DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This action is in response to the amendment filed on Feb. 5th, 2026. The amendments are linked to the original application filed on Dec. 8th, 2022.
Response to Arguments
The Examiner thanks the applicant for the remarks, edits and arguments.
Regarding Claim Rejections – 35 U.S.C. 101
The applicant has amended the claims and believes the current amended claims are patent eligible under 35 U.S.C. 101. The applicant argues that the claims recite a technical solution to a technical problem stated in the specification. The examiner would like to point to MPEP 2106.05(a) which states, “During examination, the examiner should analyze the "improvements" consideration by evaluating the specification and the claims to ensure that a technical explanation of the asserted improvement is present in the specification, and that the claim reflects the asserted improvement. Generally, examiners are not expected to make a qualitative judgement on the merits of the asserted improvement. If the examiner concludes the disclosed invention does not improve technology, the burden shifts to applicant to provide persuasive arguments supported by any necessary evidence to demonstrate that one of ordinary skill in the art would understand that the disclosed invention improves technology.”. This section of the MPEP discloses that the claims themselves must also reflect the technical improvement recited in the specification. The examiner has reviewed the claims, as a whole, and it is unclear how the claims recite training classification models when available training data is insufficient amounts. The claims recite a process of producing synthetic data sets and briefly a process of training a pre-trained model using the generated data. However, it is unclear how this provides an improvement to classification models and how this is directed to training models with insufficient training samples. The claims do not state how the generated data is specially optimized for target classification tasks. The examiner would like to note the claims do recite a process of generating synthetic training and the examiner believes the specification recites sufficient details on the improvements to technology. However, the examiner believes further details are required to be added to the claims to clearly recite the proposed technical improvements from the specification.
Next, the applicant recites Enfish, LLC v. Microsoft Corp., and believes that the amended claims are similar to this case because the claims recite a technical improvement. Further, the applicant states, “the amended claims improve the existing technological process of training ML classifiers when training data is insufficient. The claims address a specific problem in ML technology (insufficient training samples) and provide a specific solution (targeted synthetic dataset generation) that improves ML training outcomes (demonstrated accuracy improvements) that operates through specific technical means (OT maps, neural networks, geodesic hulls).”. The examiner would like to note that claims currently do not recite information on insufficient training samples. The current claims recite a process to generate synthetic data when requested and to assist in training a pre-trained model. The current claims do not address or disclose systems that might have insufficient training samples and how the claimed process will improve the training samples. Further, the examiner believes the claims do not disclose subject matter on how the claimed process improves the accuracy of datasets either. The examiner believes that further details from the specification is required to further disclose the claimed improvements.
Next, the applicant recites McRO, Inc. v. Bandai Namco Games Am. Inc., and the applicant states, “The amended claims here provide specific implementation details including computing OT maps using a neural network, operating on classification datasets with different label sets, generating synthetic samples in a feature space of the target dataset, and working with three distinct datasets in a specific structural relationship. These specific details improve the technological process of ML training, not through routine conventional activity, but through a specific, unconventional approach.”. The examiner would like to note that the current claims state, “determining an optimal transport (OT) map from the target labelled dataset to the first[second] training labelled dataset;” This process is further refined in claim 7 which recites, “wherein determining the OT map includes operating an OT neural map that includes three classifiers, a label classifier, a discriminator, and a feature classifier.”. This process uses a neural map which includes different features. However, this process fails to disclose how this architecture improves ML training in general. Neural Network architecture is varied and there are well known examples of neural networks which contain modules that perform functions such as classification, feature extraction and using discriminator modules. The disclosed process in the specification may recite a novel structure, however the claims do not reflect this structure, instead they recite generic neural network using generic neural network modules. The examiner believes further details from the specification is required to clarify the claimed improvements to training machine learning models.
Next, the applicant recites, DDR Holdings, LLC v. Hotels.com, L.P., and further states, “Here, the problem of insufficient training data for ML classifiers is a problem specifically arising in the realm of computer-implemented machine learning. The solution of synthesizing additional training data through OT interpolation in distribution space using NNs is necessarily rooted in computer technology.”. The examiner would like to note again that the claims fail to provide sufficient details to the proposed improvements to technology. As stated above, the examiner believes the claims do not recite sufficient information to allow a person of ordinary skill in the art to be able to recognize the alleged improvements to technology or machine learning training.
Next, the applicant argues that the claims are not directed to abstract ideas and/or mental concepts. The applicant states, “OT map determination and NN implementation of the same are not mental processes. The amended claims explicitly recite "determining, by a neural network (NN), an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset" and "determining, by the NN, an OT map from the target labelled dataset to the second training labelled dataset". Determining OT maps in one's head or by pen and paper is generally not feasible for anything beyond the most trivial scenarios. OT is a mathematical and computational theory that involves finding the most efficient way to move a distribution of ''mass" from one configuration to another while minimizing a given cost function. Determining an OT map involves complex mathematical operation that requires significant computational power. This is not a mental process. The Examiner's assertion ignores the computational complexity and scale involved.”. The examiner would like to note that the Optimal transport theory is a mathematical theory which can be performed using pen and paper. Next, the examiner would like to note that a neural network, at its core, is a set of sequential equations that are performed on dynamic values. This is also a process where a human can theoretically perform using pen and paper. Next, the MPEP 2106.04(a)(III)(C) states, “Claims can recite a mental process even if they are claimed as being performed on a computer. … In evaluating whether a claim that requires a computer recites a mental process, examiners should carefully consider the broadest reasonable interpretation of the claim in light of the specification. For instance, examiners should review the specification to determine if the claimed invention is described as a concept that is performed in the human mind and applicant is merely claiming that concept performed 1) on a generic computer, or 2) in a computer environment, or 3) is merely using a computer as a tool to perform the concept. In these situations, the claim is considered to recite a mental process.”. The recited claim limitation, “determining an optimal transport (OT) map from the target labelled dataset to the first[second] training labelled dataset;”, using the BRI, discloses a process of determining an OT map from one dataset to another. This process would be performed on a computing device or using a computing device as a tool. A human would reasonably be able to use a generic computer to handle the dataset and perform the abstract action of the OT computations. Further, the claims fail to disclose details on why performing this step would require significant computation power other than a generic computing device. Claims 1, 12 and 18 do not disclose the use or requirement of a complex or unique computing system other than a generic computing system containing processors connected to memory. This would lead one of ordinary skill in the art to recognize that this claim limitation recites a process which can be performed mentally using a generic computer as a tool. The examiner believes this claim limitation still recites an abstract concept and/or mental process.
Next, the applicant argues that the operations on datasets as claimed is not abstract or a mental process. The applicant states, “The claims recite computing probability distributions over these high-dimensional spaces by determining optimal transport maps between such distributions, forming geodesic hulls in distribution space (where each dataset is a point), and computing distances in these spaces. This involves optimization over joint probability distributions, integration over product spaces, and minimization of transport cost functionals. The Examiner's characterization that these operations can be performed with "pen and paper" (Office Action at 3) is incorrect The scale and complexity of these computations necessitate computer implementation and are fundamentally different from simple arithmetic.”. The examiner would like to note the above paragraph. The examiner believes that a person of ordinary skill in the art would be able to evaluate the claims using the BRI and recognize the claims as reciting a process using a generic processing system. Next, the examiner would like to point to MPEP 2106.04(a)(III)(C) again. As stated, claims limitations that are implement on a computing system can still be considered abstract ideas. Finally, the examiner would like to note that computations performed in the claims require more than simple arithmetic. However, the core elements of this computations, are capable of being performed by a human using known mathematical concepts such as: calculus, differential equations, discrete mathematics, etc. All of these examples would be considered more than simple arithmetic and within the realm of human capability.
Next, the applicant states, “The claims address a specific problem of generating training datasets from distinctly different datasets (different features, different labels, or a combination thereof). This is not an abstract problem but a concrete technological challenge in machine learning. Obtaining sufficient labeled training data is expensive and time-consuming. Many real-world classification tasks have limited labeled samples. Insufficient training data leads to poor model performance.”. The examiner would like to point to previous arguments on the technical improvements recited in the claims. As stated above, the examiner believes that the current claims fail to disclose the technical improvements.
Finally, the applicant states, “The three-way structural relationship between the training datasets and the target dataset is unconventional. The claims recite a specific structural relationship between datasets, namely a target labelled dataset, a first training labelled dataset distinct from the target, and a second training labelled dataset distinct from the target and distinct from the first, OT maps FROM target TO each training dataset, a geodesic hull formed BY the training datasets, etc. This specific structure enables targeted dataset synthesis from distinct input data. Creating a synthetic dataset optimized specifically for the target classification task by intelligently combining available training datasets. The specification explains this is unconventional "Instead of limiting this choice to those datasets already present in the pretraining collection, embodiments extend available datasets to all datasets that can be synthesized as 'combinations' of the heterogeneous datasets in a dataset space spanned by the heterogenous datasets." Specification at [0134]. This ordered combination is **not well-understood, routine, or conventional** in the field.”. The examiner would like to note that the claims were evaluated using the Alice/Mayo test. This test is not deigned to determine novelty of an invention but if the claims recite patent eligible subject matter. Using this test the examiner believes the processes of, “determining an optimal transport (OT) map from the target labelled dataset to the first[second] training labelled dataset;” and identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and” are limitation that recite abstract ideas. The examiner believes these claims recite the process of evaluating altering data using known mathematical concepts using a generic computing system. Next, the examiner must evaluate the remaining limitations for additional elements that provide a technical improvement and recite a practical application. The claims stated above may recite novel concepts, however, the key concepts of the claims fall under abstract ideas and they need to be supported by additional limitations. The additional limitations listed, including obtaining datasets and executing the abstract ideas to produce an output would fail to provide sufficiency more than judicial exception. These additional limitations would be considered known concepts such as transferring data to/from memory or transmitting data over a network. For the reasons stated above, the examiner believes that the claims fail to comply with 35 U.S.C. 101 and recite patent ineligible subject matter. Therefore, the examiner has upheld the rejection under 35 U.S.C. 101, see 101 rejection below.
Regarding Claim Rejections – 35 U.S.C. 103
The applicant believes that the proposed prior art fails to disclose or teach the added limitations. The applicant argues that the Rout fails to teach the OT map which maps data in the specific direction as the claims recite. Also, the applicant believes that Rout also fails to disclose the use of multiple datasets, a first and second separate and unique datasets. In particular the applicant states, “Rout does not teach this reverse mapping approach where maps FROM a target TO training datasets are used to determine how to combine the training datasets. Rout's maps go from source to target for the purpose of generation. The Examiner has not explained how or why one would reverse this direction or use maps in this fundamentally different way.”. Further the applicant believes that Rout fails to teach the use of a geodesic hull and that does not teach the combination of datasets as disclosed in the claims. The applicant further argues that Solomon fails to teach the elements that Rout fails to teach. Finally, the applicant argues that the combination of Rout and Solomon would not disclose the claimed limitations and these arts teach fundamentally different concepts and it would not be obvious to combine them. The examiner has considered the arguments made by the applicant and finds them persuasive. The examiner would like to note that the proposed arts Rout and Solomon may teach subject matter which is similar in nature, they fail to disclose the amended limitations in claims 1, 12 and 18. However, after each amendment to the claims the examiner is required to evaluate the claims to ensure they comply with 35 U.S.C. 102/103. After reviewing the specification, the remarks and the amended claims the examiner was able to perform a complete search and discover new subject matter which is able to teach or disclose the amendments. The examiner has found art which is able to disclose the missing elements of Rout and Solomon and no longer relies on these arts to teach or disclose claims 1, 12, and 18. The examiner still believes that Rout and Solomon can still be used in combination of the newly discovered art to teach other dependent claims. Therefore, the examiner believes, the combination of prior arts proposed teaches or disclosed the claimed subject matter and the rejection under 35 U.S.C. 103 is upheld.
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 1-20 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, 84 Fed. Reg. 50 (“2019 PEG”).
Claim 1
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
Claim 1 recites, "A computer-implemented method for generating a synthetic labelled machine learning (ML) dataset, the method comprising:" therefore it is directed to the statutory category of a process.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“defining or obtaining a target labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to define a target dataset using a generic computer as a tool. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to perform computations using a generic computer as a tool. This would include determining the OT map between datasets. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“determining an OT map from the target labelled dataset to the second training labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to perform computations using a generic computer as a tool. This would include determining the OT map between datasets. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to identify a point on a 2d surface and perform mathematical functions to determine this point and to use that same point to compute other mathematical functions. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “A computer-implemented method for generating a synthetic labelled machine learning (ML) dataset, the method comprising:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” is an insignificant extra-solution activity required for any uses of the mental processes (see MPEP § 2106.05(g)) As such, the claim is ineligible.
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset the second training labelled dataset comprising: second samples associated with second labels, at least one of the first labels is different from all of the second labels;” is an insignificant extra-solution activity required for any uses of the mental processes (see MPEP § 2106.05(g)) As such, the claim is ineligible.
“producing the synthetic labelled ML dataset by combining, based on distances between probability distribution representations of the first and second training labelled datasets in the distribution space and the point, the first and second training labelled datasets the synthetic labelled ML dataset comprises synthetic samples in a feature space of the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “A computer-implemented method for generating a synthetic labelled machine learning (ML) dataset, the method comprising:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” is an insignificant extra-solution activity required for any uses of abstract ideas (see MPEP § 2106.05(g)), and is a well-understood, routine, conventional activity (see MPEP § 2106.05(d)(II)(i); “Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset the second training labelled dataset comprising: second samples associated with second labels, at least one of the first labels is different from all of the second labels;” is an insignificant extra-solution activity required for any uses of abstract ideas (see MPEP § 2106.05(g)), and is a well-understood, routine, conventional activity (see MPEP § 2106.05(d)(II)(i); “Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
“producing the synthetic labelled ML dataset by combining, based on distances between probability distribution representations of the first and second training labelled datasets in the distribution space and the point, the first and second training labelled datasets the synthetic labelled ML dataset comprises synthetic samples in a feature space of the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 2
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein the target labelled dataset includes more, fewer, or different labels than labels of one or more of the first and second training labelled datasets.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to use a generic computing devices to generate a labeled dataset of given dimensions and restrictions. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 3
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein combining the first and second training labelled datasets includes representing labels of the first and second training labelled datasets as respective one-hot vectors of all labels in the first and second training labelled datasets.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to perform functions to combine datasets and evaluate datasets and use mathematical functions to encode the values from the dataset into vectors. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 4
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “further comprising further training, using the synthetic labelled ML dataset, a pre-trained ML model that has been trained based on the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “further comprising further training, using the synthetic labelled ML dataset, a pre-trained ML model that has been trained based on the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 5
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein identifying the point includes performing a barycentric projection of the target labelled dataset onto the geodesic hull.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to evaluate a 2D object using a generic computer and identify a point based on given restrictions. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 6
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the barycentric projection includes projection of sample data and separate label data.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the barycentric projection includes projection of sample data and separate label data.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 7
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein determining the OT map includes operating an OT neural map that includes three classifiers, a label classifier, a discriminator, and a feature classifier.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein determining the OT map includes operating an OT neural map that includes three classifiers, a label classifier, a discriminator, and a feature classifier.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 8
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein a discriminator loss of the discriminator is independent of the labels.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein a discriminator loss of the discriminator is independent of the labels.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 9
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein identifying the point proximate the target labelled dataset in the dataset space includes determining the point in the generalized geodesic hull that is closest to the target labelled dataset.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to evaluate a 2D object using a generic computer and identify a point based on given restrictions. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 10
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein identifying the point proximate the target labelled dataset includes operating a quadratic problem solver based on a (2, v) transport metric.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses mathematical concept of utilizing a mathematical formula to perform calculations. A human is able to perform know mathematical functions to produce an outcome. This claim discloses a math operation and therefore is ineligible.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 11
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A process, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “before obtaining the first or second labelled training datasets, receiving, from an application, a request for the synthetic labelled ML dataset; and” is an insignificant extra-solution activity required for any uses of the mental processes (see MPEP § 2106.05(g)) As such, the claim is ineligible.
“responsive to producing the synthetic labelled ML dataset, providing the synthetic labelled ML dataset to the application.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “before obtaining the first or second labelled training datasets, receiving, from an application, a request for the synthetic labelled ML dataset; and” is an insignificant extra-solution activity required for any uses of abstract ideas (see MPEP § 2106.05(g)), and is a well-understood, routine, conventional activity (see MPEP § 2106.05(d)(II)(i); “Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
“responsive to producing the synthetic labelled ML dataset, providing the synthetic labelled ML dataset to the application.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 12
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
Claim 12 recites, "A non-transitory machine-readable medium including instructions that, when executed by a machine, cause the machine to perform operations for generating a synthetic labelled machine learning (ML) dataset, the operations comprising:" therefore it is directed to the statutory category of a machine.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“defining or obtaining a target labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to define a target dataset using a generic computer as a tool. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to perform computations using a generic computer as a tool. This would include determining the OT map between datasets. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“determining an OT map from the target labelled dataset to the second training labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to perform computations using a generic computer as a tool. This would include determining the OT map between datasets. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to identify a point on a 2d surface and perform mathematical functions to determine this point and to use that same point to compute other mathematical functions. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “A non-transitory machine-readable medium including instructions that, when executed by a machine, cause the machine to perform operations for generating a synthetic labelled machine learning (ML) dataset, the operations comprising:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” is an insignificant extra-solution activity required for any uses of the mental processes (see MPEP § 2106.05(g)) As such, the claim is ineligible.
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset the second training labelled dataset comprising: second samples associated with second labels, at least one of the first labels is different from all of the second labels;” is an insignificant extra-solution activity required for any uses of the mental processes (see MPEP § 2106.05(g)) As such, the claim is ineligible.
“producing the synthetic labelled ML dataset by combining, based on distances between probability distribution representations of the first and second training labelled datasets in the distribution space and the point, the first and second training labelled datasets the synthetic labelled ML dataset comprises synthetic samples in a feature space of the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “A non-transitory machine-readable medium including instructions that, when executed by a machine, cause the machine to perform operations for generating a synthetic labelled machine learning (ML) dataset, the operations comprising:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” is an insignificant extra-solution activity required for any uses of abstract ideas (see MPEP § 2106.05(g)), and is a well-understood, routine, conventional activity (see MPEP § 2106.05(d)(II)(i); “Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset the second training labelled dataset comprising: second samples associated with second labels, at least one of the first labels is different from all of the second labels;” is an insignificant extra-solution activity required for any uses of abstract ideas (see MPEP § 2106.05(g)), and is a well-understood, routine, conventional activity (see MPEP § 2106.05(d)(II)(i); “Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
“producing the synthetic labelled ML dataset by combining, based on distances between probability distribution representations of the first and second training labelled datasets in the distribution space and the point, the first and second training labelled datasets the synthetic labelled ML dataset comprises synthetic samples in a feature space of the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 13
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A machine, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein the target labelled dataset includes more, fewer, or different labels than labels of one or more of the first and second training labelled datasets.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to use a generic computing devices to generate a labeled dataset of given dimensions and restrictions. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 14
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A machine, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein combining the first and second training labelled datasets includes representing labels of the first and second training labelled datasets as respective one-hot vectors of all labels in the first and second training labelled datasets.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to perform functions to combine datasets and evaluate datasets and use mathematical functions to encode the values from the dataset into vectors. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 15
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A machine, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the operations further comprise further training, using the synthetic labelled ML dataset, a pre-trained ML model that has been trained based on the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the operations further comprise further training, using the synthetic labelled ML dataset, a pre-trained ML model that has been trained based on the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 16
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A machine, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“wherein identifying the point includes performing a barycentric projection of the target labelled dataset onto the geodesic hull.” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to evaluate a 2D object using a generic computer and identify a point based on given restrictions. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
This claim does not recite any additional limitations which integrate the abstract idea into a practical application.
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea and thus the claim is subject-matter ineligible.
Claim 17
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A machine, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein the barycentric projection includes projection of sample data and separate label data.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein the barycentric projection includes projection of sample data and separate label data.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 18
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
Claim 18 recites, "A system for generating a synthetic labelled machine learning (ML) dataset, the system comprising: processing circuitry; and a memory coupled to the processing circuitry, the memory including instructions that, when executed by the processing circuitry, cause the processing circuitry to perform operations comprising:" therefore it is directed to the statutory category of a machine.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites, inter alia:
“defining or obtaining a target labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to define a target dataset using a generic computer as a tool. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to perform computations using a generic computer as a tool. This would include determining the OT map between datasets. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“determining an OT map from the target labelled dataset to the second training labelled dataset;” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to perform computations using a generic computer as a tool. This would include determining the OT map between datasets. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
“identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and” Under its broadest reasonable interpretation in light of the specification, this limitation encompasses the mental process of evaluating and observing data, which is an evaluation or observation that is practically capable of being performed in the human mind with the assistance of pen and paper. A human is able to identify a point on a 2d surface and perform mathematical functions to determine this point and to use that same point to compute other mathematical functions. The limitation is merely applying an abstract idea on generic computer system. See MPEP 2106.04(a)(2)(III)(c).
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “A system for generating a synthetic labelled machine learning (ML) dataset, the system comprising: processing circuitry; and a memory coupled to the processing circuitry, the memory including instructions that, when executed by the processing circuitry, cause the processing circuitry to perform operations comprising:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” is an insignificant extra-solution activity required for any uses of the mental processes (see MPEP § 2106.05(g)) As such, the claim is ineligible.
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset the second training labelled dataset comprising: second samples associated with second labels, at least one of the first labels is different from all of the second labels;” is an insignificant extra-solution activity required for any uses of the mental processes (see MPEP § 2106.05(g)) As such, the claim is ineligible.
“producing the synthetic labelled ML dataset by combining, based on distances between probability distribution representations of the first and second training labelled datasets in the distribution space and the point, the first and second training labelled datasets the synthetic labelled ML dataset comprises synthetic samples in a feature space of the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “A system for generating a synthetic labelled machine learning (ML) dataset, the system comprising: processing circuitry; and a memory coupled to the processing circuitry, the memory including instructions that, when executed by the processing circuitry, cause the processing circuitry to perform operations comprising:” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” is an insignificant extra-solution activity required for any uses of abstract ideas (see MPEP § 2106.05(g)), and is a well-understood, routine, conventional activity (see MPEP § 2106.05(d)(II)(i); “Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset the second training labelled dataset comprising: second samples associated with second labels, at least one of the first labels is different from all of the second labels;” is an insignificant extra-solution activity required for any uses of abstract ideas (see MPEP § 2106.05(g)), and is a well-understood, routine, conventional activity (see MPEP § 2106.05(d)(II)(i); “Receiving or transmitting data over a network, e.g., using the Internet to gather data”.
“producing the synthetic labelled ML dataset by combining, based on distances between probability distribution representations of the first and second training labelled datasets in the distribution space and the point, the first and second training labelled datasets the synthetic labelled ML dataset comprises synthetic samples in a feature space of the target labelled dataset.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 19
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A machine, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein determining the OT map includes operating an OT neural map that includes three classifiers, a label classifier, a discriminator, and a feature classifier.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein determining the OT map includes operating an OT neural map that includes three classifiers, a label classifier, a discriminator, and a feature classifier.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
Claim 20
Step 1 – Is the claim to a process, machine, manufacture or composition of matter?
A machine, as above.
Step 2A Prong 1 – Does the claim recite an abstract idea, law of nature, or natural phenomenon?
The claim recites the abstract ideas of the preceding claims from which it depends.
Step 2A Prong 2 – Does the claim recite additional elements that integrate the judicial exception into a practical application?
The claim recites the additional elements, “wherein a discriminator loss of the discriminator is independent of the labels.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Step 2B – Does the claim recite additional elements that amount to significantly more than the judicial exception?
Finally, the claim taken as a whole does not contain an inventive concept which provides significantly more than the abstract idea. The additional elements, “wherein a discriminator loss of the discriminator is independent of the labels.” amounts to generic computer components used as a tool to perform an existing process. Thus, the additional element amounts to no more than a recitation of the words "apply it" (or an equivalent) or are more than mere instructions to implement an abstract idea or other exception on a computer (see MPEP § 2106.05(f)).
Taken alone or in combination, the additional elements of the claim do not provide an inventive concept and thus the claim is subject-matter ineligible.
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.
Claims 1, 2, 12, 13, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over He et al, (He et al, “Task-adaptive Few-shot Learning on Sphere Manifold”, Aug. 25th, 2022, hereinafter “He”) in view of Liu et al, (Liu et al, “Wasserstein Task Embedding for Measuring Task Similarities”, Aug. 24th, 2022, hereinafter “Liu”).
Regarding claim 1, He discloses, “A computer-implemented method for generating a synthetic labelled machine learning (ML) dataset, the method comprising:” (Problem Description, pp. 2950; “For each task, the support set is used as the labeled dataset, and the query set is used as the unlabeled dataset, and then the model is trained by predicting the categories of NQ unlabeled examples in the query set and minimizing the objective function.” This article discloses a model that will generate different examples and datasets to train a machine learning model.)
“defining or obtaining a target labelled dataset;” (Problem Description, pp. 2950; “In this paper, we follow a typical few-shot learning setting and define the learning process as several iterations of training tasks, with each task
D
T
being an N-way K-shot classification problem. … For the learning process, the dataset D is randomly sampled several times according to the above settings to form the task set
{
D
1
T
,
D
2
T
,
…
D
n
T
}
as the training set.” This article discloses a process which takes different datasets to produce a training dataset. This is interpreted to be the target dataset.)
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” (Problem Description, pp. 2950; “Specifically, given a base dataset D, a task
D
T
⊆
D
is randomly sampled from it,
D
T
consists of two sets: a support set
D
s
contains N classes with K samples per class, K usually being 1 or 5; and a query set
D
q
contains N same classes with Q samples per class.” The support set is interpreted to be the first dataset containing labeled data.)
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset the second training labelled dataset comprising: second samples associated with second labels, at least one of the first labels is different from all of the second labels;” (Problem Description, pp. 2950; “Specifically, given a base dataset D, a task
D
T
⊆
D
is randomly sampled from it,
D
T
consists of two sets: a support set
D
s
contains N classes with K samples per class, K usually being 1 or 5; and a query set
D
q
contains N same classes with Q samples per class. We denote them as
D
s
=
{
x
i
s
,
y
i
s
}
i
=
1
N
×
K
and
D
q
=
{
x
j
q
,
y
j
q
}
j
=
1
N
×
Q
, respectively, where x is the image, y is the corresponding label. Note that
D
T
=
D
s
∪
D
q
,
D
s
∩
D
q
=
∅
” The query set is interpreted to be the second dataset containing labeled data. As stated above the intersection of the support set and query set is empty indicating there are not shared labels and the two sets are distinctly different.)
“producing the synthetic labelled ML dataset by combining, based on distances between probability distribution representations of the first and second training labelled datasets in the distribution space and the point, the first and second training labelled datasets the synthetic labelled ML dataset comprises synthetic samples in a feature space of the target labelled dataset.” (Few-shot Learning on Sphere manifold, pp. 2951-2952; “For this method, the selection of the tangent point is crucial because the tangent space is more effective for the neighborhood of the tangent point. If the points are too far from the tangent point, the original geometric properties may be lost after Logarithmic mapping to the tangent space, and the distance is not representative. Therefore, combined with the setting of few-shot learning, we propose a method that adaptively selects the tangent point according to the samples, as shown in Fig. 2.” This model will generate different examples which it uses to train a machine learning model. This will actively select generated samples using the given datasets and attempt to learn new features.)
He fails to explicitly disclose, “determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;”, “determining an OT map from the target labelled dataset to the second training labelled dataset;”, and “identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and”.
However, Liu discloses, “determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;” (Optimal Transport Dataset Distance (OTDD), pp. 4; “Let
X
=
{
x
n
∈
R
d
}
n
=
1
N
be the input set with labels (classes)
Y
=
{
y
j
}
j
=
1
J
. For each j, let
C
y
i
⊆
X
denote the class with label
y
j
. Following the OTDD framework, let
T
=
x
i
,
y
j
∈
X
×
Y
x
i
∈
C
y
j
}
i
,
j
denote the set of datalabel pairs. OTDD encodes the label
y
j
as distribution
V
y
j
, where
V
y
j
=
1
C
y
j
∑
x
i
∈
C
y
j
δ
x
i
The ground distance in T is then defined by combining the Euclidean distance between the data points and the 2-Wasserstein distance between label distributions: [See Equation (8)] Based on this metric, the OT distance between two distributions
μ
i
and
μ
j
on T is [See Equation (9)] where
∏
(
μ
i
,
μ
j
)
denotes the set of transport plans between
μ
i
and
μ
j
.” This article discloses the use of determining the OTDD of two data sets. Liu teaches a method to concatenate two different datasets a modified OTDD method and Wasserstein embedding.) and (Figure 1, pp. 5; The Wasserstein embedding is seen on the left side of the figure.)
“determining an OT map from the target labelled dataset to the second training labelled dataset;” (Optimal Transport Dataset Distance (OTDD), pp. 4; “Let
X
=
{
x
n
∈
R
d
}
n
=
1
N
be the input set with labels (classes)
Y
=
{
y
j
}
j
=
1
J
. For each j, let
C
y
i
⊆
X
denote the class with label
y
j
. Following the OTDD framework, let
T
=
x
i
,
y
j
∈
X
×
Y
x
i
∈
C
y
j
}
i
,
j
denote the set of datalabel pairs. OTDD encodes the label
y
j
as distribution
V
y
j
, where
V
y
j
=
1
C
y
j
∑
x
i
∈
C
y
j
δ
x
i
The ground distance in T is then defined by combining the Euclidean distance between the data points and the 2-Wasserstein distance between label distributions: [See Equation (8)] Based on this metric, the OT distance between two distributions
μ
i
and
μ
j
on T is [See Equation (9)] where
∏
(
μ
i
,
μ
j
)
denotes the set of transport plans between
μ
i
and
μ
j
.” This article discloses the use of determining the OTDD of two data sets. Liu teaches a method to concatenate two different datasets a modified OTDD method and Wasserstein embedding.) and (Figure 1, pp. 5; The Wasserstein embedding is seen on the left side of the figure.)
“identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and” (Wasserstein Embedding (WE), pp. 4; “The optimal transport plan
π
i
*
is the minimizer of the above optimization problem, which is solved by linear program at cost
O
N
3
log
N
,
N
being the number of input samples. To avoid mass splitting, the barycentric projection (Wang et al. 2013) assigns each
x
j
0
in the reference distribution to the center of mass it is sent to and thus outputs an approximated Monge map
T
i
. Then the Wasserstein Embedding for input
X
i
is calculated by [See Equation (7)].” The article discloses using a barycentric projection to identify centers of mass in reference distributions.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine He and Liu. He teaches a machine learning method to learn different task using feature embedding and distance metrics for classification by combining datasets and mapping them to a manifold. Liu teaches a machine learning method that is able to embed tasks and use distance metrices between datasets in a distribution space using optimal transport theory. One of ordinary skill would have motivation to combine a system that is able to embed features from different datasets, project and evaluate those values on a manifold with another machine learning system that uses Optimal Transport Theory as a distance metric instead of Euclidian metric used in He, “Fig. 3 summarizes the correlation diagrams between our proposed WTE distance and the forward/backward transferability on the aforementioned three task groups. WTE distance is negatively correlated with forward transferability, and positively correlated with catastrophic forgetting. In all scenarios, the correlation is strong and statistically significant, which confirms the efficacy of WTE distance as a measure of task similarities.” (Liu, Results, pp. 7).
Regarding claim 2, He discloses, “wherein the target labelled dataset includes more, fewer, or different labels than labels of one or more of the first and second training labelled datasets.” (Problem Description, pp. 2950; “For the learning process, the dataset D is randomly sampled several times according to the above settings to form the task set
{
D
1
T
,
D
2
T
,
…
D
n
T
}
as the training set. For each task, the support set is used as the labeled dataset, and the query set is used as the unlabeled dataset, and then the model is trained by predicting the categories of NQ unlabeled examples in the query set and minimizing the objective function.” The target data set, as interpreted above, is the training set of tasks
D
n
T
. This will contain fewer tasks from the dataset D because the task set is a subset of D.)
Regarding claim 12, He discloses, “A non-transitory machine-readable medium including instructions that, when executed by a machine, cause the machine to perform operations for generating a synthetic labelled machine learning (ML) dataset, the operations comprising:” (Datasets, pp. 2952; “Mini-ImageNet [9] is sampled from the ILSVRC-2012 (ImageNet) [56] dataset, which is used in many few-shot learning studies. It has 60,000 images, containing 100 classes of 600 images per class.” This article discloses the use of a computer-based dataset containing computer images for training and testing the proposed model. This data required the processing system to be a generic computing system containing process coupled to memory to execute computer instructions.) and (Implementation Details, pp. 2952; “We use ResNet-12 [58] as the backbone for the feature extraction and pre-train the backbones with reference to work [25], [53]. For the details of pre-training, we connect the backbone with a fully-connected classifier and do classification training using the train set of the datasets (Mini- ImageNet is 64 classification and Tiered-ImageNet is 351 classification). After training, we remove the fully-connected layer and select the model with the best performance of few-shot classification (5-way 1-shot) on the validation set.” This model also uses a machine learning model to perform the actions. This would require a generic processing system connected to memory to execute computer instructions.)
“defining or obtaining a target labelled dataset;” (Problem Description, pp. 2950; “In this paper, we follow a typical few-shot learning setting and define the learning process as several iterations of training tasks, with each task
D
T
being an N-way K-shot classification problem. … For the learning process, the dataset D is randomly sampled several times according to the above settings to form the task set
{
D
1
T
,
D
2
T
,
…
D
n
T
}
as the training set.” This article discloses a process which takes different datasets to produce a training dataset. This is interpreted to be the target dataset.)
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” (Problem Description, pp. 2950; “Specifically, given a base dataset D, a task
D
T
⊆
D
is randomly sampled from it,
D
T
consists of two sets: a support set
D
s
contains N classes with K samples per class, K usually being 1 or 5; and a query set
D
q
contains N same classes with Q samples per class.” The support set is interpreted to be the first dataset containing labeled data.)
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset, the second training labelled dataset comprising second samples associated with second labels, at least one of the first labels is different from all of the second labels;” (Problem Description, pp. 2950; “Specifically, given a base dataset D, a task
D
T
⊆
D
is randomly sampled from it,
D
T
consists of two sets: a support set
D
s
contains N classes with K samples per class, K usually being 1 or 5; and a query set
D
q
contains N same classes with Q samples per class. We denote them as
D
s
=
{
x
i
s
,
y
i
s
}
i
=
1
N
×
K
and
D
q
=
{
x
j
q
,
y
j
q
}
j
=
1
N
×
Q
, respectively, where x is the image, y is the corresponding label. Note that
D
T
=
D
s
∪
D
q
,
D
s
∩
D
q
=
∅
” The query set is interpreted to be the second dataset containing labeled data. As stated above the intersection of the support set and query set is empty indicating there are not shared labels and the two sets are distinctly different.)
“producing the synthetic labelled ML dataset by combining, based on distances between probability distribution representations of the first and second training datasets in the distribution space and the point, the first and second training labelled datasets, the synthetic labelled ML dataset comprises synthetic samples in a feature space of the target labelled dataset.” (Few-shot Learning on Sphere manifold, pp. 2951-2952; “For this method, the selection of the tangent point is crucial because the tangent space is more effective for the neighborhood of the tangent point. If the points are too far from the tangent point, the original geometric properties may be lost after Logarithmic mapping to the tangent space, and the distance is not representative. Therefore, combined with the setting of few-shot learning, we propose a method that adaptively selects the tangent point according to the samples, as shown in Fig. 2.” This model will generate different examples which it uses to train a machine learning model. This will actively select generated samples using the given datasets and attempt to learn new features.)
He fails to explicitly disclose, “determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;”, “determining an OT map from the target labelled dataset to the second training labelled dataset;”, and “identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and”.
However, Liu discloses, “determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;” (Optimal Transport Dataset Distance (OTDD), pp. 4; “Let
X
=
{
x
n
∈
R
d
}
n
=
1
N
be the input set with labels (classes)
Y
=
{
y
j
}
j
=
1
J
. For each j, let
C
y
i
⊆
X
denote the class with label
y
j
. Following the OTDD framework, let
T
=
x
i
,
y
j
∈
X
×
Y
x
i
∈
C
y
j
}
i
,
j
denote the set of datalabel pairs. OTDD encodes the label
y
j
as distribution
V
y
j
, where
V
y
j
=
1
C
y
j
∑
x
i
∈
C
y
j
δ
x
i
The ground distance in T is then defined by combining the Euclidean distance between the data points and the 2-Wasserstein distance between label distributions: [See Equation (8)] Based on this metric, the OT distance between two distributions
μ
i
and
μ
j
on T is [See Equation (9)] where
∏
(
μ
i
,
μ
j
)
denotes the set of transport plans between
μ
i
and
μ
j
.” This article discloses the use of determining the OTDD of two data sets. Liu teaches a method to concatenate two different datasets a modified OTDD method and Wasserstein embedding.) and (Figure 1, pp. 5; The Wasserstein embedding is seen on the left side of the figure.)
“determining an OT map from the target labelled dataset to the second training labelled dataset;” (Optimal Transport Dataset Distance (OTDD), pp. 4; “Let
X
=
{
x
n
∈
R
d
}
n
=
1
N
be the input set with labels (classes)
Y
=
{
y
j
}
j
=
1
J
. For each j, let
C
y
i
⊆
X
denote the class with label
y
j
. Following the OTDD framework, let
T
=
x
i
,
y
j
∈
X
×
Y
x
i
∈
C
y
j
}
i
,
j
denote the set of datalabel pairs. OTDD encodes the label
y
j
as distribution
V
y
j
, where
V
y
j
=
1
C
y
j
∑
x
i
∈
C
y
j
δ
x
i
The ground distance in T is then defined by combining the Euclidean distance between the data points and the 2-Wasserstein distance between label distributions: [See Equation (8)] Based on this metric, the OT distance between two distributions
μ
i
and
μ
j
on T is [See Equation (9)] where
∏
(
μ
i
,
μ
j
)
denotes the set of transport plans between
μ
i
and
μ
j
.” This article discloses the use of determining the OTDD of two data sets. Liu teaches a method to concatenate two different datasets a modified OTDD method and Wasserstein embedding.) and (Figure 1, pp. 5; The Wasserstein embedding is seen on the left side of the figure.)
“identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and” (Wasserstein Embedding (WE), pp. 4; “The optimal transport plan
π
i
*
is the minimizer of the above optimization problem, which is solved by linear program at cost
O
N
3
log
N
,
N
being the number of input samples. To avoid mass splitting, the barycentric projection (Wang et al. 2013) assigns each
x
j
0
in the reference distribution to the center of mass it is sent to and thus outputs an approximated Monge map
T
i
. Then the Wasserstein Embedding for input
X
i
is calculated by [See Equation (7)].” The article discloses using a barycentric projection to identify centers of mass in reference distributions.)
Regarding claim 13, He discloses, “wherein the target labelled dataset includes more, fewer, or different labels than labels of one or more of the first and second training labelled datasets.” (Problem Description, pp. 2950; “For the learning process, the dataset D is randomly sampled several times according to the above settings to form the task set
{
D
1
T
,
D
2
T
,
…
D
n
T
}
as the training set. For each task, the support set is used as the labeled dataset, and the query set is used as the unlabeled dataset, and then the model is trained by predicting the categories of NQ unlabeled examples in the query set and minimizing the objective function.” The target data set, as interpreted above, is the training set of tasks
D
n
T
. This will contain fewer tasks from the dataset D because the task set is a subset of D.)
Regarding claim 18, He discloses, “A system for generating a synthetic labelled machine learning (ML) dataset, the system comprising: processing circuitry; and a memory coupled to the processing circuitry, the memory including instructions that, when executed by the processing circuitry, cause the processing circuitry to perform operations comprising:” (Datasets, pp. 2952; “Mini-ImageNet [9] is sampled from the ILSVRC-2012 (ImageNet) [56] dataset, which is used in many few-shot learning studies. It has 60,000 images, containing 100 classes of 600 images per class.” This article discloses the use of a computer-based dataset containing computer images for training and testing the proposed model. This data required the processing system to be a generic computing system containing process coupled to memory to execute computer instructions.) and (Implementation Details, pp. 2952; “We use ResNet-12 [58] as the backbone for the feature extraction and pre-train the backbones with reference to work [25], [53]. For the details of pre-training, we connect the backbone with a fully-connected classifier and do classification training using the train set of the datasets (Mini- ImageNet is 64 classification and Tiered-ImageNet is 351 classification). After training, we remove the fully-connected layer and select the model with the best performance of few-shot classification (5-way 1-shot) on the validation set.” This model also uses a machine learning model to perform the actions. This would require a generic processing system connected to memory to execute computer instructions.)
“defining or obtaining a target labelled dataset;” (Problem Description, pp. 2950; “In this paper, we follow a typical few-shot learning setting and define the learning process as several iterations of training tasks, with each task
D
T
being an N-way K-shot classification problem. … For the learning process, the dataset D is randomly sampled several times according to the above settings to form the task set
{
D
1
T
,
D
2
T
,
…
D
n
T
}
as the training set.” This article discloses a process which takes different datasets to produce a training dataset. This is interpreted to be the target dataset.)
“obtaining a first training labelled dataset distinct from the target labelled dataset, the first training labelled dataset comprising first samples associated with first labels;” (Problem Description, pp. 2950; “Specifically, given a base dataset D, a task
D
T
⊆
D
is randomly sampled from it,
D
T
consists of two sets: a support set
D
s
contains N classes with K samples per class, K usually being 1 or 5; and a query set
D
q
contains N same classes with Q samples per class.” The support set is interpreted to be the first dataset containing labeled data.)
“obtaining a second training labelled dataset distinct from the target labelled dataset and the first training labelled dataset, the second training labelled dataset comprising second samples associated with second labels, at least one of the first labels is different from all of the second labels;” (Problem Description, pp. 2950; “Specifically, given a base dataset D, a task
D
T
⊆
D
is randomly sampled from it,
D
T
consists of two sets: a support set
D
s
contains N classes with K samples per class, K usually being 1 or 5; and a query set
D
q
contains N same classes with Q samples per class. We denote them as
D
s
=
{
x
i
s
,
y
i
s
}
i
=
1
N
×
K
and
D
q
=
{
x
j
q
,
y
j
q
}
j
=
1
N
×
Q
, respectively, where x is the image, y is the corresponding label. Note that
D
T
=
D
s
∪
D
q
,
D
s
∩
D
q
=
∅
” The query set is interpreted to be the second dataset containing labeled data. As stated above the intersection of the support set and query set is empty indicating there are not shared labels and the two sets are distinctly different.)
He fails to explicitly disclose, “determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;”, “determining an OT map from the target labelled dataset to the second training labelled dataset;”, and “identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and”.
However, Liu discloses, “determining an optimal transport (OT) map from the target labelled dataset to the first training labelled dataset;” (Optimal Transport Dataset Distance (OTDD), pp. 4; “Let
X
=
{
x
n
∈
R
d
}
n
=
1
N
be the input set with labels (classes)
Y
=
{
y
j
}
j
=
1
J
. For each j, let
C
y
i
⊆
X
denote the class with label
y
j
. Following the OTDD framework, let
T
=
x
i
,
y
j
∈
X
×
Y
x
i
∈
C
y
j
}
i
,
j
denote the set of datalabel pairs. OTDD encodes the label
y
j
as distribution
V
y
j
, where
V
y
j
=
1
C
y
j
∑
x
i
∈
C
y
j
δ
x
i
The ground distance in T is then defined by combining the Euclidean distance between the data points and the 2-Wasserstein distance between label distributions: [See Equation (8)] Based on this metric, the OT distance between two distributions
μ
i
and
μ
j
on T is [See Equation (9)] where
∏
(
μ
i
,
μ
j
)
denotes the set of transport plans between
μ
i
and
μ
j
.” This article discloses the use of determining the OTDD of two data sets. Liu teaches a method to concatenate two different datasets a modified OTDD method and Wasserstein embedding.) and (Figure 1, pp. 5; The Wasserstein embedding is seen on the left side of the figure.)
“determining an OT map from the target labelled dataset to the second training labelled dataset;” (Optimal Transport Dataset Distance (OTDD), pp. 4; “Let
X
=
{
x
n
∈
R
d
}
n
=
1
N
be the input set with labels (classes)
Y
=
{
y
j
}
j
=
1
J
. For each j, let
C
y
i
⊆
X
denote the class with label
y
j
. Following the OTDD framework, let
T
=
x
i
,
y
j
∈
X
×
Y
x
i
∈
C
y
j
}
i
,
j
denote the set of data label pairs. OTDD encodes the label
y
j
as distribution
V
y
j
, where
V
y
j
=
1
C
y
j
∑
x
i
∈
C
y
j
δ
x
i
The ground distance in T is then defined by combining the Euclidean distance between the data points and the 2-Wasserstein distance between label distributions: [See Equation (8)] Based on this metric, the OT distance between two distributions
μ
i
and
μ
j
on T is [See Equation (9)] where
∏
(
μ
i
,
μ
j
)
denotes the set of transport plans between
μ
i
and
μ
j
.” This article discloses the use of determining the OTDD of two data sets. Liu teaches a method to concatenate two different datasets a modified OTDD method and Wasserstein embedding.) and (Figure 1, pp. 5; The Wasserstein embedding is seen on the left side of the figure.)
“identifying, in a generalized geodesic hull (i) formed by the first and second training labelled datasets in a distribution space, (ii) defined by the OT maps, and (iii) that connects the first and second training labelled datasets with respect to the target dataset, a point proximate the target labelled dataset in the distribution space; and” (Wasserstein Embedding (WE), pp. 4; “The optimal transport plan
π
i
*
is the minimizer of the above optimization problem, which is solved by linear program at cost
O
N
3
log
N
,
N
being the number of input samples. To avoid mass splitting, the barycentric projection (Wang et al. 2013) assigns each
x
j
0
in the reference distribution to the center of mass it is sent to and thus outputs an approximated Monge map
T
i
. Then the Wasserstein Embedding for input
X
i
is calculated by [See Equation (7)].” The article discloses using a barycentric projection to identify centers of mass in reference distributions.)
Claims 3 and 14 are rejected under 35 U.S.C. 103 as being unpatentable over He and Liu in view of Choi et al, (Choi et al, “StarGAN: Unified Generative Adversarial Networks for Multi-Domain Image-to-Image Translation”, 2018, hereinafter “Choi”).
Regarding claim 3, Choi discloses, “wherein combining the first and second training labelled datasets includes representing labels of the first and second training labelled datasets as respective one-hot vectors of all labels in the first and second training labelled datasets.” (Mask Vector., pp. 8792; "To alleviate this problem, we introduce a mask vector m that allows StarGAN to ignore unspecified labels and focus on the explicitly known label provided by a particular dataset. In StarGAN, we use an n-dimensional one-hot vector to represent m, with n being the number of datasets. In addition, we define a unified version of the label as a vector [see equation {7)] where [·] refers to concatenation, and
c
i
represents a vector for the labels of the i-th dataset. The vector of the known label
c
i
can be represented as either a binary vector for binary attributes or a one-hot vector for categorical attributes. For the remaining n-1 unknown labels we simply assign zero values. In our experiments, we utilize the CelebA and Ra FD datasets, where n is two." This system is able to evaluate labeled and unlabeled datasets. Labels can be represented as one-hot vectors.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine He, Liu, and Choi. He teaches a machine learning method to learn different task using feature embedding and distance metrics for classification by combining datasets and mapping them to a manifold. Liu teaches a machine learning method that is able to embed tasks and use distance metrices between datasets in a distribution space using optimal transport theory. Choi teaches a machine learning method that is able to generate synthetic images using a Generative Adversarial network and use a discriminator to train the model. One of ordinary skill would have motivation to combine a system that is able to embed features from different datasets, project and evaluate those values on a manifold with another machine learning system that uses Optimal Transport Theory as a distance metric instead of Euclidian metric used in He, and use teachings of GAN architecture and use a generator and discriminator modules to generate and use the synthetic data to train a machine learning model, “In this paper, we proposed StarGAN, a scalable image-to- image translation model among multiple domains using a single generator and a discriminator. Besides the advantages in scalability, StarGAN generated images of higher visual quality compared to existing methods [15, 22, 32], owing to the generalization capability behind the multitask learning setting. In addition, the use of the proposed simple mask vector enables StarGAN to utilize multiple datasets with different sets of domain labels, thus handling all available labels from them." (Choi, Conclusion, pp. 8796).
Regarding claim 14, Choi discloses, “wherein combining the first and second training labelled datasets includes representing labels of the first and second training labelled datasets as respective one-hot vectors of all labels in the first and second training labelled datasets.” (Mask Vector., pp. 8792; "To alleviate this problem, we introduce a mask vector m that allows StarGAN to ignore unspecified labels and focus on the explicitly known label provided by a particular dataset. In StarGAN, we use an n-dimensional one-hot vector to represent m, with n being the number of datasets. In addition, we define a unified version of the label as a vector [see equation {7)] where [·] refers to concatenation, and
c
i
represents a vector for the labels of the i-th dataset. The vector of the known label
c
i
can be represented as either a binary vector for binary attributes or a one-hot vector for categorical attributes. For the remaining n-1 unknown labels we simply assign zero values. In our experiments, we utilize the CelebA and Ra FD datasets, where n is two." This system is able to evaluate labeled and unlabeled datasets. Labels can be represented as one-hot vectors.)
Claims 4, 10, 11, and 15 are rejected under 35 U.S.C. 103 as being unpatentable over He and Liu in view of Rout et al, (Rout et al, “GENERATIVE MODELING WITH OPTIMAL TRANSPORT MAPS”, Mar. 5th, 2022, hereinafter “Rout”).
Regarding claim 4, Rout discloses, “further comprising further training, using the synthetic labelled ML dataset, a pre-trained ML model that has been trained based on the target labelled dataset.” (Experiments, pp. 7; We evaluate our algorithm in generative modeling of the data distribution from a noise (§5.1) and unpaired image restoration task (§5.2). Technical details are given in Appendix B. Additionally, in Appendix B.4 we test our method on toy 2D datasets and evaluate it on theWasserstein-2 benchmark (Korotin et al., 2021b) in Appendix B.2. The code is in the supplementary material.” The methods proposed in this article use the generated maps to perform machine learning functions. This article teaches experiments where they used datasets and generated datasets to perform image restoration and data generation.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine He, Liu, and Rout. He teaches a machine learning method to learn different task using feature embedding and distance metrics for classification by combining datasets and mapping them to a manifold. Liu teaches a machine learning method that is able to embed tasks and use distance metrices between datasets in a distribution space using optimal transport theory. Rout teaches a model that is able to use Optimal transport to perform different machine learning functions such as image restoration and data generation. One of ordinary skill would have motivation to combine a system that is able to embed features from different datasets, project and evaluate those values on a manifold with another machine learning system that uses Optimal Transport Theory as a distance metric instead of Euclidian metric used in He, and use teaching form image generation machine learning models which implement Optimal Transport Theory to generate synthetic images which can be used to train machine learning models, “Our method fits OT maps for the embedded quadratic transport cost between probability distributions. Unlike predecessors, it scales well to high dimensions producing applications of OT maps directly in ambient spaces, such as spaces of images. The performance is comparable to other existing generative models while the complexity of training is similar to that of popular WGANs.” (Rout, Conclusion, pp. 9)
Regarding claim 10, Rout discloses, “wherein identifying the point proximate the target labelled dataset includes operating a quadratic problem solver based on a (2, v) transport metric.” (Equal Dimensions of Input and Output Distributions, pp. 4-5; “In this section, we
X
=
Y
=
R
D
and consider theWasserstein-2 distance (
W
2
), i.e., the optimal transport for the quadratic ground cost
c
x
,
y
=
1
2
x
-
y
2
. We use the dual form (4) to derive a saddle point problem the solution of which yields the OT map T*. We consider distributions
μ
,
ν
with finite second moments. We assume that for distributions
μ
,
ν
in view there exists a unique OT plan
π
* minimizing (3) and it is deterministic, i.e.,
π
*
=
i
d
R
D
,
T
*
#
μ
. Here T* is an OT map which minimizes (1).” The article discloses methods that take in consideration quadratic ground cost. This teaches the use of methods to determine different metrics.)
Regarding claim 11, Rout discloses “before obtaining the first or second labelled training datasets, receiving, from an application, a request for the synthetic labelled ML dataset; and” (Algorithm 1, pp. 7; This algorithm discloses the learning process for generating an optimal transport map between two sets of data. This will intake the datasets and produce a result. This algorithm must be initiated by the system in order to begin and execute. Therefore, using the broadest reasonable interpretation this algorithm must receive a request from the computer system and the proper inputs in order to execute.)
“responsive to producing the synthetic labelled ML dataset, providing the synthetic labelled ML dataset to the application.” (Algorithm 1, pp. 7; This algorithm discloses the learning process for generating an optimal transport map between two sets of data. Once the algorithm in initiated, the algorithm will execute and create a generated map of the two datasets and return it as
G
θ
.
)
Regarding claim 15, Rout discloses, “wherein the operations further comprise further training, using the synthetic labelled ML dataset, a pre-trained ML model that has been trained based on the target labelled dataset.” (Experiments, pp. 7; We evaluate our algorithm in generative modeling of the data distribution from a noise (§5.1) and unpaired image restoration task (§5.2). Technical details are given in Appendix B. Additionally, in Appendix B.4 we test our method on toy 2D datasets and evaluate it on theWasserstein-2 benchmark (Korotin et al., 2021b) in Appendix B.2. The code is in the supplementary material.” The methods proposed in this article use the generated maps to perform machine learning functions. This article teaches experiments where they used datasets and generated datasets to perform image restoration and data generation.)
Claims 5, 6, 9, 16, and 17 are rejected under 35 U.S.C. 103 as being unpatentable over He and Liu in view of Solomon et al, (Solomon et al, “Convolutional Wasserstein Distances: Efficient Optimal Transportation on Geometric Domains”, 2015, hereinafter “Solomon”).
Regarding claim 5, Solomon discloses, “wherein identifying the point includes performing a barycentric projection of the target labelled dataset onto the geodesic hull.” (Wasserstein Barycenter’s, pp. 5; “Algorithm 2 documents the barycenter method. It initializes all the
π
i
'
s
to
H
t
by taking
v
i
=
w
i
=
1
for all i and then alternatingly projects using the formulas above. The only operations needed are applications of
H
t
and elementwise arithmetic. We never need to store the matrix of
H
t
explicitly and instead apply it iteratively; this structure is key when
H
t
represents a heat kernel obtained by solving a linear system or convolution over an image.” This article discloses different methods using Convolutional Wasserstein distance. One method disclosed the uses barycenter’s as points to summarize collections of probability distributions. Algorithm 2 discloses that projection process.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine He, Liu, and Solomon. He teaches a machine learning method to learn different task using feature embedding and distance metrics for classification by combining datasets and mapping them to a manifold. Liu teaches a machine learning method that is able to embed tasks and use distance metrices between datasets in a distribution space using optimal transport theory. Solomon teaches a model that is able to implement optimal transport theory to 3D modeling and image generation using a heat kernel. One of ordinary skill would have motivation to combine a system that is able to embed features from different datasets, project and evaluate those values on a manifold with another machine learning system that uses Optimal Transport Theory as a distance metric instead of Euclidian metric used in He, and use the teaching of a machine learning model that is able to also use OPM to improve 3D modeling and generate synthetic data, "We have demonstrated the breadth of applications enabled by this framework, from rendering to image processing to geometry. Modeling via probability distributions is natural for these and other problems, and we foresee applications across several additional disciplines. Having reduced the cost of experimenting with transportation models, future studies now may incorporate transportation into graphics applications including processing of volumetric data, caustic design, dimensionality reduction, and simulation." (Solomon, Discussion and Conclusion, pp. 9).
Regarding claim 6, Solomon discloses, “wherein the barycentric projection includes projection of sample data and separate label data.” (Algorithm 2, pp. 7; This algorithm discloses the Wasserstein-barycenter using a Bregman projection. This will project data onto
C
1
and
C
2
.)
Regarding claim 9, Solomon discloses, “wherein identifying the point proximate the target label led dataset in the dataset space includes determining the point in the generalized geodesic hull that is closest to the target labelled dataset.” (Wasserstein Propagation, pp. 6; “Propagation encapsulates many other optimizations in Wasserstein space. Fig. 5 illustrates two examples. The convolutional barycenter problem (§6.1) is exactly propagation where G is a star graph, with vertices in
V
0
on the spokes and the unknown distribution μ associated with the center.” This system can be used to determine Wasserstein distance. This can include at using points in distributed space and the distances from that center point.)
Regarding claim 16, Solomon discloses, “wherein identifying the point includes performing a barycentric projection of the target labelled dataset onto the geodesic hull.” (Wasserstein Barycenter’s, pp. 5; “Algorithm 2 documents the barycenter method. It initializes all the
π
i
'
s
to
H
t
by taking
v
i
=
w
i
=
1
for all i and then alternatingly projects using the formulas above. The only operations needed are applications of
H
t
and elementwise arithmetic. We never need to store the matrix of
H
t
explicitly and instead apply it iteratively; this structure is key when
H
t
represents a heat kernel obtained by solving a linear system or convolution over an image.” This article discloses different methods using Convolutional Wasserstein distance. One method disclosed the uses barycenter’s as points to summarize collections of probability distributions. Algorithm 2 discloses that projection process.)
Regarding claim 17, Solomon discloses, “wherein the barycentric projection includes projection of sample data and separate label data.” (Algorithm 2, pp. 7; This algorithm discloses the Wasserstein-barycenter using a Bregman projection. This will project data onto
C
1
and
C
2
.)
Claims 7, 8, 19, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over He and Liu in view of Huh et al, (Huh et al, “OT-driven Multi-Domain Unsupervised Ultrasound Image Artifact Removal using a Single CNN”, 2020, hereinafter “Huh”).
Regarding claim 7, Huh discloses, “wherein determining the OT map includes operating an OT neural map that includes three classifiers, a label classifier, a discriminator, and a feature classifier.” (Neural Network Implementation, pp. 6; “Additionally, there are also two discriminators
D
φ
and
D
ψ
as shown in Fig. 3. Specifically,
D
φ
tries to find the difference between the true image x and the generated image
G
Θ
(
y
)
, whereas
D
ψ
attempts to find the fake measurement data that are generated by the synthetic measurement procedure
F
ϕ
(
x
)
. Finally, we have the domain classifier
K
η
to distinguish between deconvolution and despeckled images. In fact, the domain classifier and discriminators are both classifiers, so their structure share many commonalities. Therefore, as shown in Fig. 4(b), the discriminator
D
φ
and the classifier
K
η
are implemented using a same network architecture with double output headers composed of a domain classifier or discriminator.” Figure 3 discloses the architecture of this model. This model contains different classifiers as well as discriminator.)
It would have been obvious to one of ordinary skill in the art before the effective filing date of the present application to combine He, Liu, and Huh. He teaches a machine learning method to learn different task using feature embedding and distance metrics for classification by combining datasets and mapping them to a manifold. Liu teaches a machine learning method that is able to embed tasks and use distance metrices between datasets in a distribution space using optimal transport theory. Huh teaches a machine learning model that is able to translate data and use it to further improve and train models. One of ordinary skill would have motivation to combine a system that is able to embed features from different datasets, project and evaluate those values on a manifold with another machine learning system that uses Optimal Transport Theory as a distance metric instead of Euclidian metric used in He, and use the teachings from a machine learning system that is able to utilize translated and altered data for training, "Fig. 5 shows the comparison results using various algorithms. We show the in vivo images from the left thyroid and right thyroid. As shown in the figure, all the deconvolution output images show the improved contrast. Specifically, in the magnified images, it is easy to recognize that the deconvolution images have sharper structure. While the StarGAN output has slightly improved compared to the input image, the output from the proposed method is closer to the target image. Moreover, the side-to-side comparison with supervised and CycleGAN approach show that the proposed method provides qualitatively comparable results. However, it is remarkable that the deconvolution and despeckled images using supervised learning and CycleGAN are generated from independent network trained separately, whereas the proposed method generated both images with single generator." (Qualitative Results, Huh, pp. 8).
Regarding claim 8, Huh discloses, wherein a discriminator loss of the discriminator is independent of the labels.” (Baseline Algorithms, pp. 7; “For CycleGAN training, we used the same generator architecture as the proposed method. The discriminator also has same architecture except the last layer. There is no domain classification loss in CycleGAN. We also used WGAN with gradient penalty technique [20] similar to the proposed method.” This method teaches the use of the discriminator. This system uses labeled datasets.)
Regarding claim 19, Huh discloses, “wherein determining the OT map includes operating an OT neural map that includes three classifiers, a label classifier, a discriminator, and a feature classifier.” (Neural Network Implementation, pp. 6; “Additionally, there are also two discriminators
D
φ
and
D
ψ
as shown in Fig. 3. Specifically,
D
φ
tries to find the difference between the true image x and the generated image
G
Θ
(
y
)
, whereas
D
ψ
attempts to find the fake measurement data that are generated by the synthetic measurement procedure
F
ϕ
(
x
)
. Finally, we have the domain classifier
K
η
to distinguish between deconvolution and despeckled images. In fact, the domain classifier and discriminators are both classifiers, so their structure share many commonalities. Therefore, as shown in Fig. 4(b), the discriminator
D
φ
and the classifier
K
η
are implemented using a same network architecture with double output headers composed of a domain classifier or discriminator.” Figure 3 discloses the architecture of this model. This model contains different classifiers as well as discriminator.)
Regarding claim 20, Huh discloses, “wherein a discriminator loss of the discriminator is independent of the labels”. (Baseline Algorithms, pp. 7; “For CycleGAN training, we used the same generator architecture as the proposed method. The discriminator also has same architecture except the last layer. There is no domain classification loss in CycleGAN. We also used WGAN with gradient penalty technique [20] similar to the proposed method.” This method teaches the use of the discriminator. This system uses labeled datasets.)
Conclusion
THIS ACTION IS MADE FINAL. Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PAUL MICHAEL GALVIN-SIEBENALER whose telephone number is (571)272-1257. The examiner can normally be reached Monday - Friday 8AM to 5PM.
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, Viker Lamardo can be reached at (571) 270-5871. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/PAUL M GALVIN-SIEBENALER/Examiner, Art Unit 2147
/ERIC NILSSON/Primary Examiner, Art Unit 2151
Prosecution Insights