DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This final office action is in response to the amendment filed 12 March 2026.
Claims 1-16 and 19-25 are pending. Claims 1, 19, and 23 are independent claims.
Claim Rejections - 35 USC § 103
In the event the determination of the status of the application as subject to AIA 35 U.S.C. 102 and 103 (or as subject to pre-AIA 35 U.S.C. 102 and 103) is incorrect, any correction of the statutory basis (i.e., changing from AIA to pre-AIA ) for the rejection will not be considered a new ground of rejection if the prior art relied upon, and the rationale supporting the rejection, would be the same under either status.
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.
This application currently names joint inventors. In considering patentability of the claims the examiner presumes that the subject matter of the various claims was commonly owned as of the effective filing date of the claimed invention(s) absent any evidence to the contrary. Applicant is advised of the obligation under 37 CFR 1.56 to point out the inventor and effective filing dates of each claim that was not commonly owned as of the effective filing date of the later invention in order for the examiner to consider the applicability of 35 U.S.C. 102(b)(2)(C) for any potential 35 U.S.C. 102(a)(2) prior art against the later invention.
Claims 1-3, 5, 7, 10-11, 14, and 19-25 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma et al. (Long-Tailed Recognition Using Class-Balanced Experts, 19 October 2020, hereafter Sharma) and further in view of Hui et al. (CN 106548196, published 29 March 2017, hereafter Hui; machine translation provided by WIPO, paragraph numbers added by examiner).
As per independent claim 1, Sharma discloses a computer-implemented method for improved machine learning on imbalanced datasets (Abstract; Section 3: Here, long-tailed recognition is used where the training set is highly imbalanced), the method comprising:
obtaining, by a computing system comprising one or more computing devices, a training dataset with class imbalance (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot))
training, by the computing system, one or more teacher classification models with the training dataset using instance-based example selection (Figure 1; Sections 1-3: Here, an expert (teacher) is initialized to train a Baseline model on the entire dataset to transfer knowledge among the plurality of class subsets (Manyshot, Mediumshot, and Fewshot))
training, by the computing system, one or more student classification models with the training dataset using class-balance example selection (Figure 1; Sections 1 and 3: Here, the imbalanced classification problem is decomposed into a plurality of balanced classification problems by splitting the long-tailed training classes into balanced subsets. This prevents competition between the subsets (Manyshot, Mediumshot, and Fewshot) during training. Instead, training is performed on each of the subsets to generate a plurality of expert models (EManyshot, EMediumshot, EFewshot))
wherein training the one or more student models comprises training the one or more student classification models to predict data generated by the one or more teach classification models via distillation training (Figure 1; Section 3: Here, the feature extractor part of each export model E is initialized using the Baseline model pre-trained on the entire training set DTrain. This enables knowledge transfer between the plurality of subclass (student) models)
providing, by the computing system, the one or more student classification models as an output (Figure 1: Here, knowledge transfer is performed using the trained subclass (student) models)
Sharma fails to specifically disclose:
wherein instance-based example selection comprises randomly sampling the training dataset with the class imbalance
wherein class-balance example selection comprises equalizing probability sampling across different classes
However, Hui, which is analogous to the claimed invention because it is directed toward selecting data in a dataset having class imbalance, discloses:
wherein instance-based example selection comprises randomly sampling the training dataset with the class imbalance (paragraph 0002: Here, class imbalanced data is processed. This includes extracting a training data set proportional to samples of each cluster group (paragraph 0015))
wherein class-balance example selection comprises equalizing probability sampling across different classes (paragraph 0015: Here, samples are extracted at equal probability and proportional to the samples of each cluster group)
It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Hui with Sharma, with a reasonable expectation of success, as it would have allowed for obtaining a random forest training data subset having the proportion and probability corresponding to the larger data set (Hui: paragraph 0015).
As per dependent claim 2, Sharma discloses the method wherein:
each of the one or more teach classification models comprises a feature extraction portion configured to receive an input and generate a feature representation and a classification portion configured to receive the feature representation and generate a classification output (Figure 1: Here, the received data is classified as belonging or being outside the subset by an expert (teach) classification model. This data is then classified as one of the subsets (Manyshot, Mediumshot, Fewshot))
each of the one or more student classification models comprises a feature extraction portion configured to receive an input and generate a feature representation and a classification portion configured to receive the feature representation and generate a classification output (Figure 1; Section 3: Here, the subset datasets are then passed to the trainer to train the various models and generate expert models (EManyshot, EMediumshot, EFewshot))
training, by the computing system, the one or more student classification models to predict data generated by the one or more teach classification models via distillation training comprises:
training, by the computing system, the feature extraction portion of each student classification model to predict the feature representation generated by the feature extraction portions of the one or more teacher classification models (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This allows for using the training data to fit the appropriate model and avoid the long-tail resulting in biased classification)
training, by the computing system, the classification portion of each student classification model to predict the classification output generated by the classification portion of the one or more teach classification models (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This allows for using the training data to fit the appropriate model and avoid the long-tail resulting in biased classification)
As per dependent claim 3, Sharma discloses wherein the one or more teacher classification models comprise an ensemble of a plurality of teach classification models respectively generated from a plurality of different initialization parameters (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This partitioned data is used to generate a plurality of expert models (EManyshot, EMediumshot, and EFewshot). These models constitute an ensemble set of models trained using different parameters).
As per dependent claim 5, Sharma discloses wherein the one or more teacher classification models comprise an ensemble of a plurality of teacher classification models that have a plurality of different sets of hyperparameters (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This partitioned data is used to generate a plurality of expert models (EManyshot, EMediumshot, and EFewshot). These models constitute an ensemble set of models trained using different parameters).
As per dependent claim 7, Sharma discloses wherein the one or more student classification models comprise a convolutional neural network (Section 3.1).
As per dependent claim 10, Sharma discloses the method further comprising:
obtaining, by the computing system, a dataset, wherein the dataset comprises one or more features (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This partitioned data is used to generate a plurality of expert models (EManyshot, EMediumshot, and EFewshot). These models constitute an ensemble set of models trained using different parameters)
processing, by the computing system, the dataset with the one or more student classification models to generate one or more class confidence scores based on the one or more features (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This partitioned data is used to generate a plurality of expert models (EManyshot, EMediumshot, and EFewshot). These models constitute an ensemble set of models trained using different parameters)
determining, by the computer system, one or more classification predictions based at least in part on the one or more class confidence scores (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This partitioned data is used to generate a plurality of expert models (EManyshot, EMediumshot, and EFewshot). These models constitute an ensemble set of models trained using different parameters)
As per dependent claim 11, Sharma discloses the limitations similar to those in claim 10, and the same rejection is incorporated herein. Sharma discloses wherein the dataset comprises one or more images and the one or more classification predictions comprises one or more object classification or image classification (Abstract).
As per dependent claim 14, Sharma discloses the limitations similar to those in claim 10, and the same rejection is incorporated herein. Sharma discloses wherein the training dataset comprises images (Abstract).
With respect to independent claim 19, the claim recites the limitations substantially similar to those in claim 1. The analysis of claim 1 is incorporated herein by reference.
Sharma fails to specifically disclose one or more processors and one or more non-transitory computer readable media that collectively store instructions that, when executed by the one or more processors, cause the computing system to perform operations.
However, the examiner takes official notice that it was notoriously well-known in the art at the time of the applicant’s effective filing date to implement a method using one or more processors and one or more non-transitory computer readable media that collectively store instructions that, when executed by the one or more processors, cause the computing system to perform the operations of the method. It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined the well-known with Sharma, with a reasonable expectation of success, as it would have allowed for implementing the method on a system.
With respect to dependent claim 20, the claim recites the limitations substantially similar to those in claim 14. The analysis of claim 14 is incorporated herein by reference.
With respect to dependent claim 21, the claim recites the limitations substantially similar to those in claim 11. The analysis of claim 11 is incorporated herein by reference.
With respect to dependent claim 22, the claim recites the limitations substantially similar to those in claim 2. The analysis of claim 22 is incorporated herein by reference.
With respect to independent claim 23, the claim recites the limitations substantially similar to those in claim 1. The analysis of claim 1 is incorporated herein by reference.
With respect to dependent claim 24, the claim recites the limitations substantially similar to those in claims 11 and 14. The analysis of claims 11 and 14 is incorporated herein by reference.
With respect to dependent claim 25, the claim recites the limitations substantially similar to those in claim 5. The analysis of claim 5 is incorporated herein by reference.
Claims 4 and 6 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma and Hui and further in view of Xiang et al. (Learning From Multiple Experts: Self-paced Knowledge Distillation for Long-tailed Classification, 15 July 2020, hereafter Xiang).
As per dependent claim 4, Sharma and Hui disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Sharma discloses training the plurality of teacher classification models with the training dataset using instance-based example selection for the plurality of teacher classification models (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This partitioned data is used to generate a plurality of expert models (EManyshot, EMediumshot, and EFewshot). These models constitute an ensemble set of models trained using different parameters).
Sharma fails to specifically disclose using different initial random seeds of the training dataset.
However, Xiang, which is analogous to the claimed invention because it is directed toward training using long-tailed classification, discloses using different initial random seeds of the training dataset (Figures 1 and 3; Abstract; Section 4: Here a plurality of cardinality subsets are created and used for training the student model. After each training epoch, a measurement of the performance gap is taken and the scheme is appropriately weighted). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Xiang with Sharma, with a reasonable expectation of success, as it would have allowed for improved classification using student models (Xiang: Section 4.2).
As per dependent claim 6, Sharma and Hui disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Sharma discloses wherein the one or more teacher classification models comprise an ensemble of a plurality of teacher classification models (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This partitioned data is used to generate a plurality of expert models (EManyshot, EMediumshot, and EFewshot). These models constitute an ensemble set of models trained using different parameters).
Sharma fails to specifically disclose the models have a same initial parameterization but are trained on different randomly-selected subsets of training data.
However, Xiang, which is analogous to the claimed invention because it is directed toward training using long-tailed classification, models have a same initial parameterization but are trained on different subsets of training data (Figures 1 and 3; Abstract; Section 4: Here a plurality of cardinality subsets are created and used for training the student model. After each training epoch, a measurement of the performance gap is taken and the scheme is appropriately weighted). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Xiang with Sharma, with a reasonable expectation of success, as it would have allowed for improved classification using student models (Xiang: Section 4.2).
Further, the examiner takes official notice that it was notoriously well-known in the art at the time of the applicant’s effective filing date to randomly select datasets for training. It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined the well-known with Sharma-Xiang, with a reasonable expectation of success, as it would have allowed for selecting datasets for training the individual student models.
Claim 8 is rejected under 35 U.S.C. 103 as being unpatentable over Sharma and Hui and further in view of Zhang et al. (To Balance or Not to Balance: A Simple-yet-Effective Approach for Learning with Long-Tailed Distributions, 10 December 2019, hereafter Zhang).
As per dependent claim 8, Sharma and Hui disclose the limitation similar to those in claim 1, and the same rejection is incorporated herein. Sharma discloses wherein training, by the computing system, the one or more student classification models to predict data generated by the one or more teacher classification models via distillation training comprises a distillation loss term to train a feature extractor of the one or more student classification models to predict feature representations similar to a feature extractor of one or more teacher classification models (Section 3: Here, a class-imbalance training set is received and sorted according to class frequencies and partitioned into contiguous class balanced subsets (Manyshot, Mediumshot, and Fewshot). This partitioned data is used to generate a plurality of expert models (EManyshot, EMediumshot, and EFewshot). These models constitute an ensemble set of models trained using different parameters).
Sharma fails to specifically disclose back propagating to train a feature extractor. However, Zhang, which is analogous to the claimed invention because it is directed toward learning with long-tailed distributions, discloses back propagating to train a feature extractor (Section 3.2: Here, the losses from the CBS and RRS schemes are used to retrain the model). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Zhang with Sharma, with a reasonable expectation of success, as it would have allowed retraining the feature extractor using a loss function (Zhang: Section 3.2).
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Sharma and Hui and further in view of Kang et al. (Decoupling Representation and Classifier for Long-Tailed Recognition, 21 October 2019, hereafter Kang).
As per dependent claim 9, Sharma and Hui disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Sharma fails to specifically disclose wherein the one or more teacher classification models comprise a cosine classifier. However, Kang, which is analogous to the claimed invention because it is directed toward classifying long-tailed data, discloses a cosine classifier (Section 4). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Kang with Sharma, with a reasonable expectation of success, as it would have allowed using a known classifier for classifying data.
Claims 12-13 and 15-16 are rejected under 35 U.S.C. 103 as being unpatentable over Sharma and Hui and further in view of Wang et al. (Knowledge Distillation and Student-Teacher Learning for Visual Intelligence: A Review and New Outlooks, 6 January 2021, hereafter Wang).
As per dependent claim 12, Sharma and Hui disclose the limitations similar to those in claim 10, and the same rejection is incorporated herein. Sharma discloses a dataset and classification predictions of one or more classifications (Section 3).
Sharma fails to specifically disclose one or more samples of audio data. However, Wang discloses training and classifying based on a plurality of different data types (Section 5). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Wang with Sharma, with a reasonable expectation of success, as it would have allowed for training and classifying different types of data.
Further, the examiner takes official notice that it was notoriously well-known in the art at the time of the applicant’s effective filing date that audio data is a known data type. It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combine the well-known with Sharma-Wang, with a reasonable expectation of success, as it would have allowed for training and classifying based on audio data.
As per dependent claim 13, Sharma and Hui disclose the limitations similar to those in claim 10, and the same rejection is incorporated herein. Sharma discloses a dataset and classification predictions of one or more classifications (Section 3).
Sharma fails to specifically disclose one or more samples of audio data. However, Wang discloses training and classifying based on a plurality of different data types (Section 5). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Wang with Sharma, with a reasonable expectation of success, as it would have allowed for training and classifying different types of data.
Further, the examiner takes official notice that it was notoriously well-known in the art at the time of the applicant’s effective filing date that classifying actions to be taken by an autonomous agent (self-driving) or robot were a known data type. It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combine the well-known with Sharma-Wang, with a reasonable expectation of success, as it would have allowed for training and classifying based on actions to be taken by an autonomous agent or robot.
As per dependent claim 15, Sharma and Hui disclose the limitations similar to those in claim 10, and the same rejection is incorporated herein. Sharma discloses a dataset and classification predictions of one or more classifications (Section 3).
Sharma fails to specifically disclose one or more samples of audio data. However, Wang discloses training and classifying based on a plurality of different data types (Section 5). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Wang with Sharma, with a reasonable expectation of success, as it would have allowed for training and classifying different types of data.
Further, the examiner takes official notice that it was notoriously well-known in the art at the time of the applicant’s effective filing date that text data is a known data type. It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combine the well-known with Sharma-Wang, with a reasonable expectation of success, as it would have allowed for training and classifying based on text data.
As per dependent claim 16, Sharma and Hui disclose the limitations similar to those in claim 1, and the same rejection is incorporated herein. Sharma discloses a dataset and classification predictions of one or more classifications (Section 3).
Sharma fails to specifically disclose one or more samples of audio data. However, Wang discloses training and classifying based on a plurality of different data types (Section 5). It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combined Wang with Sharma, with a reasonable expectation of success, as it would have allowed for training and classifying different types of data.
Further, the examiner takes official notice that it was notoriously well-known in the art at the time of the applicant’s effective filing date that audio data is a known data type. It would have been obvious to one of ordinary skill in the art at the time of the applicant’s effective filing date to have combine the well-known with Sharma-Wang, with a reasonable expectation of success, as it would have allowed for training and classifying based on audio data.
Response to Arguments
Applicant’s arguments, have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new ground(s) of rejection is made in view of Sharma and Hui.
Additionally, the factual assertions set forth in the Office Action dated 14 January 2026 have not been traversed. According to MPEP 2144.03 (C) the official notice statement is taken to be admitted prior art because the appellant failed to traverse the examiner’s assertion.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant's disclosure:
Shekhar et al. (US 2022/0253630): Discloses removing class imbalance in samples through a selection policy (paragraph 0158)
Singh et al. (US 2022/0188568): Discloses recalibration to address class-imbalance samples (paragraph 0123)
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). 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 KYLE R STORK whose telephone number is (571)272-4130. The examiner can normally be reached 8am - 2pm; 4pm - 6pm.
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, Omar Fernandez Rivas can be reached at 571/272-2589. 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.
/KYLE R STORK/Primary Examiner, Art Unit 2128
Prosecution Insights