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 .
Status of Claims
This action is responsive to the application filed on 07/15/2025. Claims 1-9 and 11-21 are pending and have been examined. This action is Final.
Response to Arguments
Argument 1: The applicant argues that the 101 rejection is improper because the claims are not directed to an abstract idea but instead to a specific technical solution that improves the functioning of a computer system by enabling natural language explanations for black-box machine learning algorithms. According to applicant, the invention addresses the technical problem of the lack of transparency in complex ML models and provides a new capability for the system to explain its own operations, which is analogous to Enfish where claims were found eligible because they improved computer functionality. Even if the claims were considered directed to an abstract idea, applicant contends they include an inventive concept: the non-generic combination of a regression model and a tree-based algorithm, each trained on the inputs and outputs of the black-box algorithm, to generate a natural language explanation of outcomes. Applicant analogizes this to USPTO Example 39 (improving an ANN) and Example 47 (anomaly detection), asserting that the claims provide a concrete technical improvement in the field of explainable AI and therefore should be found patent-eligible.
Examiner Response to Argument 1: The examiner has considered the argument set forth above regarding subject matter eligibility under 35 U.S.C. 101. However, the rejection is maintained because the claims remain directed to abstract ideas. Specifically, the limitations reciting determining regression coefficients, forming decision trees, determining decision paths, and generating explanations amount to mental processes that can be performed in the human mind or with pen and paper, and reciting that such steps are performed by a generic control circuit does not change their character as abstract ideas (see MPEP 2106.04(a)). Applicant’s reliance on Enfish, Example 39, and Example 47 is unpersuasive because those cases and examples involve improvements to the operation of a computer or learning model itself, whereas the instant claims use conventional regression and tree-based algorithms in their ordinary capacity to produce explanatory text, which does not improve computer functionality. Further, the additional elements such as storing and transmitting data, applying regression or decision tree models, and outputting results to a user interface are well-understood, routine, and conventional activities that do not integrate the judicial exception into a practical application or provide significantly more (see MPEP 2106.05(d)(II), 2106.05(f), 2106.05(g), and 2106.05(h)). Likewise, the dependent claims reciting regression coefficient details (claims 2-4), sorted lists (claim 5), accessing and populating text tables (claims 6-7), specific algorithms (claim 8), field-of-use limitations such as job applicant data (claim 9), or simulated data generated by an explanation system (claim 21) all merely add mathematical concepts, mental processes, insignificant extra-solution activity, or field-of-use restrictions, which are insufficient to confer eligibility. Therefore, while the arguments have been carefully considered, the claims as a whole remain directed to abstract ideas without significantly more, and the rejection under 101 is proper.
Argument 2: The applicant argues that the 103 rejections are improper because the cited references, individually or in combination, fail to teach or suggest the claimed system that generates natural language explanations using both regression analysis and a tree-based algorithm trained on the inputs and outputs of a machine learning algorithm. Applicant notes that Saarela is merely a comparison study of linear and non-linear classifiers and does not disclose using separate regression and tree-based models trained on black-box input and output to generate explanations. While Saarela suggests using multiple techniques for reliable results, it does not describe synthesizing regression coefficients and decision tree attributes into natural language explanations. Hetherington describes decision tree rule sets but not regression analysis, and the other references (Zhang, Yanagawa, Mathew) also do not teach the required combination. Accordingly, applicant contends that the amended independent claims, as well as the dependent claims, recite a specific and unconventional integration of techniques that is not taught or suggested by the cited art and are therefore patentable over the references.
Examiner Response to Argument 2: The examiner has considered the argument set forth above, but respectfully maintains that the 103 rejections are proper. First, applicant’s stance that the art does not teach using both regression and a tree-based algorithm trained on the inputs and outputs of a machine learning algorithm to generate natural-language explanations is not persuasive in view of the mapping done in this correspondence. Saarela expressly discloses training and using L1-regularized logistic regression (regression coefficients with sign and magnitude) and decision-tree/ensemble methods (e.g., random forests) on datasets consisting of feature vectors (inputs) and class labels (outputs), and further applies local, model-agnostic explanation techniques to present human-understandable explanations of classifications. The claims do not require a single “explanation model,” nor do they require a novel training standard. They merely recite determining regression coefficients, forming a decision tree from the same input/output pairs, identifying a path and “relevant attributes,” and generating an explanation that identifies at least one relevant attribute and its effect, each of which is taught or reasonably suggested by Saarela. To the extent applicant argues that “synthesizing” coefficients and tree attributes into natural language is missing, Saarela’s use of model-agnostic, local explanations to convey which features drove the outcome is exactly the type of textual, feature-effect explanation the claims recite; the claims do not require any particular linguistic template beyond identifying attributes and their effects. Hetherington supplies the remaining path-based detail by teaching traversal of a learned tree and building rules from a subset of nodes along the traversal path, which directly answers the amended “decision path” or “subset of nodes” limitations. A person of ordinary skill in the art (PHOSITA) can combine Saarela’s established regression/forest-based explanation workflow with Hetherington’s explicit path/rule extraction, as it is a predictable use of known techniques to improve traceability of instance-level explanations, providing an articulated reason to combine (e.g., to efficiently identify the specific attributes on the path that were relevant to the select item’s categorization). For the dependent claims, the applicant’s argument that the secondary references (Zhang, Yanagawa, Mathew) do not teach the “required combination” overreads the claims: claim 5’s “pros/cons” listing and claim 6’s selection from an explanation text set are directly taught by Zhang’s generation and presentation of a set of plain-language statements (pros/cons) covering ranked factors; claim 7’s “value field” population is taught by Yanagawa’s outputs (classification label/confidence or numeric value) that fill in the statement template. Claim 8’s algorithmic specificity is satisfied by tree-construction/classification teachings (CART-like heuristics) in Hetherington. Claims 9/19’s domain limitation (job applicants/rankings) is taught by Mathew’s HR ranking/prediction. Finally, claim 21’s “simulated data from an explanation system” is taught by Yanagawa’s generated input data used to train explainable models (including decision trees and linear models). Applicant does not identify any teaching away, incompatibility, or change in principle of operation that would defeat these combinations. Accordingly, the record shows that each limitation is taught or rendered obvious by the cited combinations, and the amendments do not overcome the articulated reasons to combine under KSR. Thus, the 103 rejections are therefore proper and are maintained.
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 1-9 and 11-21 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1, (similar to claims 11 and 20),
Step 1: Claim 1 is directed to a “system”, which falls within the statutory category of a machine. Thus, it passes Step 1.
Step 2A Prong 1:
“determine a regression coefficient for each of the plurality of attributes of the select data item based on performing regression analysis…determine a decision tree based on the input data and the output data…determine a decision path of the select data item in the decision tree…determine relevant attributes of the select data item, the relevant attributes correspond to a subset of the plurality of nodes on the decision path of the select data item” -- This limitation recites determining a regression coefficient for a group of attributes once regression is performed on data, a decision tree based on collected/gathered input and output data, determining a decision path for selected data within the decision tree, and determining relevant attributes of a selected data item for which will correspond to a subset of nodes on the decision path within the data tree. The limitation is directed to a process that can be performed in the human mind using evaluation, observation, and judgement, and thus the limitation is directed to a mental process.
“generate, with the machine learning algorithm, a categorization of a select data item having a plurality of attributes” -- The limitation is directed to generating categorization of select data items that have multiple attributes using an ML algorithm. The limitation is directed to a process that can be performed in the human mind, with aid of pen and paper regarding, and thus the limitation is directed to a mental process.
“generate natural language explanation of the categorization of the select data item based on the relevant attributes and regression coefficients associated with each of the relevant attributes, wherein the natural language explanation identifies at least one relevant attribute and an effect of the at least one relevant attribute of the data item on the categorization;” – This limitation is directed to generating an explanation (which amounts to a judgment step/mental process) for data using regression coefficients, and therefore this limitation is considered to be a mental process.
“to determine positive or negative impact of attributes on the categorization;” --The limitation is directed to determining a positive/negative impact of the attributes on the categorization of the data. The limitation is directed to a process that can be performed in the human mind using evaluation, observation, and judgement, and thus the limitation is directed to a mental process.
Step 2A Prong 2 and Step 2B:
“A system for providing natural language explanation to black-box algorithm generated outcome comprising: an input data database storing input data comprising a plurality of data items, each data item comprises a plurality of attributes; an output data database storing output data determined by a machine learning algorithm, the output data comprising categorizations of the plurality of data items determined by the machine learning algorithm based on attributes of the plurality of data items;” – This limitation recites storing data to a database which is considered an insignificant, extra solution activity that cannot be integrated to a practical application (see MPEP 2106.05(g)). Furthermore, under Step 2B, the act of storing data as a well-understood, routine, and conventional activity, therefore the limitation cannot provide significantly more than the judicial exception (see MPEP 2106.05(d)(II)).
“a control circuit coupled to the input data database and the output data database, the control circuit to…using a tree-based algorithm trained on the input data and the output data of the machine learning algorithm to form the decision tree -- The limitation recites a control unit that is coupled to a database for the input/output data that will generate categorization of selected data using the machine learning algorithm, and will use a trained tree-based algorithm to form a decision tree. The limitations recited above amount to no more than mere instructions to apply onto a computer, and do not integrate to a practical application, nor provides significantly more than the judicial exception (see MPEP 2106.05(f)).
“transmit to a user interface device for display, the categorization of the select data item along with the natural language explanation of the categorization of the select data.” – This limitation is directed to transmitting gathered data over a network, and it is an insignificant, extra-solution activity that cannot be integrated to a practical application (see MPEP 2106.05(g)). Furthermore, the act of transmitting data over a network, which is a well-understood, routine and conventional activity and cannot provide significantly more than the judicial exception (see MPEP 2106.05(d)(II)).
Therefore, claim 1 is non-patent eligible. Claims 11 and 20 are analogous to claim 1, and would face the same rejection set forth above.
Regarding claim 2: (analogous to claim 12)
Step 1: Claim 2 is dependent on claim 1, which is directed to a “system”. Therefore, claim 2 also falls within the statutory category of a machine and passes Step 1.
Step 2A Prong 1:
Claim 2 recites “The system of claim 1, wherein the effect of the at least one relevant attribute of the data item on the categorization is determined based on the regression coefficient of the at least one relevant attribute and comparing a value of the at least one relevant attribute of the data item with a comparison value of the at least one relevant attribute in the input data database.”. Claim 2 is directed to determining effects of regression coefficients on at least one relevant attribute, the comparing the values of an attribute from a data item and the input data database, which can be directed to a mathematical process as regression coefficient is used, and the process of determining the effects of regression coefficient can be done by a human mind and/or if given pen and paper. Therefore, under Step 2A Prong 1, the claim recites abstract ideas. The claim recites no further elements sufficient to integrate to practical application and does not provide significantly than the judicial exception, therefore under Step 2A Prong 2 and Step 2B, claim 2 is rejected under 101.
Therefore, claim 2 is non-patent eligible. Claim 12 is analogous to claim 2, and would face the same rejection set forth above.
Regarding claim 3: (analogous to claim 13)
Step 1: Claim 3 is dependent on claim 1, which is directed to a “system”. Therefore, claim 3 also falls within the statutory category of a machine and passes Step 1.
Step 2A Prong 1:
Claim 3 recites “The system of claim 1, wherein the regression coefficient comprises a sign indicating whether the attribute has a positive or negative effect on the categorization of the data item and a numerical value indicating a significance of the attribute.”. Claim 3 is directed to regression coefficients with signs and numerical values indicating effects and significance, which is considered to be mathematical relationships and calculations. Therefore, under Step 2A Prong 1, the claim recites abstract ideas. The claim recites no further elements sufficient to integrate to practical application and does not provide significantly than the judicial exception, therefore under Step 2A Prong 2 and Step 2B, claim 3 is rejected under 101.
Therefore, claim 3 is non-patent eligible. Claim 13 is analogous to claim 3, and would face the same rejection set forth above.
Regarding claim 4: (analogous to claim 14)
Step 1: Claim 4 is dependent on claim 1, which is directed to a “system”. Therefore, claim 4 also falls within the statutory category of a machine and passes Step 1.
Step 2A Prong 1:
Claim 4 recites “The system of claim 1, wherein regression coefficients of numerical attributes are determined based on linear regression and regression coefficients of categorical attributes are determined based on logistic regression.” Claim 4 is directed to determining regression coefficients using linear regression and logistic regression, which is considered to be mathematical relationships and calculations. Therefore, under Step 2A Prong 1, the claim recites abstract ideas. The claim recites no further elements sufficient to integrate to practical application and does not provide significantly than the judicial exception, therefore under Step 2A Prong 2 and Step 2B, claim 4 is rejected under 101.
Therefore, claim 4 is non-patent eligible. Claim 14 is analogous to claim 4, and would face the same rejection set forth above.
Regarding claim 5: (analogous to claim 15)
Step 1: Claim 5 is dependent on claim 1, which is directed to a “system”. Therefore, claim 4 also falls within the statutory category of a machine and passes Step 1.
Step 2A Prong 1:
Claim 5 recites “The system of claim 1, wherein the natural language explanation comprises a list of positive attributes and a list of negative attributes sorted based on whether the attributes affected the categorization of the select data positively or negatively.” Claim 5 is directed to creating sorted lists of positive and negative attributes based on how they affect categorization, which fall under a mental process as it involves judgement and evaluation that can be performed in the human mind and/or pen and paper. Therefore, under Step 2A Prong 1, the claim recites abstract ideas. The claim recites no further elements sufficient to integrate to practical application and does not provide significantly than the judicial exception, therefore under Step 2A Prong 2 and Step 2B, claim 5 is rejected under 101.
Therefore, claim 5 is non-patent eligible. Claim 15 is analogous to claim 5, and would face the same rejection set forth above.
Regarding claim 6: (analogous to claim 16)
Step 1: Claim 6 is dependent on claim 1, which is directed to a “system”. Therefore, claim 4 also falls within the statutory category of a machine and passes Step 1.
Step 2A Prong 1:
“and select from the plurality of text strings associated with the at least one relevant attribute based on how the attribute affects the categorization of the select data.” – This limitation is directed to a process of evaluation and selection, which could be performed in the human mind, and it is considered a mental process.
Step 2A Prong 2 and Step 2B:
“The system of claim 1, wherein the control circuit is further configured to: access an explanation text table comprising text strings associated with each of the plurality of attributes, wherein one or more attributes are associated with a plurality of text strings”--This limitation recites a control circuit that is configured to accessing a table of text strings. This is considered to be mere data gathering, and is an insignificant, extra-solution activity, which cannot be integrated to a practical application (see MPEP 22106.05(g)). Furthermore, under Step 2B,
Therefore, claim 6 is non-patent eligible. Claim 16 is analogous to claim 6, and would face the same rejection set forth above.
Regarding claim 7: (analogous to claim 17)
Step 1: Claim 7 is dependent on claim 6, which is directed to a “system”. Therefore, claim 4 also falls within the statutory category of a machine and passes Step 1.
Step 2A Prong 1:
Claim 7 recites “The system of claim 6, wherein one or more text strings in the explanation text table comprises a value field, and the control circuit is configured to populate the value field of a selected text string based on a value of the at least one relevant attribute of the data item”. Claim 7 is directed to a process of populating a field with a value, which could be performed mentally or with pen and paper, therefore this is considered to be a mental process. Therefore, under Step 2A Prong 1, the claim recites abstract ideas. The claim recites no further elements sufficient to integrate to practical application and does not provide significantly than the judicial exception, therefore under Step 2A Prong 2 and Step 2B, claim7 is rejected under 101.
Therefore, claim 7 is non-patent eligible. Claim 17 is analogous to claim 7, and would face the same rejection set forth above.
Regarding claim 8: (analogous to claim 18)
Step 1: Claim 8 is dependent on claim 1, which is directed to a “system”. Therefore, claim 8 also falls within the statutory category of a machine and passes Step 1.
Step 2A Prong 1:
Claim 8 recites “The system of claim 1, wherein the decision tree is determined based on a classification and regression tree (CART) algorithm or a Chi-square automatic interaction detection (CHAID) algorithm.” Claim 8 is directed to determining a decision tree based on types of algorithms, which is considered to be a mathematical concept. Therefore, under Step 2A Prong 1, the claim recites abstract ideas. The claim recites no further elements sufficient to integrate to practical application and does not provide significantly than the judicial exception, therefore under Step 2A Prong 2 and Step 2B, claim 8 is rejected under 101.
Therefore, claim 8 is non-patent eligible. Claim 18 is analogous to claim 8, and would face the same rejection set forth above.
Regarding claim 9: (analogous to claim 19)
Step 1: Claim 9 is dependent on claim 1, which is directed to a “system”. Therefore, claim 8 also falls within the statutory category of a machine and passes Step 1.
Claim 9 recites “The system of claim 1, wherein the input data comprises job applicant data and the output data comprises rankings of applicants.” Claim 9 recites no elements under Step 2A Prong 1. Under Step 2A Prong 2, claim 9 recites further limitations of claim 1, where the claim recites a system that comprises input data of job applicant data, as well as output data that has rankings of the applicants of the job applicants. This limitation cannot be integrated to a practical application because it is merely limiting the abstract idea to a particular field of use or technological environment (MPEP 2106.05(h)). In addition, under step 2B, the claim cannot provide significantly more than the judicial exception.
Therefore, claim 9 is non-patent eligible.
Regarding claim 21,
Step 1: Claim 9 is dependent on claim 1, which is directed to a “system”. Therefore, claim 8 also falls within the statutory category of a machine and passes Step 1.
There are no elements to be evaluated under Step 2A Prong 1.
Step 2A Prong 2 and Step 2B:
“The system of claim 1, wherein input data used to train the tree-based algorithm and the regression model includes simulated data from an explanation system that generates the natural language explanation.” -- The limitation recites that the input data used to train the tree-based algorithm will further be implemented to include simulated data from a system that will generate an NL explanation. The limitation merely limits the input data to a field of use/environments that cannot be integrated to a practical application, nor provide significantly more than the judicial exception (see MPEP 2106.05(h)).
Therefore, claim 21 is non-patent eligible.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this
Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not
identically disclosed as set forth in section 102, if the differences between the claimed invention and the
prior art are such that the claimed invention as a whole would have been obvious before the effective filing
date of the claimed invention to a person having ordinary skill in the art to which the claimed invention
pertains. Patentability shall not be negated by the manner in which the invention was made.
The factual inquiries for establishing a background for determining obviousness under 35 U.S.C. 103 are
summarized as follows:
1. Determining the scope and contents of the prior art.
2. Ascertaining the differences between the prior art and the claims at issue.
3. Resolving the level of ordinary skill in the pertinent art.
4. Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claims 1-4,8,11-14,18, and 20 are rejected under 35 U.S.C. 103 as being unpatentable over NPL Saarela et. al “Comparison of feature importance measures as explanations for classification models” (referred herein as Saarela) in view of US patent Hetherington et. al, US11531915B2 (referred herein as Hetherington).
Regarding claim 1, Saarela teaches:
A system for providing natural language explanation to blackbox algorithm generated outcome comprising: an input data database storing input data comprising a plurality of data items, each data item comprises a plurality of attributes an output data database storing output data determined by a machine learning algorithm, the output data comprising categorizations of the plurality of data items determined by the machine learning algorithm based on attributes of the plurality of data items… ([Saarela, page 3, Sec. 3] “We used two different data sets with binary classification tasks. The first data set is the openly available breast cancer data from the UCI Archive. This set includes benign and malignant cell samples from 569 patients… Each sample contains thirty features…The classes in the breast cancer data are linearly separable, making the classification a simple task.”, wherein the examiner interprets samples containing thirty features to be the same as “data items comprising a plurality of attributes” because they are both directed to multiple measurable characteristics of each record, and “benign and malignant classifications” to be the same as “categorizations of the plurality of data items” because they are both directed to output labels generated by an algorithm using those attributes. Furthermore, the examiner interprets “thirty features for each cell sample” to be the same as “a plurality of attributes of a select data item” because they are both directed to multiple measurable characteristics of a single record, and “benign and malignant classifications” to be the same as “a categorization of the select data item” because they are both directed to output labels generated by a machine learning algorithm based on those attributes).).
a control circuit coupled to the input data database and the output data database the control circuit to: generate, with the machine learning algorithm, a categorization of a select data item having a plurality of attributes; determine a regression coefficient for each of the plurality of attributes of the select data item based on performing regression analysis with a regression model the regression model being trained on the input data and the output data of the machine learning algorithm to determine positive or negative impact of attributes on the categorization; ([Saarela, page 1] “Classification models have two main objectives [9]. First, they should perform well, meaning they should forecast the output for new given input features as accurately as possible.” AND [Saarela, page 2] “An example of modular global feature importances are the coefficients in L1 regularized logistic regression. L1 regularized logistic regression assigns coefficients based on the importance of a feature, forcing coefficients of unimportant features to exactly zero and providing a magnitude and direction for the remaining coefficients that directly allow an interpretation of the corresponding features…we focus on feature importance or saliency techniques, that is, techniques that explain the decision of an algorithm by assigning values that reflect the importance of input components in their contribution to that decision”, wherein the examiner interprets “L1 regularized logistic regression assigns coefficients based on the importance of a feature” to be the same as a determined regression coefficient for an attribute(s)” because they are both directed to numerical weights trained on input/output data indicating positive or negative influence of attributes on classification. Furthermore, the examiner interprets “meaning they should forecast the output for new given input features as accurately as possible…techniques that explain the decision of an algorithm by assigning values that reflect the importance of input components in their contribution to that decision” to be the same as generating a categorization of data using an ML initiated algorithm.)
a decision tree based on the input data and the output data using a tree-based algorithm trained on the input data and the output data of the machine learning algorithm to form the decision tree, the decision tree comprises a plurality of nodes each associated with an attribute of the plurality of attributes; ([Saarela, page 4, Sec. 4.2] “Random forest is a nonlinear classification and regression method that is based on building an ensemble of decision trees…Decision trees are tree-like models, where data is split recursively at each decision node into subsets using some rule. The leaf nodes represent the outcome for the observation.”, wherein the examiner interprets “decision nodes splitting data into subsets using rules” to be the same as “nodes each associated with an attribute” because they are both directed to tree structures that partition data based on attribute conditions).
generate natural language explanation of the categorization of the select data item based on the relevant attributes and regression coefficients associated with each of the relevant attributes, wherein the natural language explanation identifies at least one relevant attribute and an effect of the at least one relevant attribute of the data item on the categorization; and transmit to a user interface device for display, the categorization of the select data item along with the natural language explanation of the categorization of the select data. ([Saarela, Abstract] “In this study we compare different feature importance measures using both linear
(logistic regression with L1 penalization) and non-linear (random forest) methods and local interpretable model agnostic explanations on top of them.”, wherein the examiner interprets “local interpretable model-agnostic
explanations” to be the same as “natural language explanation identifying relevant attributes and their effects”
because they are both directed to presenting human-understandable textual descriptions of which features affected
classification results).
Saarela does not teach determine a decision path of the select data item in the decision tree; determine relevant attributes of the select data item, the relevant attributes correspond to a subset of the plurality of nodes decision path of the select data item;.
Hetherington teaches determine a decision path of the select data item in the decision tree; determine relevant attributes of the select data item, the relevant attributes correspond to a subset of the plurality of nodes decision path of the select data item; ([Hetherington, col. 4, lines 30-45] “Candidate decision rules are generated by traversing the tree. Each rule is built from one or more combinations of a subset of nodes in a traversal path of the tree.”, wherein the examiner interprets “subset of nodes in a traversal path” to be the same as “subset of the plurality of nodes decision path” because they are both directed to selecting certain nodes along a path that represent relevant features for a given item).
Saarela, Hetherington, and the instant application are analogous art because they are all directed to generating explanations for outputs of machine learning systems using attributes and decision structures.
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the invention to modify the natural language explanation system disclosed by Saarela to include the “subset of nodes in a traversal path” disclosed by Hetherington. One would be motivated to do so to effectively determine a decision path of a select data item that corresponds to relevant attributes, as suggested by Hetherington (Hetherington, col. 4, lines 30–45, “Candidate decision rules are generated by traversing the tree. Each rule is built from one or more combinations of a subset of nodes in a traversal path of the tree.”). Claims 11 and 20 are analogous to claim 1, and would face the same rejection set forth above.
Regarding claim 2, Saarela in view of Hetherington teaches The system of claim 1, (see rejection of claim 1).
Saarela further teaches wherein the effect of the at least one relevant attribute of the data item on the categorization is determined based on the regression coefficient of the at least one relevant attribute and comparing a value of the at least one relevant attribute of the data item with a comparison value of the at least one relevant attribute in the input data database. ([Saarela, page 3, Sec. 4.1] “L1 regularized logistic regression works by penalizing the feature coefficients with the L1 norm, shrinking some of the feature coefficients to exactly zero... The magnitude of feature coefficients can be interpreted as the importance of that feature, a larger coefficient meaning the feature had more relevance in the classification. In addition, the direction of the coefficient tells whether the feature increases or decreases the probability of belonging to a certain class. The model was trained with the LogisticRegressionCV function and five-fold cross-validation to choose the amount of penalization to use”, wherein the examiner interprets the magnitude and direction of feature coefficients indicating importance and effect on classification to be the same as determining the effect of an attribute based on its regression coefficient and comparing its value.)
Saarela, Hetherington, and the instant application are analogous art because they are all directed to determining relevant attributes based on regression coefficients and compare a value of the relevant attribute to another in the input data database.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of claim 1 disclosed by Saarela and Hetherington to include the “magnitude of feature coefficients can be interpreted as the importance of that feature, a larger coefficient meaning the feature had more relevance in the classification. In addition, the direction of the coefficient tells whether the feature increases or decreases the probability of belonging to a certain class. The model was trained with the LogisticRegressionCV function and five-fold cross-validation to choose the amount of penalization to use” disclosed by Saarela. One would be motivated to do so to efficiently determine the attribute’s effect based on coefficients and then comparing its value as suggested by Saarela (see [Saarela, page 3, Sec. 4.1] above). Claim 12 is analogous to claim 2, and therefore will face the same rejection as set forth above.
Regarding claim 3, Saarela in view of Hetherington teaches The system of claim 1, (see rejection of claim 1).
Saarela further teaches wherein the regression coefficient comprises a sign indicating whether the attribute has a positive or negative effect on the categorization of the data item and a numerical value indicating a significance of the attribute. ([Saarela, page 3, Sec. 4.1] “The magnitude of feature coefficients can be interpreted as the importance of that feature, a larger coefficient meaning the feature had more relevance in the classification. In addition, the direction of the coefficient tells whether the feature increases or decreases the probability of belonging to a certain class. The model was trained with the LogisticRegressionCV function and five-fold cross-validation to choose the amount of penalization to use”, wherein the examiner interprets “the magnitude of feature coefficients” to be the same as “numerical value indicating the significance of the attribute” and “the direction of the coefficient tells whether the feature increases or decreases the probability of belonging to a certain class” to be the same as “the regression coefficient comprises a sign indicating whether the attribute has a positive or negative effect on the categorization of the data item”.)
Saarela, Hetherington, and the instant application are analogous art because they are all directed to categorizing (signing) an attribute to be positive or negative effective data.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of claim 1 disclosed by Saarela and Hetherington to include the “the magnitude of feature coefficients…the direction of the coefficient tells whether the feature increases or decreases the probability of belonging to a certain class” as disclosed by Saarela. One would be motivated to do so to efficiently signate if a coefficient is negative/decreasing or positive/increasing value as suggested by Saarela (see [Saarela, page 3, Sec. 4.1] quote above). Claim 13 is analogous to claim 3, and therefore will face the same rejection as set forth above.
Regarding claim 4, Saarela in view of Hetherington teaches The system of claim 1, (see rejection of claim 1).
Saarela further teaches wherein regression coefficients of numerical attributes are determined based on linear regression and regression coefficients of categorical attributes are determined based on logistic regression. ([Saarela, Abstract] “In this study we compare different feature importance measures using both linear (logistic regression with L1 penalization) and non-linear (random forest) methods” and [Saarela, page 3, Sec. 4.1] “L1 regularized logistic regression works by penalizing the feature coefficients with the L1 norm, shrinking some of the feature coefficients to exactly zero… The magnitude of feature coefficients can be interpreted as the importance of that feature, a larger coefficient meaning the feature had more relevance in the classification. In addition, the direction of the coefficient tells whether the feature increases or decreases the probability of belonging to a certain class L1 regularized logistic regression works by penalizing the feature coefficients with the L1 norm, shrinking some of the feature coefficients to exactly zero. The magnitude of feature coefficients can be interpreted as the importance of that feature, a larger coefficient meaning the feature had more relevance in the classification. In addition, the direction of the coefficient tells whether the feature increases or decreases the probability of belonging to a certain class”, wherein the examiner interprets logistic regression being used to determine feature coefficients and their importance for classification to be the same as determining regression coefficients for categorical attributes based on logistic regression.)
Saarela, Hetherington, and the instant application are analogous art because they are all directed to determining coefficients based on logistic regression.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of claim 1 disclosed by Saarela and Hetherington to include the “logistic regression works by penalizing the feature coefficients…importance of that feature, a larger coefficient meaning the feature had more relevance in the classification. In addition, the direction of the coefficient tells whether the feature increases or decreases the probability of belonging to a certain class” as disclosed by Saarela. One would be motivated to do so to effectively determine regression coefficients for attributes based on logistic regression as suggested by Saarela (see [Saarela, Abstract] and [Saarela, page 3, Sec. 4.1] quotes above). Claim 14 is analogous to claim 4, and therefore will face the same rejection as set forth above.
Regarding claim 8, Saarela in view of Hetherington teaches The system of claim 1, (see rejection of claim 1).
Hetherington further teaches wherein the decision tree is determined based on a classification and regression tree (CART) algorithm or a Chi-square automatic interaction detection (CHAID) algorithm. (Hetherington, [Col. 6, line 4] “A decision tree may operate as a classifier. Any of examples 0-7 may be more or less accurately classified by descending through the tree according to how the example does or does not satisfy conditions within the tree during descent.” and [Col. 4, line 59] “Construction of the decision tree may occur according to the techniques and heuristics discussed above and later herein or according to other techniques” wherein the examiner interprets “operate as a classifier” and “construction of the decision tree... according to techniques and heuristics” to be equivalent to determining a decision tree based on classification algorithms similar to CART and/or CHAID).
Saarela, Hetherington, and the instant application are analogous art because they are all directed to creating a decision tree and determining classifications using techniques, like an algorithm.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of claim 1 disclosed by Saarela and Hetherington to include the “A decision tree may operate as a classifier. Any of examples 0-7 may be more or less accurately classified by descending through the tree… Construction of the decision tree may occur according to the techniques and heuristics” as disclosed by Hetherington. One would be motivated to do so to efficiently create a decision tree using techniques and heuristics to classify by iterating through the tree as suggested by Hetherington (see [Hetherington, col. 6 line 4 and col.4 line 59] quote above). Claim 18 is analogous to claim 8, and therefore will face the same rejection as set forth above.
Claims 5,6,15, and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Saarela in view of Hetherington in further view of Zhang et. al, US20160070867A1 (referred herein as Zhang).
Regarding claim 5, Saarela in view of Hetherington teaches The system of claim 1, (see rejection of claim 1).
Saarela in view of Hetherington does not teach wherein the natural language explanation comprises a list of positive attributes and a list of negative contributes sorted based on whether the attributes affected the categorization of the select data positively or negatively.
Zhang teaches wherein the natural language explanation comprises a list of positive attributes and a list of negative contributes sorted based on whether the attributes affected the categorization of the select data positively or negatively. (Zhang, [0007] “The natural language translation module is configured to determine a common set of pros and cons for each of a plurality of treatment options described in the quantitative personalized decision content...present natural language explanations to the patient describing a selected set of the treatment options in terms of the common set of pros and cons in natural language”, wherein the examiner interprets determining pros and cons to be the same as listing positive and negative attributes that affected the categorization.)
Saarela, Hetherington, Zhang, and the instant application are analogous art because they are all directed to natural language explanations that has positive and negative attributes.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of claim 1 disclosed by Saarela and Hetherington to include the “The natural language translation module is configured to determine a common set of pros and cons for each of a plurality of treatment options described in the quantitative personalized decision content” as disclosed by Zhang. One would be motivated to do so to effectively determine a set of pro attribute and con attributes to decipher which may be affected by the categorization of selected data as suggested by Zhang (see Zhang, [0007] quote above). Claim 15 is analogous to claim 5, and therefore will face the same rejection as set forth above.
Regarding claim 6, Saarela in view of Hetherington teaches The system of claim 1, (see rejection of claim 1).
Saarela in view of Hetherington does not teach wherein the control circuit is further configured to: access an explanation text table comprising text strings associated with each of the plurality of attributes, wherein one or more attributes are associated with a plurality of text strings; and select from the plurality of text strings associated with the at least one relevant attribute based on how the attribute affects the categorization of the select data.
Zhang teaches wherein the control circuit is further configured to: access an explanation text table comprising text strings associated with each of the plurality of attributes, wherein one or more attributes are associated with a plurality of text strings; and select from the plurality of text strings associated with the at least one relevant attribute based on how the attribute affects the categorization of the select data. (Zhang, [0031] “a set of natural language explanations to exhaustively cover all the situations in treatment option rankings and contributing factor rankings are generated”, and [0032] “Natural language explanations from the generated set describing the selected treatment option are presented 112 to the patient.”, wherein the examiner interprets this set of natural language explanations to be equivalent to an explanation text table with text strings associated with attributes, and presenting natural language explanations describing how the selected option ranks as selecting text strings based on how attributes affect categorization.)
Saarela, Hetherington, Zhang, and the instant application are analogous art because they are all directed to using an explanation text table with strings associated to attributes and methods of selecting it.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of claim 1 disclosed by Saarela and Hetherington to include the “a set of natural language explanations to exhaustively cover all the situations in treatment option rankings…Natural language explanations from the generated set describing the selected treatment option are presented” as disclosed by Zhang. One would be motivated to do so to efficiently select strings (treatment options) in a generated set of explanations presented as suggested by Zhang (See [Zhang, 0031 and 0032] quote above). Claim 16 is analogous to claim 6, and therefore will face the same rejection as set forth above.
Claims 7 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over Saarela in view of Hetherington in further in view of Zhang further in view of Yanagawa et. al, US20220067558A1 (referred herein as Yanagawa).
Regarding claim 7, Saarela in view of Hetherington further in view of Zhang The system of claim 6, (see rejection of claim 6).
Saarela in view of Hetherington further in view of Zhang does not teach wherein one or more text strings in the explanation text table comprises a value field, and the control circuit is configured to populate the value field of a selected text string based on a value of the at least one relevant attribute of the data item.
Yanagawa teaches wherein one or more text strings in the explanation text table comprises a value field, and the control circuit is configured to populate the value field of a selected text string based on a value of the at least one relevant attribute of the data item. ([0034] “In an embodiment, if the text to be explained is a classification problem, then the output is a classification label and its confidence value. In an embodiment, if the text to be explained is a regression problem, then the output is the value of the input text itself.”, wherein the examiner interprets “classification label and its confidence value” or “value of the input text” to be equivalent to a value field, and “output” to be analogous to “populate the value field”.)
Saarela, Hetherington, Zhang, Yanagawa, and the instant application are analogous art because they are all directed to text strings of the explanation table comprising a field of values and to populate the value field.
It would have been obvious to one of ordinary skill in the art before the effective filing date of the claimed invention to modify the system of claim 6 disclosed by Saarela and Hetherington to include the “if the text to be explained is a classification problem, then the output is a classification label and its confidence value” as disclo