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 .
Continued Examination Under 37 CFR 1.114
A request for continued examination under 37 CFR 1.114, including the fee set forth in 37 CFR 1.17(e), was filed in this application after final rejection. Since this application is eligible for continued examination under 37 CFR 1.114, and the fee set forth in 37 CFR 1.17(e) has been timely paid, the finality of the previous Office action has been withdrawn pursuant to 37 CFR 1.114. Applicant's submission filed on 12/22/2025 has been entered.
Response to Arguments
Applicant’s arguments in light of the amendment filed on 12/22/2025 with respect to the 35 U.S.C. 112(b) rejections have been fully considered and are persuasive. The 35 U.S.C. 112(b) rejections of claims 1-24 have been withdrawn.
Claim Rejections - 35 USC § 102
The following is a quotation of the appropriate paragraphs of 35 U.S.C. 102 that form the basis for the rejections under this section made in this Office action:
A person shall be entitled to a patent unless –
(a)(1) the claimed invention was patented, described in a printed publication, or in public use, on sale, or otherwise available to the public before the effective filing date of the claimed invention.
Claims 1, 2, 7, 9, 10, 15, 17, 18, 23 and 25 are rejected under 35 U.S.C. 102(a)(1) as being anticipated by Feature Selection Based on Mutual Information: Criteria of Max-Dependency, Max-Relevance, and Min-Redundancy to Peng et al. (hereinafter Peng).
Per claim 1, Peng discloses One or more machine-readable storage media comprising instructions (Abstract…implementing two-state feature selection algorithm, “Because of the difficulty in directly implementing the maximal dependency condition, we first derive an equivalent form, called minimal-redundancy-maximal-relevance criterion (mRMR), for first-order incremental feature selection. Then, we present a two-stage feature selection algorithm by combining mRMR and other more sophisticated feature selectors (e.g., wrappers). This allows us to select a compact set of superior features at very low cost. We perform extensive experimental comparison of our algorithm and other methods using three different classifiers (naive Bayes, support vector machine, and linear discriminate analysis) and four different data sets (handwritten digits, arrhythmia, NCI cancer cell lines, and lymphoma tissues)”) to cause at least one processor (Section 5.2.1…the two-stage feature selection algorithm implemented in Matlab software and executed on eight 3.06G Xeon CPUs, “…compared the average computational time cost to select the top 50 mRMR and MaxDep features for both continuous data sets NCI and LYM, based on parallel experiments on a cluster of eight 3.06G Xeon CPUs running Redhat Linux 9, with the Matlab implementation”) to at least:
obtain dataset features indicative of a plurality of characteristics of an input dataset (Section 1… obtaining features from multiple datasets, each dataset having a plurality of characteristics, "Given the input data D tabled as N samples and M features…", Section 5.1 and Table 1…features from four datasets are obtained, "HDR" (649 features for handwritten digits), "ARR" (278 features for arrhythmia), "NCI" (9703 gene features for cancer cell lines), and "LYM" (4,026 gene features for lymphoma));
execute multiple ranking algorithms to rank the dataset features based at least in part on a performance metric (Section 2.2 and Section 3.2…a two-stage feature selection algorithm with multiple ranking approaches including, 1) mRMR incremental selection per equation (7) ranks features based on relevance/redundancy metrics (2) Wrappers (both forward and backward selection) rank features with the direct goal to minimize the classification error, “By using mRMR feature selection in the first-stage, we intend to find a small set of candidate features, in which the wrappers can be applied at a much lower cost in the second-stage….In this paper, we consider two selection schemes of wrapper, i.e., the backward and forward selections:…”; Section 3.2…the minimization of the classification error associated with the wrapper is the performance metric to meet, “A wrapper [15], [18] is a feature selector that convolves with a classifier (e.g., naive Bayes classifier), with the direct goal to minimize the classification error of the particular classifier.”; Section 5…additionally, Max-Relevance and Max-Dependency are discussed as alternative ranking methods where performance metrics include mutual information values and classification error rates), at least one of the multiple ranking algorithms to correlate the dataset features based on mutual information shared between the dataset features (Section 2.2…the Min-Redundancy criterion uses mutual information between features per equation (5), where I(xi, xj) is mutual information between features xi and xj, "The following minimal redundancy (Min-Redundancy) condition can be added to select mutually exclusive features…"; Section 2.2…per equation (7), the mRMR incremental algorithm correlates the candidate feature xj with already-selected features xi based on their shared mutual information; Section 1.. equation (1) defines mutual information, “Given two random variables x and y, their mutual information is defined in terms of their probabilistic density functions…”.
identify feature subsets for respective ones of the ranked dataset features (Section 3.1… identifying and creating sequential feature subsets for respective ranked positions, where subset Sk contains the top k ranked features, S1 for the 1st ranked feature, S2 for the top 2 ranked features, etc., “Use mRMR incremental selection (7) to select n (a preset large number) sequential features from the input X. This leads to n sequential feature sets S1… Sn”; Section 3.3…Multiple algorithms used to select n sequential feature sets, "We use both mRMR and Max-Relevance methods to select n sequential feature sets S1…Sn");
prior to training a machine learning model, predict performance metrics based on the feature subsets (Section 3.1…cross-validation errors serve as predicted performance metrics evaluated in the first stage, before final feature subset selection and final model training in the second stage, “Compare all the n sequential feature sets S1… Sn (1≤k≤n) to find the range of k, called Ω, within which the respective (cross-validation classification) error ek is consistently small (i.e., has both small mean and small variance)”; Section 6…”Our method uses an optimal first-order incremental selection to generate a candidate list of features that cover a wider spectrum of characteristic features. These candidate features have similar generalization strength on different classifiers (as seen in Figs. 3 and 4 and Tables 2 and 3). They facilitate effective computation of wrappers to find compact feature subsets with superior classification accuracy (as shown in Fig. 5 and Table 4)”); and
select a final subset of the dataset features for training the machine learning model based on the predicted performance metrics (Section 3.1…selects a final subset based on the performance metrics: "Within Ω, find the smallest classification error e = min ek*. The optimal size of the candidate feature set, n, is chosen as the smallest k that corresponds to e**"; Table 4…shows "Lowest Error Rate (%)" with selected feature subsets across datasets and classifiers; Section 3.2…the final compact subset is selected based on the predicted/evaluated performance metrics, "Once the termination condition is satisfied, the selected number of features, m, is chosen as the dimension for which the lowest error is first reached").
Per claim 2, Peng discloses claim 1, further disclosing determine whether a budget for feature extraction is reached (Section 3.2…termination condition from which feature selection is stopped and m features are chosen, is construed as a budget for feature extraction, “Once the termination condition is satisfied, the selected number of features, m, is chosen as the dimension for which the lowest error is first reached. For example, suppose the sequence of classification errors of the first six features is [10, 8, 4, 4, 4, 7]. The forward selection will terminate at five features, but only return the first three features as the result; in this way we obtain a more compact set of features that minimizes the error.”).
Per claim 7, Peng discloses claim 1, further disclosing the instructions, when executed, cause one or more of the at least one processor to identify a feature subset for respective ones of the ranked dataset features including at least a threshold amount of ranked dataset features (Section 3.1… Peng teaches creating sequential feature subsets, "S1 ⊂ S2 ⊂ ... ⊂ Sn " where each Sk contains k features and algorithm identifies the range Ω and determines the optimal size n*: "The optimal size of the candidate feature set, n*, is chosen as the smallest k that corresponds to e*", the value n* serves as a threshold for the minimum number of features; Section 3.1…additionally, "a preset large number" n establishes an upper threshold, these thresholds ensure feature subsets contain at least a threshold amount of ranked features.
Per claim 25, Peng discloses claim 1, further disclosing the instructions, when executed, cause one or more of the at least one processor to generate a correlation data structure representative of the mutual information shared between the dataset features (Section 2.2, equation (5)…The mRMR algorithm requires computing the mutual information I(xi, xj) between feature pairs for the Min-Redundancy criterion from equation (5) intrinsically requiring generation and maintaining a correlation data structure (e.g., matrix, table, or computed values) that represents the mutual information relationships between features; Section 2.2, equation (7)…the incremental algorithm from equation (7) requires: "(1/(m-1)) Σ I(xj, xi)" which necessitates storing or computing pairwise mutual information values), the at least one of the multiple ranking algorithms based on the correlation data structure (Section 2.2, equation (7)…The mRMR ranking algorithm from equation (7) expressly uses the mutual information between features I(xi, xj) in its ranking criterion where the ranking decisions are based on the computed mutual information values between features, necessitating the correlation data structure.
Claims 9, 10 and 15 are substantially similar in scope and spirit to claims 1, 2 and 7, respectively. Therefore, the rejections of claims 1, 2 and 7 are applied accordingly.
Claims 17, 18 and 23 are substantially similar in scope and spirit to claims 1, 2 and 7, respectively. Therefore, the rejections of claims 1, 2 and 7 are applied accordingly.
Allowable Subject Matter
Claims 3-6, 8, 11-14, 16, and 19-22 are objected to as being dependent upon a rejected base claim, but would be allowable if rewritten in independent form including all of the limitations of the base claim and any intervening claims.
The following is the statement of reasons for the indication of allowable subject matter: The prior art disclosed by the applicant and cited by the Examiner fail to teach or suggest, alone or in combination, all the limitations of the independent and intervening claims (claims 1, 2, 9, 10, 17 and 18), further including the particular notable limitations provided below:
Claims 3, 11 and 19: the budget is a time budget
Claims 4, 12 and 20: generate, in response to a determination that the budget for feature extraction is not reached, a second performance metric based on the one or more performance metrics.
Claims 5, 13 and 21: generate the performance metrics based on a cost metric indicative of an implementation cost of a corresponding feature.
Claims 6, 14 and 22: evaluate the feature subsets using a proxy model to predict the performance metrics.
Claims 8 and 16: the performance metrics are predicted by a plurality of proxy models.
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ALAN CHEN whose telephone number is (571)272-4143. The examiner can normally be reached M-F 10-7.
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, Kamran Afshar can be reached at (571) 272-7796. 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.
/ALAN CHEN/Primary Examiner, Art Unit 2125