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 .
Response to Arguments
Applicant's arguments filed 4/6/2026 have been fully considered but they are not persuasive.
Applicant asserts a table on pages 11 and 12. First, the argued “Claims” in the left column are not claims. For example, the word “data stream” does not appear in the claims. Second, Examiner disagrees with Applicant’s characterization of the references in the right three columns. Third, the only argued, and claimed, element is “sampling the training data at a predetermined rate…” Claim 1. This is taught by Bagus sec. II (B) p. 3 “’Random’ selection chooses n samples per class from each sub-task…” The predetermined rate is n per class. In conclusion, because applicant argues unclaimed elements, mischaracterizes the office action’s cited references, and because the art teaches or makes obvious all the claim elements, Applicant’s assertion is not persuasive.
Applicant states that the references “address fundamentally different technical problems.” Remarks 12. Examiner disagrees. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to sample hard negatives “it yields consistent and significant boosts in detection performance on benchmarks…” Shri abs. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to use a memory mapped method and predetermined sampling rate because these replay method “outperform other methods under a broad variety of application-oriented constraints.” Bagus sec. IV (E) p. 8. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to compare model performance in order to select “the best model from this set [of models].” Ali abs. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to use an F1 score for training because it is an industry standard for measuring accuracy.
Applicant mischaracterizes Shri and Bagus and then argues against the combination of the mischaracterized references. Remarks 13 sec. 4b&c. Examiner disagrees with Applicants unsupported assertions.
Applicant opines that “three+ references requires impermissible hindsight.” Remarks 14. This is not true and it is not a rule.
Applicant argues “technical advantages” of a “claimed triple-integration pipeline”. Remarks 13. Triple integration pipeline is not claimed. No argument is necessary for unclaimed elements.
Claim Rejections - 35 USC § 101
All 101 rejections withdrawn, thank you.
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.
Claims 1, 2, 5, 7, 10, 11, 14 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over A Survey on Active Deep Learning: From Model Driven to Data Driven by Peng et al, Training Region-based Object Detectors with Online Hard Example Mining by Shrivastava et al (Shri) and An Investigation of Replay-based Approaches for Continual Learning by Bagus et al.
Claims 3 and 12 are rejected under 35 U.S.C. 103 as being unpatentable over A Survey on Active Deep Learning: From Model Driven to Data Driven by Peng et al, Training Region-based Object Detectors with Online Hard Example Mining by Shrivastava et al (Shri), An Investigation of Replay-based Approaches for Continual Learning by Bagus et al and Active Learning with Model Selection by Ali et al.
Claims 6 and 15 are rejected under 35 U.S.C. 103 as being unpatentable over A Survey on Active Deep Learning: From Model Driven to Data Driven by Peng et al, Training Region-based Object Detectors with Online Hard Example Mining by Shrivastava et al (Shri), An Investigation of Replay-based Approaches for Continual Learning by Bagus et al and US 20180260735 A1 to Arad et al.
Peng teaches claims 1 and 10. A method for operating a learning model executed by an apparatus for operating a learning model, the method comprising:
selecting at least one training data among first training data selected through (Peng fig. 4 p. 221:13 “Using uncertainty as a single metric to select samples, many similar samples tend to cluster into a small area. (b) Combining uncertainty with representativeness, it can prevent the selected points from being too similar and lacking representativeness.” The representativeness is the data labeled by active learning of a previous mode. Sampling high uncertainty data is sampling. Peng sec. 3.5 and p. 221:14 also teaches this hybrid method “The “multi-criteria” selected informative samples by simultaneously considering density, similarity, uncertainty, and label-based measure…”)
learning the previous learning model anew using the at least one selected training data, wherein learning the previous learning model anew performs an operation of continual-learning using a (Peng abs “Fig. 1. Active learning trains the predictor with initial training samples and uses its selector to select a few of unlabeled samples, labels them, adds them into the training dataset, and then re-trains the predictor.” And fig.1 below. Retraining the predictor is learning the previous model anew.)
PNG
media_image1.png
162
388
media_image1.png
Greyscale
Peng doesn’t teach an explicit hard-negative.
However, Shri teaches hard-negative sampling (Shri sec. 1 p. 2 “training examples are sampled according to a non-uniform, non-stationary distribution that depends on the current loss of each example under consideration.”)
The claims, Peng and Shri all sample training data to increase model performance. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to sample hard negatives “it yields consistent and significant boosts in detection performance on benchmarks…” Shri abs.
Peng doesn’t teach memory mapping and a predetermined rate.
However, Bagus teaches a memory mapping method and a predetermined rate. (Bagus sec. II p. 2 “The sample selection strategy defines two criteria: • which sample is a potential candidate to be included in the buffer (referred-to as sample-in) • which sample is a potential candidate to be removed from the buffer (referred-to as sample-out)…” Bagus sec. II (B) p. 3 “’Random’ selection chooses n samples per class from each sub-task,”)
The claims, Peng and Bagus all sample training data for active/continual learning. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to use a memory mapped method and predetermined sampling rate because these replay method “outperform other methods under a broad variety of application-oriented constraints.” Bagus sec. IV (E) p. 8.
Peng teaches claims 2 and 11. The method of claim 1, wherein the selecting at least one training data evaluates the at least one selected training data through the previous learning model and selects the training data with the highest ratio of a predetermined evaluation index among the at least one selected training data. (Peng sec. 3.1 p. 221:8 “Uncertainty may be the most often used metric by the selectors in active learning. This kind of method assumes that the uncertainty samples provide more information for the training of predictors if they are labeled.” Peng sec. 3.1 p. 221:9 “a new active selection criterion based on the sensitiveness of unlabeled examples to adversarial attacks such as DeepFool [41].”)
Peng teaches claims 3 and 12. The method of claim 1, further including:
learning the previous learning model anew using the rest of the selected training data (Peng abs “Fig. 1. Active learning trains the predictor with initial training samples and uses its selector to select a few of unlabeled samples, labels them, adds them into the training dataset, and then re-trains the predictor.” Retraining the predictor is learning the previous model anew and updating to the newly learned model.)
Peng doesn’t teach model comparison.
However, Ali teaches comparing the newly learned learning model with the previous learning model in terms of at least one of performance, accuracy, and speed; and (Ali sec. 3.2 “The value of information for training VOIT ,M(x) measures the improvement in model M’s accuracy after training on a point x ∈ P.” VOIT is a value of information for training metric that is used by the selector to pick training data to then compare across models. Ali sec. 3 “At the end of active learning, ALMS outputs the model M∗ ∈ M it expects will perform best on future test data after re-training that model on all data in the union of the train and validation sets.”)
learning the previous learning model anew using the rest of the selected training data if the newly learned learning model is inferior to the previous learning model in terms of at least one of performance, accuracy, and speed and updating the previous learning model to the newly learned learning model if the newly learned learning model is better than the previous learning model in terms of at least two or more of performance, accuracy, and speed. (Ali sec. 3 “At the end of active learning, ALMS outputs the model M∗ ∈ M it expects will perform best on future test data after re-training that model on all data in the union of the train and validation sets.”)
The claims, Ali and Peng are all active learners with selective training data. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to compare model performance in order to select “the best model from this set [of models].” Ali abs.
Shri teaches claims 5 and 14. The method of claim 4, wherein the selecting at least one training data selects first training data satisfying conditions for hard-negative sampling by sampling training data exceeding a predetermined prediction error value from previous training data. (Shri sec. 4.1 “Given a list of RoIs and their losses, NMS works by iteratively selecting the RoI with the highest loss, and then removing all lower loss RoIs that have high overlap with the selected region.” RoI with the highest loss’ loss is the threshold.)
Peng teaches claims 6 and 15. The method of claim 4, wherein the selecting at least one training data calculates a performance index value for determining the performance obtained by applying specific training data in the previous training model using at least one performance index among (Peng abs “Fig. 1. Active learning trains the predictor with initial training samples and uses its selector to select a few of unlabeled samples, labels them, adds them into the training dataset, and then re-trains the predictor.” Peng sec. 3.1 p. 221:8 “Uncertainty may be the most often used metric by the selectors in active learning…”)
Peng doesn’t teach selecting based on F1 Score, accuracy etc.
However, Arad teaches selecting at least one training data calculates a performance index value for determining the performance obtained by applying specific training data in the previous training model using at least one performance index among F1-score, accuracy, mean Average Precision (mAP), and mean Intersection over Union (mIoU). (Arad para 18 “comparing the first F1-score and the second F1-score, wherein in response to a determination that the first F1-score is greater than the second F1-score, removing the misclassified sample from the training set.” F1 score is also a measure of accuracy.)
Peng, Arad, and the claims all train on sampled data. It would have been obvious to a person having ordinary skill in the art, at the time of filing, to use an F1 score for training because it is an industry standard for measuring accuracy.
Peng teaches claims 7 and 16. The method of claim 4, wherein the selecting at least one training data calculates an uncertainty score through a score model from the new training data and labels new training data for which the calculated uncertainty score exceeds a predetermined threshold value as the second training data. (Peng sec. 3.1 p. 221:8 “Uncertainty may be the most often used metric by the selectors in active learning…” Peng sec. 2.2 p. 221:6 “Pool-Based Sampling usually use one selector to measure all un-annotated samples and then it ranks all data-point in the un-annotated set and finally selects the top N samples to be annotated.”)
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 Austin Hicks whose telephone number is (571)270-3377. The examiner can normally be reached Monday - Thursday 8-4 PST.
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, Mariela Reyes can be reached at (571) 270-1006. 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.
/AUSTIN HICKS/Primary Examiner, Art Unit 2142
Prosecution Insights