enabling richer, distribution-matching samples.
Claim 19 is rejected under 35 USC 103 as being unpatentable over Pichara in view of A Unified Approach to Interpreting Model Predictions to Lundberg et al. (hereinafter Lundberg).
Pichara discloses claim 13.
Pichara does not expressly disclose, but with Lundberg does teach: wherein the set of updated variable values are determined using explainability methods applied to the trained model (Lundberg: Abstract and §1…Lundberg introduces SHAP (SHapley Additive exPlanations), a unified framework that explains a model's prediction by assigning each input feature an importance value for that prediction, such that applying SHAP to Pichara's prediction model surfaces which features (and therefore which ingredient variable values) drive the predicted classification, which constitutes determined using explainability methods applied to the trained model, "We present a unified framework for interpreting predictions, SHAP (SHapley Additive exPlanations). SHAP assigns each feature an importance value for a particular prediction")
Pichara and Lundberg are analogous art because they are both within the same machine-learning field of using trained predictive models, and Lundberg's interpretability framework is broadly applicable to any trained model (including the tree-based gradient boosting models that Pichara expressly uses (Pichara: ¶0027)).
Before the effective filing date of the claimed invention, it would have been obvious to a person of ordinary skill in the art to apply Lundberg's SHAP/feature-importance framework to Pichara's trained model to identify the ingredient features driving the model's classification of a candidate formulation, and then to update the corresponding ingredient variable values based on those importance scores. Lundberg's SHAP was a widely adopted, model-agnostic interpretability technique, and using it to guide formula-iteration decisions is a routine application yielding predictable results.
The suggestion/motivation for doing so would have been to obtain Lundberg's principled, mathematically grounded feature-importance values to determine which ingredient adjustments will most efficiently push a candidate formulation toward the target, a direct improvement of Pichara's iterative formula refinement process (Pichara: FIG. 6 and ¶0073…sensory feedback loop).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALAN CHEN whose telephone number is (571)272-4143. The examiner can normally be reached M-F 10-7.
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, Kamran Afshar can be reached at (571) 272-7796. 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.
/ALAN CHEN/Primary Examiner, Art Unit 2125
Prosecution Insights