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 .
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.
According to the first part of the analysis, in the instant case, claims 1-7 are directed to a method, claims 8-14 are directed to a non-transitory computer readable medium, and claims 15-20 are directed to a system. Each of these claims fall within one of the four statutory categories (i.e., process, machine, manufacture, or composition of matter).
For claim 1,
Step 2A Prong One
identifying a plurality of feature type candidates corresponding to the plurality of data subsets; … scoring each of the different sets of feature type candidates based on respective accuracies, relative to the dataset, of each first machine learning model of the plurality of first machine learning models that respectively corresponds to each different set of feature type candidates; selecting a final set of feature types from the different sets of feature type candidates based on the scores of the different sets of feature type candidates;
(These steps for identifying feature type candidates, scoring sets of feature type candidates, and selecting a final set of feature types can feasibly be performed in the mind and are considered mental processes)
Step 2A Prong Two
A method comprising: accessing a dataset including a plurality of data subsets;… building a plurality of first machine learning models using different sets of feature type candidates;… and training a second machine learning model using a labeled dataset that is generated by applying the final set of feature types to the dataset.
(These steps for accessing data, building machine learning models, and training the models are insignificant extra solution activity. See MPEP § 2106.05(g))
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes such as identifying, scoring, and selecting while the additional elements are insignificant extra-solution activity recited at a high degree of generality.
For claim 2,
Step 2A Prong One
(Claim 2 depends on claim 1, which has been determined to recite abstract ideas including mental processes. Therefore, claim 2 also recites an abstract idea.)
Step 2A Prong Two
wherein the feature type candidates of the plurality of feature type candidates are identified based on an inference type machine learning analysis of the dataset.
(This step for identifying the feature type candidates using a machine learning analysis a high degree of generality is mere instructions to apply an exception. See MPEP § 2106.05(f))
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes while the additional elements are mere-instructions to apply the exception.
For claim 3,
Step 2A Prong One
wherein the method further comprises filtering the feature type candidates based on respective probability values corresponding to likelihoods that the feature type candidates are actual feature types corresponding to their respective data subsets.
(This step for filtering candidates can feasibly be performed in the mind and is considered a mental process)
Step 2A Prong Two
The claim does not include additional elements, when considered separately and in combination, that integrate the judicial exception into a practical application.
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes without any technological improvement or inventive step.
For claim 4,
Step 2A Prong One
(Claim 4 depends on claim 1, which has been determined to recite abstract ideas including mental processes. Therefore, claim 4 also recites an abstract idea.)
Step 2A Prong Two
wherein the different sets of feature type candidates are based on different combinations of feature type candidates corresponding to different data subsets.
(This step for selecting a particular data source for the sets of feature type candidates is insignificant extra solution activity. See MPEP § 2106.05(g))
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes while the additional elements are insignificant extra-solution activity recited at a high degree of generality.
For claim 5,
Step 2A Prong One
wherein the different sets of feature type candidates are selected based on respective combined probability values for each set of feature type candidates, the respective combined probability values being determined based on respective individual probability values of individual feature type candidates included in a corresponding set of feature type candidates,
(These steps for determining probabilities and selecting based on probabilities are mental processes)
Step 2A Prong Two
the respective individual probability values corresponding to likelihoods that the corresponding feature type candidates are actual feature types corresponding to their respective data subsets.
(This step for selecting a particular type or source for the individual probabilities is insignificant extra solution activity. See MPEP § 2106.05(g))
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes while the additional elements are insignificant extra-solution activity recited at a high degree of generality.
For claim 6,
Step 2A Prong One
(Claim 6 depends on claim 1, which has been determined to recite abstract ideas including mental processes. Therefore, claim 6 also recites an abstract idea.)
Step 2A Prong Two
wherein the building of the plurality of first machine learning models includes training the plurality of first machine learning models using data sampled from the dataset.
(This step for building machine learning models through training is insignificant extra solution activity. See MPEP § 2106.05(g))
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes while the additional elements are insignificant extra-solution activity recited at a high degree of generality.
For claim 7,
Step 2A Prong One
further comprising determining the respective accuracies of the plurality of first machine learning models based on data sampled from the dataset that is used as validation data for the plurality of first machine learning models.
(This step for determining accuracies is a mental process)
Step 2A Prong Two
The claim does not include additional elements, when considered separately and in combination, that integrate the judicial exception into a practical application.
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes without any technological improvement or inventive step.
For claim 8,
Step 2A Prong One
identifying a plurality of feature type candidates corresponding to the plurality of data subsets; … scoring each of the different sets of feature type candidates based on respective accuracies, relative to the dataset, of each first machine learning model of the plurality of first machine learning models that respectively corresponds to each different set of feature type candidates; selecting a final set of feature types from the different sets of feature type candidates based on the scores of the different sets of feature type candidates;
(These steps for identifying feature type candidates, scoring sets of feature type candidates, and selecting a final set of feature types can feasibly be performed in the mind and are considered mental processes)
Step 2A Prong Two
A system, comprising: one or more processors; and one or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause the system to perform operations, the operations comprising:
(This step for implementing the methods using generic hardware is mere instructions to apply an exception. See MPEP § 2106.05(f))
A method comprising: accessing a dataset including a plurality of data subsets;… building a plurality of first machine learning models using different sets of feature type candidates;… and training a second machine learning model using a labeled dataset that is generated by applying the final set of feature types to the dataset.
(These steps for accessing data, building machine learning models, and training the models are insignificant extra solution activity. See MPEP § 2106.05(g))
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes such as identifying, scoring, and selecting while the additional elements are mere instructions to apply an exception and insignificant extra-solution activity recited at a high degree of generality.
For claims 9-14,
Claims 9-14 are non-transitory computer readable medium claims directly corresponding to method claims 2-7, and are rejected under 101 using similar reasoning.
For claim 15,
Step 2A Prong One
identifying a plurality of feature type candidates corresponding to the plurality of data subsets; … scoring each of the different sets of feature type candidates based on respective accuracies, relative to the dataset, of each first machine learning model of the plurality of first machine learning models that respectively corresponds to each different set of feature type candidates; selecting a final set of feature types from the different sets of feature type candidates based on the scores of the different sets of feature type candidates;
(These steps for identifying feature type candidates, scoring sets of feature type candidates, and selecting a final set of feature types can feasibly be performed in the mind and are considered mental processes)
Step 2A Prong Two
A system, comprising: one or more processors; and one or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause the system to perform operations, the operations comprising:
(This step for implementing the methods using generic hardware is mere instructions to apply an exception. See MPEP § 2106.05(f))
A method comprising: accessing a dataset including a plurality of data subsets;… building a plurality of first machine learning models using different sets of feature type candidates;… and training a second machine learning model using a labeled dataset that is generated by applying the final set of feature types to the dataset.
(These steps for accessing data, building machine learning models, and training the models are insignificant extra solution activity. See MPEP § 2106.05(g))
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes such as identifying, scoring, and selecting while the additional elements are mere instructions to apply an exception and insignificant extra-solution activity recited at a high degree of generality.
For claim 16,
Step 2A Prong One
and the operations further comprise filtering the feature type candidates based on respective probability values corresponding to likelihoods that the feature type candidates are actual feature types corresponding to their respective data subsets.
(This step for filtering candidates can feasibly be performed in the mind and is considered a mental process)
Step 2A Prong Two
wherein: the feature type candidates of the plurality of feature type candidates are identified based on an inference type machine learning analysis of the dataset;
(This step for identifying the feature type candidates using a machine learning analysis a high degree of generality is mere instructions to apply an exception. See MPEP § 2106.05(f))
Step 2B
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception because, when considered individually and in combination, they do not add significantly more (also known as an inventive concept) to the exception. The claim recites mental processes while the additional elements are mere instructions to apply an exception recited at a high degree of generality.
For claims 17-20
Claims 17-20 are system claims directly corresponding to method claims 4-7, and are rejected using similar reasoning.
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.
Claim(s) 1, 4, 6-7, 8, 11, 13-14, 15, 17, 19-20 is/are rejected under 35 U.S.C. 103 as being unpatentable over Laurence Louis Eric Rouesnel (hereinafter Rouesnel) (US 11593705 B1, 2023-02-28) in view of Nitin S. Sharma et al. (hereinafter Sharma) (US 20220374784 A1, 2022-11-24).
Regarding claim 1, Rouesnel teaches;
A method comprising: accessing a dataset including a plurality of data subsets;
([col 2, ln 10-13] a feature engineering engine utilizes multiple analyzers to produce different sets of data facts describing the values of each column in a dataset.)
NOTE: Rouesnel teaches accessing a dataset including a plurality of data subsets (columns).
identifying a plurality of feature type candidates corresponding to the plurality of data subsets;
([col 6, ln 48-58] Embodiments may thus implement a variety of processing strategies 120A-120H to apply different strategies for feature set generation. For example, a first processing strategy 120A may implement the following logic: if, according to the data statements, a column is at least 50% phone numbers, ... treat the column as including phone number data values. As another example, another processing strategy could implement relatively simple logic where if at least 99% of values of a column are of a particular type (e.g., integers), the column is treated as that type (e.g., integers))
NOTE: Rouesnel teaches identifying potential (i.e., candidate) feature types for each column / subset, such as phone numbers or integers. Thus, Rouesnel teaches identifying a plurality of feature type candidates (phone numbers, integers, etc.) corresponding to the plurality of data subsets / columns.
building a plurality of first machine learning models using different sets of feature type candidates;
([col 6, ln 48-50] Embodiments may thus implement a variety of processing strategies 120A-120H to apply different strategies for feature set generation… [col 3, ln 29-32] Embodiments disclosed herein provide a flexible and powerful way to analyze datasets to determine how to transform the data into useful feature sets that can be used to train high-quality ML models.)
NOTE: As previously taught, each column in a generated feature set is represented using one of the aforementioned feature type candidates. Each feature set comprises a plurality of columns each having a corresponding feature type candidate, and thus, each feature set comprises a set of feature type candidates.
Each feature set can comprise a different set of feature type candidates, based on the different processing strategies for generating each feature set.
The feature sets are used to train / build ML models (‘models’ is plural, indicating a plurality).
Thus, Rouesnel teaches training / building a plurality of first machine learning models using different sets of feature type candidates
Rouesnel fails to explicitly teach but Sharma teaches;
scoring each of the different sets of feature type candidates based on respective accuracies, relative to the dataset, of each first machine learning model of the plurality of first machine learning models that respectively corresponds to each different set of feature type candidates; selecting a final set of feature types from the different sets of feature type candidates based on the scores of the different sets of feature type candidates;
([0019] As will be appreciated by one of skill in the art, the term “data feature” (or simply “feature”) as used herein refers to an attribute of a data sample)
PNG
media_image1.png
293
586
media_image1.png
Greyscale
NOTE: Each feature represents an attribute / type for a column of data in the dataset. Thus, each feature can alternatively be considered a feature type under the broadest reasonable interpretation.
([Abstract] the feature-selection operations include processing the training dataset based on an optimization model, where an objective function utilized in the optimization model utilizes performance feedback information corresponding to machine learning models that are trained based on candidate feature sets. Based on the feature-selection operation, the computer system may generate an output value that indicates the subset of features to include in the reduced feature set… [0051] In various embodiments, the disclosed techniques include selecting the reduced feature set as the one that produces a machine learning model, trained using the reduced feature set, which results in the highest accuracy score.)
NOTE: Sharma teaches training a plurality of machine learning models each corresponding to a plurality of candidate feature / feature type sets, then selecting the best, ‘reduced’ feature type set as the candidate feature type set that results in the highest accuracy score for its respective model on the dataset, indicating that each of candidate feature type sets are scored based on their accuracy scores.
Thus, Sharma teaches scoring each of the different sets of feature type candidates based on respective accuracies, relative to the dataset, of each first machine learning model of the plurality of first machine learning models that respectively corresponds to each different set of feature type candidates, and selecting a final set of feature types (the selected reduced feature type set) from the different sets of feature type candidates (from the plurality of feature type candidates) based on the scores of the different sets of feature type candidates (based on the set having the highest accuracy score).
and training a second machine learning model using a labeled dataset that is generated by applying the final set of feature types to the dataset.
([0037] In the depicted embodiment, training dataset 110 includes data samples 112A-112N, each of which includes a corresponding feature vector 202A-202N and a label 204A-204N (respectively).)
NOTE: Sharma teaches the training dataset being labeled.
([0064] In various embodiments, method 500 includes using the reduced feature set to train one or more machine learning models. For example, in some embodiments, method 500 further includes generating an updated training dataset that includes data values for the subset of features that are included in the reduced feature set)
NOTE: Sharma teaches using the aforementioned reduced / final feature type set to generate an updated [labeled] training dataset, which is used to train at least one additional / second machine learning model.
Thus, Sharma teaches training a second machine learning model using a labeled dataset (the labeled, updated training dataset) that is generated by applying the final set of feature types to the dataset.
OBVIOUSNESS TO COMBINE SHARMA WITH ROUESNEL:
Rouesnel and Sharma are both analogous art to the present disclosure as they pertain to processing feature data used to train machine learning models.
Rouesnel teaches the base candidate feature type identification for each data subset and generating models using different sets of candidate feature types, while Sharma provides a means for selecting a final feature set based on corresponding model accuracies, and using the final set to train another machine learning model.
Additionally, Sharma states;
([0003] The feature-selection process aims to reduce the number of data features included in the training dataset, which may provide various improvements to the training process (e.g., reducing the training time) and the resulting machine learning models (e.g., improving the accuracy of the model, reducing overfitting, etc.).)
NOTE: Sharma indicates that the feature selection methods of their disclosure can lead to various improvements, such as reducing training time of models, improving accuracy, and reducing overfitting.
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the process of Sharma to identify the final set of feature types of the sets of candidate feature types taught by Rouesnel, to reduce training time, improve accuracy, and reduce overfitting of resulting models using the final set.
Regarding claim 4, Rouesnel teaches;
wherein the different sets of feature type candidates are based on different combinations of feature type candidates corresponding to different data subsets.
([col 6, ln 48-58] Embodiments may thus implement a variety of processing strategies 120A-120H to apply different strategies for feature set generation. For example, a first processing strategy 120A may implement the following logic: if, according to the data statements, a column is at least 50% phone numbers, ... treat the column as including phone number data values. As another example, another processing strategy could implement relatively simple logic where if at least 99% of values of a column are of a particular type (e.g., integers), the column is treated as that type (e.g., integers))
NOTE: Each feature set comprises of a set of columns each having a candidate feature type (phone numbers, integers, etc.), and thus comprises a set of candidate feature types. This excerpt details that each distinct feature set can use different logic when assigning candidate feature types to columns / subsets, thus, each feature set can have a different combination of feature type candidates corresponding to each different column / data subset.
Thus, Rouesnel teaches the different sets of feature type candidates (the different sets of candidate feature types associated with a given feature set) are based on different combinations of feature type candidates (each feature set can comprise a different combination of candidate feature types) corresponding to different data subsets (each feature set can assign different candidate feature types to each column / data subset).
Regarding claim 6, Rouesnel teaches;
wherein the building of the plurality of first machine learning models includes training the plurality of first machine learning models using data sampled from the dataset.
([col 3, ln 29-32] Embodiments disclosed herein provide a flexible and powerful way to analyze datasets to determine how to transform the data into useful feature sets that can be used to train high-quality ML models.)
NOTE: Rouesnel teaches training / building a plurality of ML models using feature sets of data sampled from a dataset.
Thus, Rouesnel teaches building the plurality of first machine learning models by training the plurality of first machine learning models using data sampled from the dataset.
Regarding claim 7, Rouesnel fails to explicitly teach but Sharma teaches;
further comprising determining the respective accuracies of the plurality of first machine learning models based on data sampled from the dataset that is used as validation data for the plurality of first machine learning models.
([0197] model evaluation module… evaluate[s] the performance of the machine learning models 1005 across multiple different test splits of test dataset 1010… [0198] the test dataset 1010 itself may be a subset of the training dataset 110.)
NOTE: Sharma teaches using a model evaluation module to evaluate performances of the previously taught ML models using the test dataset. The test dataset can be a subset of the training dataset held back during the training phase; thus, the test dataset can reasonably be considered validation data under the broadest reasonable interpretation.
([0199] Note that the model evaluation module 1006 may evaluate the machine learning model 1005 using various suitable metrics… may include any other suitable performance metric, such as accuracy)
NOTE: Sharma teaches determining the accuracies of the previously taught plurality of machine learning models using the model evaluation module.
Thus, Sharma teaches determining the respective accuracies of the plurality of first machine learning models based on data sampled from the dataset that is used as validation data (test splits of the test dataset, which is a sampled from the training dataset, and can be considered validation data under BRI) for the plurality of first machine learning models.
OBVIOUSNESS:
Using the same reasoning from claim 1, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the process of Sharma to identify the final set of feature types of the sets of candidate feature types taught by Rouesnel, to reduce training time, improve accuracy, and reduce overfitting of resulting models using the final set.
Regarding claim 8,
Claim 8 is a non-transitory computer readable media claim that is substantially similar to method claim 1, with one added limitation, which is taught by Rouesnel;
One or more non-transitory computer-readable media storing instructions that, in response to being executed by one or more processors, cause a system to perform operations, the operations comprising:
([col 10, ln 32-37] The code is stored on a computer-readable storage medium, for example, in the form of a computer program comprising instructions executable by one or more processors. The computer-readable storage medium is non-transitory.)
NOTE: The remaining limitations are substantially similar to claim 1 and are taught using the same reasoning.
Regarding claims 11, 13-14
Claims 11, 13-14 are non-transitory computer readable media claims that are substantially similar to method claims 4, 6-7, respectively, and are taught using the same reasoning as their corresponding method claims.
Regarding claim 15,
Claim 15 is a system claim that is substantially similar to method claim 1, with one added limitation, which is taught by Rouesnel;
A system, comprising: one or more processors; and one or more non-transitory computer-readable storage media configured to store instructions that, in response to being executed, cause the system to perform operations, the operations comprising:
([col 10, ln 32-37] The code is stored on a computer-readable storage medium, for example, in the form of a computer program comprising instructions executable by one or more processors. The computer-readable storage medium is non-transitory.)
NOTE: The remaining limitations are substantially similar to claim 1 and are taught using the same reasoning.
Regarding claims 17, 19-20
Claims 17, 19-20 are system claims that are substantially sim ilar to method claims 4, 6-7, respectively, and are taught using the same reasoning as their corresponding method claims.
Claim(s) 2-3, 5, 9-10, 12, 16, 18 is/are rejected under 35 U.S.C. 103 as being unpatentable over Rouesnel in view of Sharma as applied to claims 1, 8, and 15 above, and further in view of Dan Zhang et al. (hereinafter Zhang) (“Sato: Contextual Semantic Type Detection in Tables”, 2020-06-03).
Regarding claim 2, Rouesnel and Sharma fail to teach but Zhang teaches;
wherein the feature type candidates of the plurality of feature type candidates are identified based on an inference type machine learning analysis of the dataset.
([pg. 1] In response, recent work [22] introduced Sherlock, a deep learning model for semantic type detection trained on a massive table corpora [21].)
NOTE: The Sherlock model is a machine learning model used for type detection of a dataset.
[pg. 2]
PNG
media_image2.png
420
897
media_image2.png
Greyscale
NOTE: Zhang teaches identifying a possible / candidate feature type, t_i, of a plurality of feature type candidates (t_1, … t_m), for a given column, c_i, using a prediction / inference from a machine learning model (the Sherlock model, for example).
Thus, Zhang teaches feature type candidates of a plurality of feature type candidates are identified based on an inference type machine learning analysis of the dataset (the dataset being the tabular data disclosed by Zhang).
OBVIOUSNESS TO COMBINE ZHANG WITH ROUESNEL AND SHARMA:
Zhang is analogous art to the present disclosure as it pertains to identifying candidate feature types using machine learning inference.
Rouesnel teaches the base of using identified feature type candidates corresponding to data subsets to generate machine learning models, while Zhang provides a means of identifying feature type candidates using machine learning inference / prediction.
By using machine learning to identify feature type candidates, Zhang provides a way to automate the process of identifying feature type candidates.
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the process disclosed by Zhang to identify feature type candidates to be used in the processes disclosed by Rouesnel, to automate the process of identifying feature type candidates.
Regarding claim 3, Rouesnel and Sharma fail to teach but Zhang teaches;
wherein the method further comprises filtering the feature type candidates based on respective probability values corresponding to likelihoods that the feature type candidates are actual feature types corresponding to their respective data subsets.
([pg. 5] unary potential of a semantic type for a given column can be considered the probability of that semantic type based on the values of the column...., the unary potentials calculate column-wise prediction scores, which are used to select semantic type candidates for each column)
NOTE: Zhang teaches a unary potential for each semantic type (candidate feature type) of a column (data subset) which is a probability indicating the likelihood of the type being the actual type corresponding to a column.
The unary potentials are used to select specific semantic type candidates (candidate feature types) for each column, indicating that other semantic type candidates are not selected, i.e., filtered out.
Thus, Zhang teaches filtering the feature type candidates (semantic type candidates) based on respective probability values (unary potentials) corresponding to likelihoods that the feature type candidates are actual feature types corresponding to their respective data subsets (columns).
OBVIOUSNESS:
Rouesnel provides the base of identifying a plurality of candidate feature types for each column, while Zhang provides a method of selecting / filtering the candidate types for each column based on probabilistic values representing the likelihood for a given column type being the actual type.
A person of ordinary skill in the art, before the effective filing date of the claimed invention, would have been motivated to incorporate the type inference and unary-potential scoring of Zhang into the column analysis process of Rouesnel to provide more probabilistically relevant candidate feature types, thereby enabling Rouesnel’s interpretation engines to generate more appropriate feature engineering pipelines for machine learning.
Regarding claim 5; Rouesnel and Sharma fail to teach but Zhang teaches;
wherein the different sets of feature type candidates are selected based on respective combined probability values for each set of feature type candidates,
[pg. 5]
PNG
media_image3.png
25
265
media_image3.png
Greyscale
PNG
media_image4.png
133
1073
media_image4.png
Greyscale
NOTE: Zhang teaches a set (sequence) of columns c with each column in c having potential type, where each type can be considered a feature type candidate (potential / candidate type for a column of feature data), and the set (sequence) of types t associated with the set of columns c can be considered a set of feature type candidates. For a given set of columns, c, a set of feature type candidates, t, is selected if it has the highest P(t|c) (which can be considered a combined probability value, further explained below) compared to the other sets of feature type candidates.
Thus, Zhang teaches different sets of feature type candidates (t) are selected based on respective combined probability values for each set of feature type candidates (P(t|c)).
the respective combined probability values being determined based on respective individual probability values of individual feature type candidates included in a corresponding set of feature type candidates, the respective individual probability values corresponding to likelihoods that the corresponding feature type candidates are actual feature types corresponding to their respective data subsets.
([pg. 5] unary potential of a semantic type for a given column can be considered the probability of that semantic type based on the values of the column....)
NOTE: As previously taught, a unary potential represents the individual probability that a feature type candidate t_i is the actual feature type corresponding to its respective data subset (column) c_i. Thus, the unary potentials can be considered respective individual probability values of individual feature type candidates.
[pg. 5]
PNG
media_image3.png
25
265
media_image3.png
Greyscale
PNG
media_image4.png
133
1073
media_image4.png
Greyscale
NOTE: The combined probability value P(t|c) is determined based on a combination of respective individual probability values (sum of unary potentials (t_i, c_i)) included in a corresponding set of feature type candidates (the set of types, t).
Thus, Zhang teaches the respective combined probability values (P(t|c)) being determined based on respective individual probability values (sum of unary potentials (t_i, c_i)) of individual feature type candidates (t_i) included in a corresponding set of feature type candidates (t), the respective individual probability values (unary potentials) corresponding to likelihoods that the corresponding feature type candidates are actual feature types corresponding to their respective data subsets.
OBVIOUSNESS:
Rouesnel provides the base of having different sets of feature type candidates, while Zhang provides a means for selecting the different sets of feature type candidates based on probability values.
Zhang further states;
([pg. 5] Sato uses a linear-chain CRF to model the dependencies between columns types given their values. (b) For each column, Sato plugs in the column-wise prediction scores for each type as the unary potentials of the corresponding node in the CRF model. Then Sato learns the pairwise potential through backpropagation updates using stochastic gradient descent, maximizing the posterior probability P(t|c).)
NOTE: Zhang details that using the aforementioned combined probability values (P(t|c)) based on the individual probability values (unary potentials) allows the system to model dependencies between column types, providing a more nuanced representation of the sets of feature type candidates, allowing for a more informed selection.
Therefore, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the process disclosed by Zhang to select the different sets of feature type candidates disclosed by Rouesnel to provide a more nuanced representation of the sets of feature type candidates, allowing for a more informed selection.
Regarding claim 9-10, 12
Claims 9-10, 12 are non-transitory computer-readable media claims that are substantially similar to method claims 2-3, 5, respectively, and are rejected using the same reasoning.
Regarding claim 16, Zhang teaches;
The system of claim 15, wherein: the feature type candidates of the plurality of feature type candidates are identified based on an inference type machine learning analysis of the dataset;
These limitations are substantially similar to method claim 2, and are rejected using the same reasoning.
Using the same reasoning from claim 2, it would have been obvious to one of ordinary skill in the art, before the effective filing date of the claimed invention, to use the process disclosed by Zhang to identify feature type candidates to be used in the processes disclosed by Rouesnel, to automate the process of identifying feature type candidates.
and the operations further comprise filtering the feature type candidates based on respective probability values corresponding to likelihoods that the feature type candidates are actual feature types corresponding to their respective data subsets.
These limitations are substantially similar to method claim 3, and are rejected using the same reasoning.
Using the same reasoning from claim 3, a person of ordinary skill in the art, before the effective filing date of the claimed invention, would have been motivated to incorporate the type inference and unary-potential scoring of Zhang into the column analysis process of Rouesnel to provide more probabilistically relevant candidate feature types, thereby enabling Rouesnel’s interpretation engines to generate more appropriate feature engineering pipelines for machine learning.
Regarding claim 18
Claim 18 is a non-transitory computer-readable media claim that is substantially similar to method claim 5, and is rejected using the same reasoning.
CONCLUSION
Any inquiry concerning this communication or earlier communications from the examiner should be directed to Matthew Alan Cady whose telephone number is (571) 272-7229. The examiner can normally be reached Monday - Friday, 7:30 am - 5:00 pm ET.
Examiner interviews are available via telephone, in-person, and video conferencing using a USPTO supplied web-based collaboration tool. To schedule an interview, applicant is encouraged to use the USPTO Automated Interview Request (AIR) at http://www.uspto.gov/interviewpractice.
If attempts to reach the examiner by telephone are unsuccessful, the examiner’s supervisor, Cesar Paula can be reached on (571)272-4128. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of published or unpublished applications may be obtained from Patent Center. Unpublished application information in Patent Center is available to registered users. To file and manage patent submissions in Patent Center, visit: https://patentcenter.uspto.gov. Visit https://www.uspto.gov/patents/apply/patent-center for more information about Patent Center and https://www.uspto.gov/patents/docx for information about filing in DOCX format. For additional questions, contact the Electronic Business Center (EBC)
at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/MATTHEW ALAN CADY/ Examiner, Art Unit 2145
/CESAR B PAULA/Supervisory Patent Examiner, Art Unit 2145
Prosecution Insights