DETAILED ACTION
This action is responsive to the application filed on 06/12/2026. Claims 1-13 are pending and have been examined. This action is Final.
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 .
Priority
Applicant’s claim for the benefit of a prior-filed application under 35 U.S.C. 119(e) or under 35 U.S.C.
120, 121, 365(c), or 386(c) is acknowledged.
Response to Arguments
Argument 1: The applicant asserts that the amendments to claims 1-10 have resolved the indefiniteness the Examiner identified, and therefore requests withdrawal of the 112(b) rejection. In other words, the applicant treats the indefiniteness rejection as mooted by the claim amendments (which recast the “factor” as one of four enumerated categories, recast “combination” as a “logical relationship between at least two Boolean-type evaluation results,” and recast the action as one “recommended as an attempt to eliminate” the factor) rather than contesting the Examiner’s original reasoning.
Response to Argument 1: The examiner has considered the arguments set forth above. In light of the amendments, the examiner finds the arguments persuasive, thus the rejection is withdrawn.
Argument 2: The applicant argues that amended claim 1 is patent-eligible under both prongs. Under Step 2A Prong Two, the applicant contends the claim, even assuming it recites a mathematical concept or mental process, integrates that concept into a practical application because it does not merely calculate and evaluate metrics or output an identified factor. Rather, it identifies a prediction-error factor as one of a limited set (error other than the model or data, local error, distribution shift, or abnormality in a prediction error sample) according to a logical relationship between at least two Boolean-type evaluation results, and then determines and outputs proposal information indicating a concrete corresponding action (operation test of a system, hyperparameter adjustment and re-learning, re-learning with appropriate data, or investigating the cause). The applicant analogizes to the USPTO 2024 AI eligibility examples, arguing the claim is like Example 47 claim 3 (eligible because the detected anomaly is used to improve network security) and Example 48 claim 2 (eligible because AI results improve speech-separation technology), and unlike the ineligible Example 47 claim 2 and Example 48 claim 1 that merely detect or output analytical results, and points to specification paragraphs [0006] and [0046] as showing the claim connects the analysis to a concrete improvement action so that “the user can immediately start an action necessary for improvement.” In the alternative, under Step 2B, the applicant argues that even if the claim is directed to an abstract idea, the ordered combination of the four recited operations amounts to “significantly more,” because the prediction-error analysis is transformed into a proposal for a concrete improvement action rather than merely output as a classification, and thus is not a mere implementation of an abstract idea on a generic computer. The applicant states that independent claims 9 and 10 are eligible for the same reasons.
Response to Argument 2: The examiner has considered the applicant’s arguments above, but does not find them persuasive. The recited factor identification and rule-based action selection are themselves the abstract idea; the only additional elements are the generic processor and memory, which are mere instructions to apply the exception on a computer under Step 2A Prong 2/Step 2B (MPEP 2106.05(f)), and the creating and outputting of proposal information, which merely presents the result of the analysis and is insignificant extra-solution activity (MPEP 2106.05(g) under Step 2A Prong 2 and MPEP 2106.05(d)(II) under Step 2B). The applicant’s reliance on Example 47 claim 3 and Example 48 claim 2 is misplaced because those claims performed a further action that improved a technology (blocking network traffic; improving speech separation), whereas amended claim 1, by its own terms, recites only “an action to be recommended as an attempt to eliminate the identified factor” and to “create and output proposal information indicating the determined action.” The recited operation test, hyperparameter adjustment and re-learning, re-learning with appropriate data, and cause investigation are recited only as the content of the recommendation and are not performed by the claim, placing it with the ineligible Example 47 claim 2 and Example 48 claim 1. The specification passages the applicant cites confirm this, stating the output is an “action proposal” so the user “can immediately start an action,” that is, a person, and not the claim, performs any improvement. Under Step 2B, the additional elements add only a generic computer performing the abstract analysis and outputting its result, which is well-understood, routine, and conventional. Accordingly, claim 1, and claims 9 and 10 for the same reasons, remain ineligible under 35 U.S.C. 101.
Argument 3: The applicant argues that the 102-anticipation rejection must show every claim element, and a 103 rejection must show all limitations taught or suggested, and then arguing that the cited references, whether alone or in combination, fail to teach or suggest every feature of amended independent claim 1. The applicant’s central contention is that the references do not disclose “identify a factor of an error in prediction … according to a logical relationship between at least two Boolean-type evaluation results … wherein … a different factor is identified in a case where the logical relationship is different.” Specifically, the applicant argues that Fly’s disclosure of a “combined result” exceeding a threshold is merely a single numerical comparison, not a logical relationship between at least two Boolean-type evaluation results in which different logical relationships yield different identified factors, and that Fly does not disclose the amended claim’s enumerated factors or the requirement that a different factor be identified when the logical relationship differs. The applicant further argues that Fly does not disclose “determine, based on a rule that assigns different actions to different factors, an action to be recommended as an attempt to eliminate the identified factor,” including the four specific factor-to-action mappings, and that Mewald, Wexler, Walters, and Chung likewise do not teach or suggest these features. On that basis the applicant asserts that independent claim 1 is patentable, that independent claims 9-10 are patentable for at least the same reasons, that dependent claims 2-8 are patentable at least by virtue of their dependency, and that new claims 11-13 are patentable both by dependency and for reciting additional features.
Response to Argument 3: The examiner has considered the applicant’s arguments above, but does not find them persuasive. The applicant’s arguments are directed to Fly and to a theory that the claimed “logical relationship between at least two Boolean-type evaluation results” can be met only by a “combined result” exceeding a threshold. Those arguments are moot because the present grounds, necessitated by the amendments, do not rely on Fly or on any combined result or numerical comparison. As set forth in the rejection in this action, Peh discloses the recited Boolean-type evaluation results by storing each failed rule check as a “1” and each passed check as a “0”, and Qureshi supplies the contested limitation, disclosing that a logic rule is “a logical combination of conditions” whose truth is interpreted “as an indication of a problem,” with different logic rules identifying different problems (see mapping below for further details). Thus, the combination thus teaches identifying a different factor according to the logical relationship among the Boolean results. The factor-to-action determination the applicant disputes is likewise mapped above, with Qureshi teaching a remedy stored in association with each problem, Peh teaching retraining with new hyperparameters, and Walters teaching correcting the model based on detected data drift. The applicant's assertion that “Mewald, Wexler, Walters and Chung also do not teach or suggest these features” is unpersuasive and, as to Wexler, moot, because the present rejection relies on Peh in view of Qureshi, Mewald, and Walters, and further in view of Chung for claim 8, not Fly or Wexler. Because the applicant's arguments do not address the combination applied or identify any limitation it fails to teach, claims 1-13 remain rejected under 35 U.S.C. 103.
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 therefore, subject to the
conditions and requirements of this title.
Claims 1-13 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
Regarding claim 1,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Step 2A Prong 1:
(a) “calculate and evaluate a plurality of types of metrics with respect to a prediction model, data of explanatory variables used in the prediction model, or data of target variables used in the prediction model” - The limitation is directed to calculating and evaluating metrics associated with a prediction model and associated data. The limitation recites mathematical calculations, operations, and relationships, and thus is directed to math.
(b) “identify a factor of an error in prediction by the prediction model according to a logical relationship between at least two Boolean-type evaluation results of the plurality of types of the metrics, wherein the identified factor is any one of: an error other than the prediction model and data; a local error; a distribution shift in data; and an abnormality in a prediction error sample, a different factor is identified in a case where the logical relationship is different” - The limitation is directed to applying a logical relationship between Boolean-type evaluation results to identify a corresponding prediction error factor. The limitation involves evaluating information and identifying a result according to a logical relationship and thus recites a mathematical concept and a mental process involving evaluation and judgment, and thus the limitation is directed to math and/or a mental process.
(c) “determine, based on a rule that assigns different actions to different factors, an action to be recommended as an attempt to eliminate the identified factor, wherein: based on the identified factor being the error other than the prediction model and data, the action is determined as performing an operation test of a system; based on the identified factor being the local error, the action is determined as adjusting a hyperparameter for learning the prediction model and re-learning the prediction model; based on the identified factor being the distribution shift in data, the action is determined as re-learning the prediction model using appropriate data; and based on the identified factor being the abnormality in the prediction error sample, the action is determined as investigating a cause of occurrence of the prediction error sample” - The limitation is directed to applying a rule to an identified factor to select a corresponding recommended action. Such rule-based evaluation and selection constitutes evaluation and judgment and thus recites a mental process.
Step 2A Prong 2 and Step 2B:
(a) “An analysis device comprising: at least one memory storing instructions; and at least one processor configured to execute the instructions to” - The limitation recites generic computer components used to perform the judicial exception. The memory stores instructions and the processor executes the instructions and therefore amount to mere instructions to apply the judicial exception on a computer. The limitation does not integrate into a practical application, nor provide significantly more than the judicial exception (see MPEP 2106.05(f)).
(b) “create and output proposal information indicating the determined action” - The limitation creates and communicates information representing the result of the abstract determination. The limitation merely outputs the result of the preceding analysis and constitutes insignificant extra-solution activity. See MPEP 2106.05(g).
Thus, claim 1 is non-patent eligible. Claims 9 and 10 are analogous to claim 1, aside from claim type and minute differences, and thus same rejection can be applied.
Regarding claim 2,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Step 2A Prong 1:
“identify a factor of an error in prediction by the prediction model according to a rule for associating logical relationship between Boolean-type evaluation results of the plurality of types of the metrics with factors” - The limitation is directed to applying a rule associating logical relationships between Boolean-type evaluation results with corresponding factors. The limitation involves evaluating information according to a rule and identifying a corresponding result and thus recites a mental process.
Step 2A Prong 2 and Step 2B:
“The analysis device according to claim 1, wherein the processor is configured to execute the instructions to” -- The limitation recites a processor being configured to execute the instructions to perform the task highlighted in prong 1. The limitation amounts to no more than mere instructions to apply onto a computer, and it cannot be integrated to a practical application, and cannot provide significantly more than the judicial exception (see MPEP 2106.05(f)).
Thus, claim 2 is non-patent eligible.
Regarding claim 3,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Step 2A Prong 1:
“identify a factor of an error in prediction by the prediction model according to a logical relationship between a Boolean-type evaluation result of a predetermined metric among the plurality of types of the metrics and a Boolean-type evaluation result of the metric selected according to the Boolean-type evaluation result of the predetermined metric” - The limitation is directed to using one Boolean-type evaluation result to select another metric and applying a logical relationship between Boolean-type evaluation results to identify a factor. The limitation recites mathematical or logical relationships and evaluation and judgment and thus recites a mathematical concept and a mental process.
Step 2A Prong 2 and Step 2B:
“The analysis device according to claim 2, wherein the processor is configured to execute the instructions to” -- The limitation recites a processor being configured to execute the instructions to perform the task highlighted in prong 1. The limitation amounts to no more than mere instructions to apply onto a computer, and it cannot be integrated to a practical application, and cannot provide significantly more than the judicial exception (see MPEP 2106.05(f)).
Therefore, claim 3 is non-patent eligible.
Regarding claim 4,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Step 2A Prong 1:
“calculate and evaluate the metrics by the calculation algorithm or the evaluation algorithm” -- The limitation is directed to calculate and evaluate metrics by the calculation/evaluation algorithm. The limitation is directed to the use of a mathematical calculation/operation/concept, and thus the limitation is directed to math.
Step 2A Prong 2 and Step 2B:
“The analysis device according to claim 1, wherein the processor is configured to execute the instructions to: “receive an instruction to designate a calculation algorithm or an evaluation algorithm for the metrics… designated by the instruction.” -- The limitation recites a processor being configured to execute the instructions to perform the task highlighted in prong 1. Furthermore, the limitation recites receiving an instruction to designate algorithms for calculation relating to the metrics. The limitation amounts to no more than mere instructions to apply onto a computer, and it cannot be integrated to a practical application, and cannot provide significantly more than the judicial exception (see MPEP 2106.05(f)).
Therefore, claim 4 is non-patent eligible.
Regarding claim 5,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Step 2A Prong 1:
“identify a factor of an error in prediction by the prediction model according to the rule designated by the instruction.” -- The limitation is directed to identifying factor in error in prediction model according to the rule that is instructed. The limitation is directed to a process that can be performed in the human mind using observation and judgement, with aid of pen and paper, and thus the limitation is directed to a mental process.
Step 2A Prong 2 and Step 2B:
“The analysis device according to claim 2, wherein the processor is further configured to execute the instructions to: receive an instruction to designate the rule,” -- The limitation recites a processor being configured to execute the instructions to perform the task highlighted in prong 1. Furthermore, the limitation recites receiving an instruction to designate a rule that is set. The limitation amounts to no more than mere instructions to apply onto a computer, and it cannot be integrated to a practical application, and cannot provide significantly more than the judicial exception (see MPEP 2106.05(f)).
Therefore, claim 5 is non-patent eligible.
Regarding claim 6,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Regarding claim 6,
Step 2A Prong 1:
“determine an action to be recommended as an attempt to eliminate the identified factor” - The limitation is directed to determining a recommended action based on an identified factor. Such determination can be performed through evaluation and judgment and thus recites a mental process.
Step 2A Prong 2 and Step 2B:
“The analysis device according to claim 1, wherein the processor is further configured to execute the instructions to determine an action for eliminating the identified factor.” -- The limitation recites a processor being configured to execute the instructions to perform the task of determining an action for eliminating the factor that is identified (deleting). The limitation amounts to no more than mere instructions to apply onto a computer, and thus the limitation does not integrate to a practical application, nor provides significantly more than the judicial exception (see MPEP 2106.05(f)).
Therefore, claim 6 is non-patent eligible.
Regarding claim 7,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Step 2A Prong 1:
“generate image data of a predetermined graph according to the metrics.” -- The limitation is directed to generating image data of a predetermined graph according to metrics. is directed to a process that can be performed in the human mind using observation and judgement, with aid of pen and paper, and thus the limitation is directed to a mental process.
Step 2A Prong 2 and Step 2B:
“The analysis device according to claim 1, wherein the processor is configured to execute the instructions to” -- The limitation recites a processor being configured to execute the instructions to perform the task highlighted in prong 1. The limitation amounts to no more than mere instructions to apply onto a computer, and it cannot be integrated to a practical application, and cannot provide significantly more than the judicial exception (see MPEP 2106.05(f)).
Therefore, claim 7 is non-patent eligible.
Regarding claim 8,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
There are no elements to be evaluated under Step 2A Prong 1.
Step 2A Prong 2 and Step 2B:
“The analysis device according to claim 1, wherein the processor is configured to execute the instructions to” -- The limitation recites a processor being configured to execute the instructions to perform the task highlighted in prong 1. The limitation amounts to no more than mere instructions to apply onto a computer, and it cannot be integrated to a practical application, and cannot provide significantly more than the judicial exception (see MPEP 2106.05(f)).
“to generate image data representing a flowchart defining the metric used to identify the factor and an order of using the metric and a transition history in the flowchart.” -- The limitation recites to generate image data that represents a flowchart defining metrics and will be used to identify the factor and order using metric and transition history (gathered data) that is within the flowchart. Generating data based on gathered information (gathered data) is an insignificant, extra-solution activity that cannot be integrated into a practical application (see MPEP 2106.05(g)). Furthermore, under Step 2B, the act of generating data merely to represent it and define new information is a well-understood routine, and conventional activity (WURC) that cannot provide significantly more than the judicial exception (see MPEP 2106.05(d)(II)).
Therefore, claim 8 is non-patent eligible.
Regarding claim 11,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Step 2A Prong 1:
(a) “The analysis device according to claim 1, wherein the plurality of types of the metrics include at least one of: an accuracy of the prediction model based on a mean square error; an abnormality score of the prediction error sample with respect to training data, calculated using an abnormality detection method; and a magnitude of distribution shift of data calculated from an inter-distribution distance between a distribution of the training data and a distribution of operation data” - The limitation is directed to calculating or evaluating a mean square error, an abnormality score, or an inter-distribution distance used to determine a magnitude of distribution shift. The limitation recites mathematical calculations, operations, and relationships and thus recites a mathematical concept.
The limitations defining the training data and operational data identify the data upon which the mathematical analysis is performed and remain part of the recited data analysis.
There are no elements to be evaluated under Step 2A Prong 2 and Step 2B.
Thus, claim 11 is non-patent eligible.
Regarding claim 12,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
Step 2A Prong 1:
(a) “The analysis, at least one of the abnormality score or the magnitude of distribution shift of data is the metric calculated for the target variables” - The limitation further defines the mathematical metric calculation by specifying that the abnormality score or magnitude of distribution shift is calculated for target variables. The limitation remains directed to mathematical analysis of data and thus recites a mathematical concept.
There are no elements to be evaluated under Step 2A Prong 2 and Step 2B.
Thus, claim 12 is non-patent eligible.
Regarding claim 13,
Step 1: This claim is directed to a device, which falls under a category of machine. The claim satisfies Step 1.
There are no elements to be evaluated under Step 2A Prong 1.
Step 2A Prong 2 and Step 2B:
“The analysis device according to claim 1, the output proposal information comprises a proposal sentence indicating the action to be recommended.” - The limitation merely specifies the form in which the result of the abstract determination is presented. Presenting the recommendation as a proposal sentence constitutes insignificant extra-solution activity and does not integrate the judicial exception into a practical application (see MPEP 2106.05(g)). Furthermore, under Step 2B, presenting information in sentence form using conventional computer output functionality which is directed to a well understood, routine, and conventional activity (WURC) for which does not provide significantly more than the judicial exception (see MPEP 2106.05(d)(II)).
Thus, claim 13 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.
Claim(s) 1-7, 9-13 are rejected under 35 U.S.C. 103 as being unpatentable over US20200364618A1, by Peh et. al. (referred herein as Peh) in view of US8001527B1, by Qureshi et. al. (referred herein as Qureshi) further in view of NPL reference “Introducing TensorFlow Model Analysis: Scaleable, Sliced, and Full-Pass Metrics”, by Mewald et. al. (referred herein as Mewald) and further in view of US10599957B2, by Walters et. al. (referred herein as Walters).
Regarding claim 1, Peh teaches:
An analysis device comprising: at least one memory storing instructions; and at least one processor configured to execute the instructions to: ([Peh, 0099-0101] “The example computer system 900 includes a processor 902…a main memory 904…The storage unit 916 includes a machine-readable medium 922 on which is stored instructions 924 (e.g., software) embodying any one or more of the methodologies or functions described herein,” wherein the examiner interprets “processor 902” to be the same as “at least one processor configured to execute the instructions” and “a machine-readable medium 922 on which is stored instructions 924” to be the same as “at least one memory storing instructions” because they are both directed to a processor that executes instructions stored in a memory.)
calculate and evaluate a plurality of types of metrics with respect to a prediction model, data of explanatory variables used in the prediction model, or data of target variables used in the prediction model; ([Peh, 0035] “These checks can be largely grouped under the major categories of checks for: model goodness-of-fit; prediction accuracy; prediction stability; input data; business understanding; and exceptions,” AND [Peh, 0027] “the early warning system 210 ingests information from existing models, such as fitted and forecast values, model coefficients, residuals, and the like,” wherein the examiner interprets the recited plurality of check categories to be the same as “a plurality of types of metrics,” and interprets the ingested “input data” and “fitted and forecast values” to be the same as “data of explanatory variables used in the prediction model, or data of target variables used in the prediction model,” because they are both directed to evaluating multiple types of metrics computed from a prediction model and its input and output data)
identify a factor of an error in prediction by the prediction model … at least two Boolean-type evaluation results of the plurality of types of the metrics; ([Peh, 0036] “For each rule that fails, the result is stored in a binary matrix as a ‘1’ and each rule that passes is stored into the matrix as a ‘0’,” AND [Peh, 0005] “generating diagnostic reports to help data scientists assess the health of the model, identify sources of error, perform speedier root cause analysis,” wherein the examiner interprets storing each rule-check result as a binary ‘1’ or ‘0’ to be the same as “Boolean-type evaluation results of the plurality of types of the metrics,” and interprets “identify sources of error” to be the same as “identify a factor of an error in prediction by the prediction model,” because they are both directed to determining a cause of a prediction error from two-valued results of evaluated metrics)
the identified factor is any one of:…an abnormality in a prediction error sample; ([Peh, 0056] “Rule 1.4 ‘Model Residual Distribution Check’. The early warning system 210 may compare the distribution of model residuals against the distribution of historical model residuals…determines if the distribution of the residual has changed significantly…and generates a warning,” wherein the examiner interprets a flagged/anomalous model-residual condition to be the same as “an abnormality in a prediction error sample” because they are both directed to an anomalous prediction-error result flagged for diagnosis)
based on the identified factor being the local error, the action is determined as adjusting a hyperparameter for learning the prediction model and re-learning the prediction model; ([Peh, 0038, 0044] “the early warning system may…attempt corrective maintenance of the model (e.g., by retraining the model with new hyperparameters)…may perform automated model update and hyperparameter tuning,” wherein the examiner interprets “retraining the model with new hyperparameters” to be the same as “adjusting a hyperparameter for learning the prediction model and re-learning the prediction model” because they are both directed to changing a hyperparameter and retraining the model)
based on the identified factor being the abnormality in the prediction error sample, the action is determined as investigating a cause of occurrence of the prediction error sample. ([Peh, 0005] “The early warning system… generating diagnostic reports to help data scientists assess the health of the model, identify sources of error, perform speedier root cause analysis, and quickly update/fix the model, as appropriate” wherein the examiner interprets generating a diagnostic report that identifies sources of error and performs root cause analysis to be the same as “investigating a cause of occurrence of the prediction error sample” because they are both directed to investigating the cause of the flagged prediction-error condition).
Peh does not teach according to a logical relationship between at least two Boolean-type evaluation results of the plurality of types of the metrics, wherein; the identified factor is any one of: an error other than the prediction model and data; a local error [and] a distribution shift in data; a different factor is identified in a case where the logical relationship is different; determine, based on a rule that assigns different actions to different factors, an action to be recommended as an attempt to eliminate the identified factor, wherein; based on the identified factor being the error other than the prediction model and data, the action is determined as performing an operation test of a system; based on the identified factor being the distribution shift in data, the action is determined as re-learning the prediction model using appropriate data; create and output proposal information indicating the determined action.
Qureshi teaches:
identify a factor of an error in prediction by the prediction model according to a logical relationship between at least two Boolean-type evaluation results of the plurality of types of the metrics, wherein … a different factor is identified in a case where the logical relationship is different; ([Qureshi, page 40, col. 7-8, lines , 38-41, 45-46, ] “a logic rule can be thought of as a logical combination of conditions; if each condition is met, the rule itself is true,…the meta-application 20 interprets the truth of a logic rule as an indication of a problem AND [Qureshi, page 37, col. 2, lines 18-20] “the encoded knowledge preferably maps known problems to logical combinations of ‘features’…or other conditions,” wherein the examiner interprets a logic rule that is a “logical combination of conditions” whose truth “indicate[es] … a problem” to be the same as “a logical relationship between at least two Boolean-type evaluation results,” and interprets different logic rules mapping to different problems to be the same as “a different factor is identified in a case where the logical relationship is different,” because they are both directed to identifying a different cause depending on which logical combination of two-valued conditions is satisfied)
the identified factor is any one of: an error other than the prediction model and data; ([Qureshi, Abstract] The system also includes a root cause analysis module configured to identify one or more problematic objects of the application model…[and finds] root cause candidates”, wherein the examiner interprets identifying a hardware or software system component as the root-cause object to be the same as “an error other than the prediction model and data” because they are both directed to attributing the fault to a component of the surrounding system rather than to the prediction model or its data)
determine, based on a rule that assigns different actions to different factors, an action to be recommended as an attempt to eliminate the identified factor, wherein: ([Qureshi, page 42, col. 12, lines 9-15] “The knowledge base 22 preferably stores information about remedial actions, or ‘remedies’ that may be performed…Each remedy is stored in the local knowledge base 22 in association with a particular problem…remedy selector 40 determines the preferred order in which to execute the remedies,” wherein the examiner interprets storing, for each problem, an associated remedy selected by a remedy selector to be the same as “a rule that assigns different actions to different factors” and “an action to be recommended as an attempt to eliminate the identified factor” because both are directed to selecting, per identified factor, a corresponding corrective action)
based on the identified factor being the error other than the prediction model and data, the action is determined as performing an operation test of a system; ([Qureshi, page 41, col. 9, lines 14-19] The application model 24 also preferably contains information about what telemetry ‘metrics’ are relevant to an object, as well as parameterized troubleshooting procedures (‘unit tests’), or references to such procedures (which can reside elsewhere), that can be used to measure the health of an object,” AND [Qureshi, Abstract] “The root cause analysis module can be further configured to apply diagnostic unit tests on one or more objects associated with the root cause candidates”, wherein the examiner interprets performing a diagnostic unit test on the suspect system component to be the same as “performing an operation test of a system” because they are both directed to testing the operation of a system component)
create and output proposal information indicating the determined action. ([Qureshi, page 42, col. 12, lines 36-40] “In one embodiment, as problems are detected, they are reported by the meta-application 20 to associated IT personnel, together with associated remedial actions and their plans that may be executed to address the detected problem,” AND [Qureshi, page 39, col. 6, lines 28-29] “user interface 29 (illustrated as a graphical user interface or ‘GUI’),” wherein the examiner interprets reporting the detected problem together with its associated remedial action to a user to be the same as “create and output proposal information indicating the determined action” because they are both directed to outputting, to the user, information indicating the determined corrective action).
Peh and Qureshi do not teach the identified factor is any one of:…a local error; [and] a distribution shift in data;…based on the identified factor being the distribution shift in data, the action is determined as re-learning the prediction model using appropriate data.
Mewald teaches:
the identified factor is any one of:…a local error; ([Mewald, “Aggregate vs sliced metrics”] “Most model evaluation results look at aggregate metrics. A model may have an acceptable AUC over the entire eval dataset, but underperform on specific slices,” wherein the examiner interprets a model that performs acceptably in aggregate but “underperform[s] on specific slices” to be the same as “a local error” because they are both directed to a localized region of the data in which the model fails while overall performance appears acceptable).
Peh, Qureshi, and Mewald do not teach “the identified factor is any one of: … a distribution shift in data”; and “based on the identified factor being the distribution shift in data, the action is determined as re-learning the prediction model using appropriate data.”
Walters teaches:
the identified factor is any one of: … a distribution shift in data; and based on the identified factor being the distribution shift in data, the action is determined as re-learning the prediction model using appropriate data. ([Walters, page 24, col. 2, lines 49-52] “detecting data drift based on a difference in a trained model parameter from a baseline model parameter … The operations may include correcting the model based on the detected data drift,” wherein the examiner interprets “detecting data drift” to be the same as “a distribution shift in data,” and interprets “correcting the model based on the detected data drift” to be the same as “re-learning the prediction model using appropriate data,” because they are both directed to retraining the model in response to a detected change in the data distribution).
Peh, Qureshi, Mewald, Walters, and the instant application are analogous art because they are all directed to monitoring a machine-learning prediction model, evaluating a plurality of metrics, and identifying and remedying the cause of prediction errors.
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 analysis device with memory and stored instructions disclosed by Peh to include the “logic rule[s] that describe problems” disclosed by Qureshi. One would be motivated to do so to efficiently associate each logical combination of the model’s pass/fail metric results with a specific identified factor and its corresponding corrective action, as suggested by Qureshi ([Qureshi, Abstract] The system also includes a root cause analysis module configured to identify one or more problematic objects of the application model…[and finds] root cause candidates”.)
It would also have been obvious to a person of ordinary skill in the art before the effective filing date of the invention to include the “sliced” metrics that reveal a model that “underperform[s] on specific slices” disclosed by Mewald. One would be motivated to do so to effectively expose localized prediction failures that aggregate metrics would otherwise conceal, as suggested by Mewald ([Mewald, “Aggregate vs sliced metrics”] “A model may have an acceptable AUC over the entire eval dataset, but underperform on specific slices.”). It would have also been obvious to a person of ordinary skill in the art before the effective filing date of the invention to include the “detecting data drift” disclosed by Walters. One would be motivated to do so to effectively restore model accuracy when the data distribution has shifted, as suggested by Walters (Walters, [col. 2, lines 52-54] “correcting the model based on the detected data drift.”). Claims 9 and 10 are analogous to claim 1, aside from claim type and minute differences, thus the same rejection can apply as above.
Regarding claim 2, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 1 (see the rejection of claim 1).
Qureshi further teaches wherein the processor is configured to execute the instructions to identify a factor of an error in prediction by the prediction model according to a rule for associating logical relationship between Boolean-type evaluation results of the plurality of types of the metrics with factors ([Qureshi, page 37, col. 2, lines 18-20] “the encoded knowledge preferably maps known problems to logical combinations of ‘features’…or other conditions,” wherein the examiner interprets encoded knowledge that maps a logical combination of conditions to a problem to be the same as “a rule for associating logical relationship between Boolean-type evaluation results of the plurality of types of the metrics with factors” because both are directed to a stored rule associating a logical relationship of conditions with an identified factor).
Peh, Qureshi, Mewald, Walters, and the instant application are analogous art because they are all directed to associating, by rule, a logical relationship between Boolean-type evaluation results of a prediction model’s metrics with corresponding error factors.
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 apparatus claim 1 disclosed by Peh, Qureshi, Mewald, and Walters to include the encoded knowledge that “maps known problems to logical combinations of … conditions” disclosed by Qureshi. One would be motivated to do so to efficiently and systematically associate each logical combination of metric results with its corresponding error factor, as suggested by Qureshi ([Qureshi, page 37, col. 2, lines 18-20] “the encoded knowledge preferably maps known problems to logical combinations of ‘features’…or other conditions,”).
Regarding claim 3, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 2 (see the rejection of claim 2).
Qureshi further teaches wherein the processor is configured to execute the instructions to identify a factor of an error in prediction by the prediction model according to a logical relationship between a Boolean-type evaluation result of a predetermined metric among the plurality of types of the metrics and a Boolean-type evaluation result of the metric selected according to the Boolean-type evaluation result of the predetermined metric ([Qureshi, Abstract] “apply diagnostic unit tests on one or more objects associated with the root cause candidates, the diagnostic unit tests configured to narrow down a list of possible root causes of the problems.”, wherein the examiner interprets evaluating an initial condition and then, based on that result, selecting and evaluating a further condition to narrow the diagnosis to be the same as “a Boolean-type evaluation result of the metric selected according to the Boolean-type evaluation result of the predetermined metric.)
Peh, Qureshi, Mewald, Walters, and the instant application are analogous art because they are all directed to identifying an error factor by conditionally selecting and evaluating a further metric based on the Boolean-type evaluation result of a predetermined metric.
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 apparatus claim 1 disclosed by Peh, Qureshi, Mewald, and Walters to include the diagnostic tests “configured to narrow down a list of possible root causes” disclosed by Qureshi. One would be motivated to do so to efficiently and precisely localize the cause of a prediction error by selecting each subsequent check based on the result of a prior check, as suggested by Qureshi ([Qureshi, Abstract] “apply diagnostic unit tests on one or more objects associated with the root cause candidates, the diagnostic unit tests configured to narrow down a list of possible root causes of the problems.”).
Regarding claim 4, the combination of Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 1 (see the rejection of claim 1 above).
Peh further teaches wherein the processor is further configured to execute the instructions to: receive an instruction to designate a calculation algorithm or an evaluation algorithm for the metrics, and calculate and evaluate the metrics by the calculation algorithm or the evaluation algorithm designated by the instruction ([Peh, ¶ 0036] “A full list of rules may be provided to the user who can specify which rules to include in the set that is applied (e.g. by indicating rules to turn on an off. The user may also set custom parameters for the rule checks (e.g., … in configuration files)…when the rule check model 212 applies rules it takes in the configuration files and runs through the rule checks as specified by the configuration file,” wherein the examiner interprets a user specifying which rule checks and parameters to apply, and the system then applying those designated rule checks, to be the same as “receive an instruction to designate a…evaluation algorithm for the metrics, and calculate and evaluate the metrics by the…evaluation algorithm designated by the instruction”).
Regarding claim 5, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 2 (see the rejection of claim 2).
Peh further teaches wherein the processor is further configured to execute the instructions to: receive an instruction to designate the rule, and identify a factor of an error in prediction by the prediction model according to the rule designated by the instruction ([Peh, [0004, 0050] “Users can choose the rules to apply to their models…Users may specify which rules to apply as an alternative to or in addition to the default,” wherein the examiner interprets a user designating which rule to apply, and the system identifying the factor based on the user-designated rule, to be the same as “receive an instruction to designate the rule, and identify a factor of an error in prediction by the prediction model according to the rule designated by the instruction”).
Regarding claim 6, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 1 (see the rejection of claim 1 above).
Qureshi further teaches wherein the processor is further configured to execute the instructions to determine an action to be recommended as an attempt to eliminate the identified factor ([Qureshi, page 42, col. 12, lines 11-15] “Each remedy is stored in the local knowledge base 22 in association with a particular problem…remedy selector 40 determines the preferred order in which to execute the remedies,” wherein the examiner interprets selecting the remedy associated with the identified problem to be the same as “determine an action to be recommended as an attempt to eliminate the identified factor”).
Peh, Qureshi, Mewald, Walters, and the instant application are analogous art because they are all directed to determining a recommended action for eliminating an identified error factor.
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 apparatus claim 1 disclosed by Peh, Qureshi, Mewald, and Walters to include the remedy “stored in the local knowledge base 22 in association with a particular problem” disclosed by Qureshi. One would be motivated to do so to effectively recommend a corrective action directed at eliminating the identified factor, as suggested by Qureshi (Qureshi, [Qureshi, page 42, col. 12, lines 11-15] “Each remedy is stored in the local knowledge base 22 in association with a particular problem.”).
Regarding claim 7, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 1 (see the rejection of claim 1 above).
Mewald further teaches wherein the processor is further configured to execute the instructions to generate image data of a predetermined graph according to the metrics ([Mewald, “How does the TensorFlow Model Analysis work?”] “TFMA uses the graph in this SavedModel to compute (sliced) metrics and provides visualization tools to analyze those metrics,” [Mewald, “Aggregate vs sliced metrics”] “Figure 3: TFMA allows us to slice a metric by different segments of our eval dataset, enabling more fine grained analysis,” wherein the examiner interprets generating a visualization of the computed metrics to be the same as “generate image data of a predetermined graph according to the metrics”).
Peh, Qureshi, Mewald, Walters, and the instant application are analogous art because they are all directed to generating a graphical representation of a prediction model’s evaluation metrics.
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 apparatus claim 1 disclosed by Peh, Qureshi, Mewald, and Walters to include the “visualization tools to analyze those metrics” disclosed by Mewald. One would be motivated to do so to effectively enable a user to inspect and interpret the evaluation results in graphical form, as suggested by Mewald ([Mewald, “How does the TensorFlow Model Analysis work?”] “TFMA uses the graph in this SavedModel to compute (sliced) metrics and provides visualization tools to analyze those metrics,” [Mewald, “Aggregate vs sliced metrics”] “Figure 3: TFMA allows us to slice a metric by different segments of our eval dataset, enabling more fine grained analysis,”.)
Regarding claim 11, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 1 (see the rejection of claim 1).
Peh further teaches the training data is data of the explanatory variables or data of the target variables used for training the prediction model, and the operational data is data obtained at the time of operation of the prediction model, and is data including data of the explanatory variables used for prediction by the prediction model or data of actual values of the target variables ([Peh, ¶¶ 0053, 0007] the model is evaluated using data “split [into] training and test samples” with limits “calculated based on the training data,” and thereafter receives data upon a “model refresh [that] may indicate the presence of new data in the output of the model,” wherein the examiner interprets the training/test samples used for the model to be the same as “the training data” and the new data obtained upon operation/refresh to be the same as “the operational data”).
Walters further teaches wherein the plurality of types of the metrics include at least one of: an accuracy of the prediction model based on a mean square error; an abnormality score of the prediction error sample with respect to training data, calculated using an abnormality detection method; and a magnitude of distribution shift of data calculated from an inter-distribution distance between a distribution of the training data and a distribution of operation data ([Walters, col. 2, lines 49-50] “detecting data drift based on a difference in a trained model parameter from a baseline model parameter,” AND [Walters, page 37, col. 27, lines 36-40] “In some aspects, a prediction accuracy check can determine the accuracy of predictions made by a model (e.g., recurrent neural network, kernel density estimator, or the like) given a dataset.” AND [Walters, page 44, col. 41, lines 22-23] “For example, detecting data drift at step 1812 may be based on at least one of a least squares error method”, wherein the examiner interprets a computed difference between a current data distribution and a baseline distribution and prediction accuracy check made by a model, as well as detecting drift based on least squares error method to be the same as “a magnitude of distribution shift of data calculated from an inter-distribution distance between a distribution of the training data and a distribution of operation data,” it being sufficient under the recited “at least one of” that one enumerated metric is taught)
Peh, Qureshi, Mewald, Walters, and the instant application are analogous art because they are all directed to computing a magnitude of distribution shift from an inter-distribution distance between a training-data distribution and an operation-data distribution.
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 apparatus claim 1 disclosed by Peh, Qureshi, Mewald, and Walters to include the “detecting data drift based on a difference…from a baseline” disclosed by Walters. One would be motivated to do so to effectively detect a distribution shift between the training data and the operation data so that the model may be corrected, as suggested by Walters ([Walters, col. 2, lines 49-50] “detecting data drift based on a difference in a trained model parameter from a baseline model parameter,” AND [Walters, page 37, col. 27, lines 36-40] “In some aspects, a prediction accuracy check can determine the accuracy of predictions made by a model (e.g., recurrent neural network, kernel density estimator, or the like) given a dataset.” AND [Walters, page 44, col. 41, lines 22-23] “For example, detecting data drift at step 1812 may be based on at least one of a least squares error method”.)
Regarding claim 12, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 11 (see the rejection of claim 11).
Peh further wherein at least one of the abnormality score or the magnitude of distribution shift of data is the metric calculated for the target variables ([Peh, 0056] “Rule 1.4, “Model Residual Distribution Check,” belongs to the “goodness-of-fit” rule family and is recommended and is recommended for evaluating statistical models, machine learning models or forecasting models using a criticality level of “Warning” and risk score of 1. The early warning system 210 may compare the distribution of model residuals against the distribution of historical model residuals from previous refreshes using the Anderson-Darling (AD) test. The early warning system 210 determines if the distribution of the residual has changed significantly”, wherein the examiner interprets a distributional metric computed from the model residuals, i.e., from the target variables, to be the same as “at least one of the abnormality score or the magnitude of distribution shift of data … calculated for the target variables” because both are directed to a metric computed on the target-variable (actual-value) data).
Regarding claim 13, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 1 (see the rejection of claim 1).
Qureshi further teaches wherein the output proposal information comprises a proposal sentence indicating the action to be recommended. ([Qureshi, page 42, col. 12, lines 37-40] “In one embodiment, as problems are detected, they are reported by the meta-application 20 to associated IT personnel, together with associated remedial actions and their plans that may be executed to address the detected problem.”, wherein the examiner interprets a textual report indicating the associated remedial action to be the same as “a proposal sentence indicating the action to be recommended”).
Peh, Qureshi, Mewald, Walters, and the instant application are analogous art because they are all directed to outputting proposal information, in sentence form, indicating a recommended corrective action.
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 apparatus claim 1 disclosed by Peh, Qureshi, Mewald, and Walters to include the reporting of a problem “together with associated remedial actions” disclosed by Qureshi. One would be motivated to do so to effectively inform the user of the recommended corrective action in a human-readable form, as suggested by Qureshi ([Qureshi, page 42, col. 12, lines 37-40] “In one embodiment, as problems are detected, they are reported by the meta-application 20 to associated IT personnel, together with associated remedial actions and their plans that may be executed to address the detected problem.”, wherein the examiner interprets a textual report indicating the associated remedial action to be the same as “a proposal sentence indicating the action to be recommended”).
Claim(s) 8 is rejected under 35 U.S.C. 103 as being unpatentable over Peh in view of Qureshi, Mewald, and Walters, and further in view of NPL reference “Automated Data Slicing for Model Validation: A Big data-AI Integration Approach”, by Chung et. al. (referred herein as Chung).
Regarding claim 8, Peh, Qureshi, Mewald, and Walters teaches The analysis device according to claim 3 (see the rejection of claim 3 above).
Peh, Qureshi, Mewald, and Walters do not teach wherein the processor is further configured to execute the instructions to generate image data representing a flowchart defining the metric used to identify the factor and an order of using the metric and a transition history in the flowchart.
Chung teaches wherein the processor is further configured to execute the instructions to generate image data representing a flowchart defining the metric used to identify the factor and an order of using the metric and a transition history in the flowchart. ([Chung, page 4, sec. 3] “Slice Finder provides interactive visualization tools for the user to explore the recommended slices,” AND [Chung, page 5, sec. 3.1.2] “Slice Finder…traversing the slice lattice in a breadth-first manner, one level at a time…checks if it has an effect size at least T…tests for statistical significance,” wherein the examiner interprets the interactive visualization to be the same as “generate image data representing a flowchart,” and interprets the breadth-first traversal applying metric checks in a specified sequence to be the same as “defining the metric used to identify the factor and an order of using the metric and a transition history in the flowchart”).
Peh, Qureshi, Mewald, Walters, Chung, and the instant application are analogous art because they are all directed to generating a visual representation of the metrics used to identify a prediction-error factor and the order in which those metrics are applied.
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 apparatus claim 1 disclosed by Peh, Qureshi, Mewald, and Walters to include the “interactive visualization tools for the user to explore the recommended slices” disclosed by Chung. One would be motivated to do so to effectively improve interpretability and user understanding of how the evaluation metrics identify problematic prediction behavior, as suggested by Chung ([Chung, page 4, sec. 3] “Slice Finder provides interactive visualization tools for the user to explore the recommended slices,” AND [Chung, page 5, sec. 3.1.2] “Slice Finder…traversing the slice lattice in a breadth-first manner, one level at a time…checks if it has an effect size at least T…tests for statistical significance,”.)
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 DEVAN KAPOOR whose telephone number is (703)756-1434. The examiner can normally be reached Monday - Friday: 9:00AM - 5:00 PM EST (times may vary).
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, David Yi can be reached at (571) 270-7519. 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.
/DEVAN KAPOOR/Examiner, Art Unit 2126
/DAVID YI/Supervisory Patent Examiner, Art Unit 2126
Prosecution Insights