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 .
Priority
Claims priority CON of 18/567,875 12/07/2023 18/567,875 is a 371 of PCT/JP2023/032508 09/06/2023 are acknowledge.
Information Disclosure Statement
The information disclosure statements (IDS) submitted on 01/07/26, 06/24/26, and 12/04/2025 are in accordance with the provisions of 37 CFR 1.97 and are considered by the Examiner.
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.
Subject Matter eligibility Rejection 35 U.S.C 101
Claims 1-18 are rejected under 35 U.S.C. § 101 for the reasons below.
Step 1: Statutory Category
Step 1 addresses whether the claims recite statutory categories.
Claims 1-6 recite an information-processing device and thus fall within the machine category.
Claims 7-12 recite an information-processing method and thus fall within the process category.
Claims 13-18 recite a non-transitory computer-readable storage medium storing a program and thus fall within the manufacture category. Each claim satisfies Step 1. MPEP § 2106.03.
Claims recite statutory categories; therefore, below are analyses in the context of prong one.
Step 2A, Prong One: Judicial Exception
Prong one addresses whether the claims recite an abstract idea.
Claim 1.
An information processing device comprising:
at least one memory configured to store instructions;
and at least one processor configured to execute the instructions to:
acquire a first model associated with a first elapsed period and a second model associated with a second, later elapsed period, wherein each model has been trained by machine learning to output a measure for a human in response to input of a plurality of types of feature values representing a condition of the human;
determine, for the first model, a first order of priority associated with a first set of the types of feature values;
determine, for the second model, a second order of priority associated with a second set of the types of feature values;
(f) generate, for the second model, aggregated evaluation data including pairs (X', y') by, for each of a plurality of varied subsets of the second set of the types of feature values:
(g) determining a first output obtained when all types of feature values in the second set are input to the second model;
(h) and determining a second output by providing, as input to the second model, actual values for feature values included in the subset and a predetermined reference value for feature values not included in the subset, wherein X' identifies the subset and y' indicates whether the second output is identical to the first output;
train, for the second model, a binary determination model using the aggregated evaluation data;
determine, for the second model, based on outputs of the binary determination model, a second required number of the types of feature values for which data acquisition can be omitted without changing an output of the second model;
reset the first order of priority by inserting, at a position corresponding to the second required number, one or more types of feature values extracted from the second order of priority, into the first order of priority;
and output acquisition-instruction data according to the reset first order of priority to a user terminal to support correction of a data acquisition plan for the first elapsed period.
Note: Non-bold language above recites an abstract idea, while bold language introduces elements that are further evaluated under prong two and step 2B.
Claim 1 is treated as representative for the independent-claim Prong One analysis because independent claims 7 and 13 recite substantively the same abstract idea and differ principally in statutory categories.
Under their broadest reasonable interpretation, the non-bold portions of limitations c–h and j–k recite evaluating feature values representing a human condition to determine a corresponding measure, ranking types of that information, selecting and evaluating subsets, comparing outputs obtained from complete and reduced information, determining from those comparisons the amount of information necessary to preserve the result, and revising an informational priority order accordingly. These acts constitute observations, evaluations, and judgments that can be performed by a human using written information and therefore fall within the mental-process grouping of MPEP 2106.04(a)(2)(III).
For example, a person provided with the relevant information could obtain two decision guides corresponding to earlier and later periods and use condition information to determine a measure for the individual. The person could rank the types of condition information by priority for each period, write different subsets of the later-period information, determine the result when considering all information and the result when considering only each subset while assigning a reference value to omitted information, record the subset as X′ and record whether the two results are the same as y′, examine the resulting binary determinations to determine how many types of information may be omitted without changing the result, and insert selected highly ranked later-period information into the corresponding position of the earlier-period priority list. These acts directly mirror the claimed determine, generate, determining whether identical, determine required number, and reset order of priority operations.
Dependent Claims 2-6 mirror Claims 8-12 and 14-18, and recite the inherent abstract idea of Claim 1; Claims 7 and 13 are explained above.
Claim 2 further recites determining a required number from evaluation-data pairs or binary outputs. A person given the same results can inspect which subsets preserve the output and determine the required number, an evaluation and judgment.
Claim 3 further conditions insertion on whether selected feature types are absent from a specified range of the first priority order. A person can inspect the ranked list, determine whether an item appears within the specified range, and insert it when absent.
Claim 4 further specifies that the binary model determines whether a subset-based output is identical to the all-feature output. The substantive determination is the same mental comparison already recited in Claim 1: observe two results and judge same or different.
Claim 5 further specifies determining the priority orders based on feature weights. Given the weights, a person can compare their relative values and rank the corresponding feature types; thus the limitation further specifies the parent evaluation and prioritization rather than replacing it with a non-mental operation.
Claim 6 further presents the reset priority order to support decision-making by a user regarding correction of the acquisition plan.
Claims 2–6, and their mirrored Claims 8–12 and 14–18, retain Claim 1’s mental process and, where noted above, further specify its observation, comparison, evaluation, judgment, and prioritization operations. MPEP 2106.04(a)(2)(III) places such practically human-performable evaluations and judgments within the mental-process grouping.
Accordingly, Claims 1–18 recite the same underlying mental process of evaluating, comparing, prioritizing, and using information to guide a decision. Because the claims therefore recite an abstract idea under Step 2A, Prong One, the analysis proceeds to Prong Two to determine whether the additional elements integrate that abstract idea into a practical application.
Step 2A, Prong Two: Integration into a Practical Application
Prong Two evaluates the additional elements identified above, individually and as an ordered combination, to determine whether they integrate the recited mental process into a practical application.
Independent Claims 1, 7, and 13
The additional elements are the memory and processor of Claim 1, the non-transitory computer-readable storage medium and computer of Claim 13, the requirement that the treatment models have been trained by machine learning, training a binary determination model using the aggregated evaluation data, and outputting acquisition-instruction data according to the reset first order of priority to a user terminal. Claim 7 recites substantially the same ML model and terminal-output implementation as Claim 1, without Claim 1's device structure.
The memory, processor, computer, and storage medium provide the machinery on which the mental evaluation is executed but do not change how those components operate. Under MPEP § 2106.05(f), merely instructing a computer to perform the judicial exception does not integrate it into a practical application; no claimed improvement to processor, memory, storage, or computer operation is recited.
The requirements that the treatment models have been trained by machine learning and that the processor train ... a binary determination model using the aggregated evaluation data likewise do not provide practical application, because they provide the computerized mechanism for performing the evaluations identified in Prong One, rather than an improvement to machine-learning technology. The claim recites no training algorithm, model architecture, loss function, parameter-update technique, or other technological mechanism by which the training itself is improved.
The additional element output acquisition-instruction data according to the reset first order of priority to a user terminal occurs after the priority information has been determined. The specification confirms that the information is presented so that the user can check the priority feature value types, and make decisions regarding correction. Spec. [0030]. Thus, the terminal communicates the result of the preceding analysis; it does not itself correct the data-acquisition plan or change a physical process. MPEP § 2106.05(g) identifies outputting a report after information analysis as insignificant post-solution activity, and Example 47 similarly treats outputting anomaly data as an additional output activity that does not integrate the abstract analysis.
Viewed as an ordered combination, the additional elements use generic computing and machine-learning tools to execute the mental process and communicate its resulting priority information. They do not recite an improved computer or ML technique, a particular machine integral to the exception, a transformation of an article, a particular treatment, or another meaningful limitation beyond implementing and reporting the abstract evaluation. Accordingly, Claims 1, 7, and 13 do not integrate the judicial exception into a practical application.
Dependent Claims
Claims 2, 4, and 5, mirrored by Claims 8, 10–11 and 14, 16–17, add use of aggregated evaluation data, binary determination models, or machine-learning weights to perform the evaluations identified in Prong One. These limitations specify computerized tools or information used to determine required feature numbers, compare model outputs, or establish priority orders, but do not recite an improvement to those tools or another technological process. They therefore amount to instructions to apply the abstract evaluation using computer or machine-learning technology rather than integrating it into a practical application. MPEP § 2106.05(f).
Claim 3 (and mirrored Claims 9 and 15) merely requires checking if selected feature types are absent from a priority range and inserting them if not present. This is part of the abstract idea, not a separate practical application.
Claim 6, mirrored by Claims 12 and 18, additionally requires presenting the reset priority order on a user-terminal display. The evaluation and decision-making have already produced the priority information; the display merely communicates that result so a user can consider correction of the acquisition plan. This is insignificant post-solution activity under MPEP § 2106.05(g), rather than a technological application of the mental process.
Accordingly, the additional elements do not change the character of the claimed mental process. The processor, memory, storage medium, and machine-learning models merely execute the evaluations identified in Prong One; the dependent claims either further specify those evaluations or use the same computer/ML tools to perform them; and the terminal/display only communicates the resulting priority information after the evaluation is complete. MPEP §§ 2106.05(f), 2106.05(g). Thus, individually and as an ordered combination, the additional elements do not integrate the judicial exception into a practical application. Step 2A, Prong Two is NO for Claims 1–18; the analysis proceeds to Step 2B.
Step 2B: Inventive Concept
Step 2B evaluates whether the additional elements, individually or as an ordered combination, amount to significantly more than the recited mental process. Unlike Prong Two, Step 2B may consider whether an additional element is well-understood, routine, and conventional, and any such finding must be factually supported. MPEP § 2106.05(d); 2024 AI Guidance.
Independent Claims 1, 7, and 13
The processor, memory, computer, and non-transitory storage medium do not provide an inventive concept. The specification itself describes the hardware as a typical information processing device containing a CPU, ROM, RAM, storage, communication interface, input/output interface, and bus, with the CPU loading and executing stored programs. Spec. [0033]-[0035]. This record supports that these elements perform their ordinary processing, storage, and input/output functions rather than a claimed unconventional hardware operation. MPEP § 2106.05(d).
The requirements that models have been trained by machine learning and that the system train ... a binary determination model using the aggregated evaluation data likewise do not supply significantly more. The claims instruct the ML model to implement the abstract evaluation but do not recite how the model is technologically improved or how its training departs from generic model training. These limitations remain mere instructions to use an ML model to perform the abstract evaluation under MPEP § 2106.05(f). Example 47 applies the same reasoning to the use of a trained neural network.
The user-terminal output also does not provide an inventive concept. It reports the priority information after completion of the evaluation, and the specification states that the user then checks that information and decides whether correction is appropriate. Spec. [0030]. Example 47 treats high-level outputting of resulting data as well-understood computer activity and insignificant extra-solution activity when reconsidered under Step 2B.
When considered together, these elements produce the same arrangement identified in Prong Two: ordinary computer components execute the evaluative instructions, ML models automate portions of those evaluations, and a terminal reports the resulting priority information. The combination does not change the operation of the processor, memory, model-training technology, or terminal and does not recite an unconventional technological arrangement comparable to the types recognized by MPEP § 2106.05. It therefore does not provide an inventive concept.
Dependent Claims
Claims 2, 4, and 5, mirrored by Claims 8, 10–11 and 14, 16–17, do not provide an inventive concept. Their use of aggregated data, binary models, or machine-learning weights merely supplies computerized implementation for determining, comparing, and prioritizing information already identified as the abstract process. Because these limitations do not recite a technological improvement or another meaningful application beyond instructing the computer or model to perform that evaluation, the § 2106.05(f) conclusion from Prong Two carries forward at Step 2B.
Claim 3, mirrored by Claims 9 and 15, recites no additional element beyond the abstract idea.
Claim 6, mirrored by Claims 12 and 18, adds presentation of the completed priority result on a user-terminal display. Reconsidered at Step 2B, the display performs its ordinary function of presenting information after the analysis; it does not alter the evaluation, improve display technology, or cause a further technological operation. Thus, the output remains insignificant post-solution activity under MPEP § 2106.05(g) and does not provide an inventive concept; see example 47 claim2.
Accordingly, the additional elements, individually and as an ordered combination, do not amount to significantly more than the mental process. Step 2B is NO. Claims 1-18 therefore do not satisfy the Alice/Mayo eligibility analysis and are rejected under 35 U.S.C. § 101.
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:
Determining the scope and contents of the prior art.
Ascertaining the differences between the prior art and the claims at issue.
Resolving the level of ordinary skill in the pertinent art.
Considering objective evidence present in the application indicating obviousness or nonobviousness.
Claim(s) 1-4 and 6, and 7-10, 12-16, and 18 are rejected under 35 U.S.C. 103 as being unpatentable over US20210374562A1-Chang and further in view of NPL-Ahn, et al. refer to PTO-892-U and further in view of Degeest refer to PTO-892-V.
Claim 1.
An information processing device comprising:
at least one memory configured to store instructions; (Chang, par. 0094-0096)
and at least one processor configured to execute the instructions to:(Chang, par. 0094-0096)
acquire a first model associated with a first elapsed period and a second model associated with a second, later elapsed period, wherein each model has been trained by machine learning to output a measure for a human in response to input of a plurality of types of ; (Chang evaluates rankings output by a baseline version of the machine learning model over a period of time (e.g., a number of days or weeks) [0016-0018], retains different baseline versions that have been updated over time [0053-0057], and periodically (e.g., every week, two weeks, month, etc.) retrains a new baseline version using the latest feature values [0067]; profile and/or activity data of the candidates are inputted into the machine learning model(s) [0033])
Chang states that the initial baseline version is in use in production (the first model), while a later retrained baseline or simplified version (the second model) is also kept. The first model has importance scores based on rankings that have been compiled over time. The second model is periodically retrained using the most recent feature values. Both versions analyse the human-associated profile, activity, and sensor-derived feature values in order to produce candidate-strength or interest scores together with the corresponding rankings [0033], these rankings being a measure of a human.
determine, for the first model, a first order of priority associated with a first set of the types of feature values;(Chang, Analysis apparatus 204 calculates importance scores 230 … for a baseline version 208 of the machine learning model deployed in production.[0041]. See also, [0042], [0051], [0077])
determine, for the second model, a second order of priority associated with a second set of the types of feature values;(Chang periodically retrains a new baseline version using the latest feature values and/or latest importance scores [0067], while retaining distinct updated baseline versions over time [0053], [0057]. For each baseline version, Chang calculates feature-importance scores reflecting each feature’s impact on that version’s rankings [0041], [0077], and identifies the features with the highest or lowest scores [0051], [0077]. Thus, the most reasonable reading of Chang’s recurring process is that an earlier baseline version has a first feature-priority order and a later retrained baseline version has a second feature-priority order associated with its later feature set. Figures 2–4 confirm the recurring sequence from model rankings, through feature-impact scoring, to priority-based feature selection)
generate, for the second model, aggregated evaluation data including pairs (X', y') by, for each of a plurality of varied subsets of the second set of the types of feature values: determining a first output obtained when all types of feature values in the second set are input to the second model; and determining a second output by providing, as input to the second model, actual values for feature values included in the subset and a predetermined reference value for feature values not included in the subset, wherein X' identifies the subset and y' indicates whether the second output is identical to the first output;
Chang describes the functional logic as follows:
X′, with regard to a varied subset: Chang repeatedly creates altered sets in which each altered set of feature values… includes a different subset of modified feature values. This is stated by Chang in [0044-0045, 0091-0092]. Each of these altered sets shows which feature types have been replaced and which have stayed at their original, actual values, when interpreted in a reasonable way in relation to X′.
y′, indicating whether the two outputs are the same: Chang derives an original ranking from the original feature values [0089], gets a modified ranking after having input the modified set [0091], and then calculates the rank-biased overlap between the two rankings. According to Chang [0047]–[0050], it is expressly stated that an overlap value of 1 means that the two rankings are identical; the comparison value thus shows whether the second output is the same as the first, even though Chang keeps the more detailed numeric similarity value rather than assigning a Boolean label.
The evaluation data is compiled by Chang in the form of importance scores, which are derived from the rank-biased overlaps between each feature's modified rankings and the respective original rankings. Chang [0089-0093]. This aggregation process adequately includes the data associations between each modified-feature subset and its corresponding output-comparison result. See also Chang [0047]–[0050], [0067], [0093]
;(Chang, The value of the rank-biased overlap falls in the range [0, 1], where 0 indicates that S and T are disjoint and 1 indicates that S and T are identical. [0048]. Simplification apparatus 202 uses importance scores 230 and a feature removal threshold 232 to identify a set of high-importance features 238. [0051]. High-importance features 238 include a subset of features … with importance scores 230 that meet or exceed feature removal threshold 232. [0051]. Training apparatus 210 trains a simplified version 214 … using only high-importance features 238 … [and] the corresponding labels from feature repository 234. [0054].After rank-biased overlaps 228 are calculated between a set of original rankings 250. [0050-0051])
Chang works out a binary-capable comparison metric, applies a binary feature-selection rule, and makes a binary performance/deployment decision; he also trains a separate simplified ranking model.
determine, for the second model, ;(Chang, rank-biased overlap … 1 indicates that S and T are identical [0048]; the rank-biased overlap is either 0 or 1 … [0049]; feature removal threshold [is] an estimated number of features to be removed [0053]; the reduced number of features with the lowest importance scores is excluded during retraining [0065-0067]; excluded features are removed from workflows/pipelines for training and executing simplified version 214 [0053-0055]. the number of features to be removed from the baseline version is calculated as the difference between the resource overhead of the baseline version and the target resource overhead divided by the per-feature resource overhead, rounded up to the nearest whole number. ([0076].)
Chang compares the original ranking with the altered one, and the rank-biased-overlap result can be used to indicate identity; it also establishes a numerical threshold for removal and removes the relevant features from the workflow of the simplified model. With regard to the previously adopted iterative mapping, the same procedure can be carried out for a later baseline version which makes up the second model.
; (Chang, high-importance features may be a predetermined number with the largest importance scores ([0051]), and Chang adjusts the threshold so “five fewer features are removed” ([0064]) and “a greater number of high-importance features” is included in, and used to retrain, the simplified model ([0068]).)
Chang's threshold, which is based on scores, establishes a priority line between features that are retained and those that are excluded. If the threshold is changed, features that had previously been excluded then become important retained features; Chang thus provides a reasonable method for controlling the number of feature types that are reincluded.
and output.(Chang’s model-generated rankings are “outputted to users of an online system,” and Chang’s computer system includes a display. (Chang, [0042]; [0094]-[0096].)
Chang shows users their rankings, the recommendations, the search results, the content feeds, and their position within the member rankings.
Obvious Rationales:
Chang states that the profile and/or activity data of the candidates are fed into the machine learning model(s) and that the models then produce scores indicating the strength of the candidates or the level of interest. Chang [0033]. However, it does not describe that the inputs are associated with human condition.
Degeest describes medical sensors and medical tests associated with human condition. (Degeest, sensors p. 1; medical tests are acquired p. 4; mammography tests, 8 features resulting from diabetes p.6;)
A POSITA could replace the feature values for Chang derived from his profile or activity with medical-test or sensor-derived feature values. Since both types of values function as inputs to a machine-learning model in order to eliminating redundant, or less useful, features.
Chang already produces a variety of feature subsets, compares the rankings that result from them with the ranking obtained from all the features, clearly identifies cases where the RBO equals 1, and decides on or eliminates a number of features that are of low importance, but it does not train a binary model based on those observations or deduce the number of features to be removed from such a model.
Ahn describes a separate data-sufficiency indicator which functions as a classifier and which is trained using partial-input observations linked to binary sufficiency labels, being applied in this way until it produces an output of 1, at which point further data collection can be stopped. [This indicator is likewise a classifier and is trained using a bearing fault dataset, the same dataset that is used to train the classifier f. Ahn, figure 4, p.1, p.5–p.7]
A POSITA would have applied Ahn’s learned sufficiency technique to Chang’s existing subset comparisons by using Chang’s subset as the training input and the RBO=1 identity as the binary target, and would then gradually assess Chang’s feature subsets until the trained indicator is able to predict output preservation. In this way, the POSITA would apply Ahn's learned sufficiency indicator to Chang's existing reduced-feature evaluations in order to predict when the retained feature subset preserves the full-input result, thus advancing Chang's stated aim of reducing unnecessary feature retrieval and processing while still retaining the model output. See Chang 0015, 0021, and Ahn p. 5 to p. 8.
Chang identifies high-importance features as the predetermined number having the largest importance scores ([0051]) and changes the feature-removal threshold so that fewer features are removed and more high-importance features are used to retrain the simplified model ([0064], [0068]). Chang therefore supplies a score-ordered feature-selection recommendation and a number-controlled retained-feature boundary, but does not disclose taking a feature from a later priority order and inserting it into the earlier order at that boundary. Degeest identifies the reason to make that modification: replacing an earlier ranking with one derived from limited later data may be unreliable and discard available information (p. 1). Degeest instead preserves the earlier reliable ranking RAR_A and uses RBR_B to insert new DBD_B features into RAR_A (pp. 5–6).
A POSITA would combine Degeest’s rank-merging method with Chang’s score order because Degeest expressly identifies that, when later data introduce new features, performing feature selection only from the later, smaller dataset can be unreliable and is not optimal because it discards most of the available information (p. 1). Chang likewise uses a feature-importance order to decide which input features remain in its simplified model ([0051], [0064], [0068]). Thus, when updating Chang’s feature selection to account for later-ranked feature types, a POSITA would use Degeest’s known insertion method rather than replace Chang’s existing score order with a later-only order. That modification preserves the information already reflected in Chang’s earlier importance ranking while incorporating the later feature types, which is the precise reliability and information-preservation benefit Degeest teaches.
Claim 2.
The information processing device according to claim 1, wherein the at least one processor is configured to execute the instructions to: determine a first required number of the types of feature values for the first model based on aggregated evaluation data including pairs (X', y') or based on outputs of a binary determination model trained using the aggregated evaluation data. (Chang generates different modified-feature subsets X′X′, where each altered set … includes a different subset of modified feature values ([0044]–[0045], [0091]–[0092]), and associates each subset with an RBO result y′y′, where 1 indicates S and T are identical ([0048]). Chang aggregates those evaluations into importance scores and applies a feature-removal threshold to determine the number of features retained or removed ([0051], [0053], [0076]). Applied to Chang’s production baseline version, this reads on the first alternative.)
Claim 3.
The information processing device according to claim 2, wherein inserting the one or more types of feature values extracted from the second order of priority is performed when those types are not included within a range of the first order of priority corresponding to the first required number.
Chang teaches a first priority order and a first-N range because it selects a predetermined number of features with the largest importance scores as high-importance features, Chang para. 0051. In the claim 2 combination, the first required number defines that first-N range. Chang does not teach performing the insertion only when the feature type from the second order is not included in that range.
According to Degeest, the initial ranking R_A ranks the original features in D_A, while the later data D_B introduces new features. Degeest makes use of R_B only for inserting the new features from D_B into the ranking R_A and determines the positions of these features by comparing them with the features that are already ranked in R_A, Degeest pp. 5–6. A feature from D_B that has been identified as new is completely missing from R_A and is therefore necessarily missing from the first-N part of R_A.
A POSITA would have combined Chang’s priority order with Degeest’s condition for inserting new features since Degeest states that it is unreliable to replace the original ranking with one based solely on later data and that this approach throws away most of the available information, Degeest p. 1. It would therefore keep Chang’s existing first-N feature selection and apply only Degeest’s insertion procedure to a feature type that is later in the ranking and not already included in that selected range. Reinserting a feature that is already in the range would add no new feature type and would result in duplicating information that had already been retained; by inserting only a later feature that is missing, the POSITA preserves Chang’s original ranking while at the same time incorporating the new information identified by Degeest.
Claim 4.
The information processing device according to claim 1, wherein the second required number is determined using the binary determination model, which is trained to determine whether an output of the second model, when a subset of the types of feature values is input, is identical to an output of the second model when all types of feature values are input.
Chang describes the process of generating different feature subsets, deriving a ranking from all the actual feature values and a ranking from each of the modified subsets, and concludes that the rankings are the same when the rank-biased overlap is 1, as stated in Chang paragraphs 0044 to 0050 and 0089 to 0093. However, Chang does not teach the use of a trained binary determination model to establish that the rankings are identical and to determine the second required number from the model's output.
Ahn teaches the missing learned determination. Ahn’s data-sufficiency indicator is also a classifier and is trained using partial-input observations, and its positive result represents that the partial-input and full-input classifier decisions … are the same, Ahn p. 6. Ahn then identifies the minimum partial input that produces that result, Ahn pp. 5–7.
A POSITA combined Chang with Ahn since Chang already produces the necessary training examples: each subset of Chang's features specifies the partial input, and Chang's rank-biased-overlap value of 1 provides the positive label indicating that the subset and the full-feature outputs are identical. As Ahn shows on pages 2–3, it teaches the use of these partial-input/equality observations to train a classifier which can determine the minimum sufficient input so that further collection need not take place; therefore, applying Ahn's indicator to the records of Chang's subsets would predict the smallest number of features that preserves Chang's full-feature ranking, rather than simply making an individual post hoc overlap comparison for each tested subset.
Claim 6.
The information processing device according to claim 1, wherein the at least one processor is configured to execute the instructions to: cause the reset first order of priority to be presented on a display of a user terminal to support decision-making by a user regarding correction of a data acquisition plan for the first elapsed period.(Refer claim 1 last limitation obviousness rational)
Note: Claims 7-10, 12, 13-16 and 18 are rejected with the same analysis above for be very similar than claims 1-4 and 6.
Claim(s) 5, and 11 and 17 are rejected under 35 U.S.C. 103 as being unpatentable over US20210374562A1-Chang and further in view of and further in view of NPL-Ahn, et al. refer to PTO-892-U and further in view of Degeest refer to PTO-892-V and in further view of Weston: US 20050216426
Claim 5.
The information processing device according to claim 1, wherein the first order of priority and the second order of priority are determined based on weights assigned to respective types of feature values during machine-learning of the first model and the second model.
Chang's approach involves choosing a set of a predetermined number of important features that have the highest importance scores ([0051]), altering the threshold in such a way that five fewer features are eliminated ([0064]), and including a larger number of highly important features when retraining the simplified model ([0068]). As a result, Chang provides the first and second feature priority orders specific to the model as set out in claim 1, although he does not disclose the method of determining those orders from the weights assigned during the machine learning of the respective models. Weston, on the other hand, states that the weights which multiply the classifier's inputs can be used as feature ranking coefficients and that the inputs with the largest weights have the greatest effect on the classification decision ([0090]). Moreover, Weston trains the classifier, calculates its weight vector, uses the magnitude of the weights as the criterion for ranking, and produces a feature-ranked list ([0094], [0102]–[0122]).
A POSITA could have combined Weston’s method of using learned-weight rankings with Chang’s feature selection and retraining procedure since Chang simplifies his model by keeping only those features that have a high importance and meet a removal threshold ([0051], [0064], [0068]). Weston also deals with the same feature selection issue: his recursive feature elimination method automatically eliminates gene redundancy and produces better and more compact gene subsets ([0074]). Weston gives a reason for the advantage of his additional criterion, pointing out that the inputs with the largest classifier weights have the greatest effect on the classification decision and therefore represent the most informative features ([0090]). It follows that using Weston’s learned feature weights as the importance scores for Chang would cause Chang’s threshold to remove the feature types that have less influence on the particular model decision, while retaining those that have greater influence for the simplified model.
Note: Claims 11 and 17 are rejected with claim 5 for being very similar.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JOSHUA DAMIAN RUIZ whose telephone number is (571)272-0409. The examiner can normally be reached 0800-1800.
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, Shahid Merchant can be reached at (571) 270-1360. 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.
/J.D.R./Examiner, Art Unit 3684
/Shahid Merchant/Supervisory Patent Examiner, Art Unit 3684
Prosecution Insights