NON-FINAL REJECTION, FIRST DETAILED ACTION
Status of Prosecution
The present application 18/497,400, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
The application was filed in the Office on Oct. 30, 2023. This application serves as the priority application to PCT/US24/53667 filed on Oct. 30, 2024.
Claims 1-20 are pending and are all rejected in this rejection. Claims 1, 11 and 20 are independent claims.
Status of Claims
Claims 1, 10-11, 19 and 20 are rejected under 35 USC § 103 as being unpatentable over Hoffman et al. (“Hoffman”), United States Patent Application Publication 2023/0315532 published on Oct. 5, 2023, in view of non-patent literature Banerjee et al. (“Banerjee”), “Budgeted Subset Selection for Fine-tuning Deep Learning Architectures in Resource-Constrained Applications,” published in 2020.
Claims 2-5 and 12-15 are rejected under 35 USC § 103 as being unpatentable over Hoffman in view of Banerjee in further view of Capelo et al. (“Capelo”), United States Patent 11,379,140 published on July 5, 2022.
Claims 6-7 and 16-17 are rejected under 35 USC § 103 as being unpatentable over Hoffman in view of Banerjee in view of Capelo and in further view of Ospenica et al. (“Ospenica”), United States Patent 11,640,323 published on May 2, 2023.
Claims 8 and 18 are rejected under 35 USC § 103 as being unpatentable over Hoffman in view of Banerjee in further view of Bowers et al. (“Bowers”), United States Patent Application Publication 9,996,804 published on June 12, 2018.
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 of this title, 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.
-A.
Claims 1, 10-11, 19 and 20 are rejected under 35 USC § 103 as being unpatentable over Hoffman et al. (“Hoffman”), United States Patent Application Publication 2023/0315532 published on Oct. 5, 2023, in view of non-patent literature Banerjee et al. (“Banerjee”), “Budgeted Subset Selection for Fine-tuning Deep Learning Architectures in Resource-Constrained Applications,” published in 2020.
As to Claim 1, Hoffman teaches: A computer-implemented method comprising:
accessing training data, a computing resource limit setting, and parameters of a machine learning model (Hoffman: Fig. 2, par. 0105, data defining a compute budget [210] (i.e. a computing resource limit setting); par. 0106 allocation mapping parameters (i.e. parameters of a machine learning model));
forming, at a server, a machine learning training strategy based on the training data and the computing resource limit setting (Hoffman: par. 0106, at [220], using the training data and the compute budget and allocation mapping parameters, an allocation tuple is generated (i.e. a machine learning training strategy), that defines the target model size and a target amount of training data for training the machine learning model);
forming a machine learning model configuration based on the machine learning training strategy (Hoffman: par. 0107, the training system instantiates the machine learning model per the strategy at [230]).
selecting data from the training data based on the machine learning training strategy (Hoffman: pars. 0108, “The system can identify a subset of the corpus of training data that includes the target amount of training data, e.g., by randomly sampling training data from the corpus of training data, and then retrieve the selected training data for use in training the machine learning model”); and
providing the machine learning model configuration and the data to a machine learning platform (Hoffman: par. 0110, at [250], the model is trained with the given configuration and data).
PNG
media_image1.png
751
684
media_image1.png
Greyscale
Hoffman may not explicitly teach: selecting sampled data from the training data based on the machine learning training strategy; and
providing the machine learning model configuration and the sampled data to a machine learning platform.
Banerjee teaches in general concepts related to budgeted subset selection for fine-tuning deep learning architectures in a manner addressing data privacy and cost (Banerjee: Abstract). Specifically, Banerjee teaches that a specific set of samples are selected from the training data in a manner to consider resource constraints so that maximum generalization capability is possible (Banerjee: p. 1-2, “We are given a parameter k (k << N), which is the maximum allowable training set size (in addition to the initial training data), considering the resource constraints. Which k samples should we select from the set of N for fine-tuning, so that we derive a model with maximum generalization capability?”).
It would have been obvious to a person having ordinary skill in the art at a time before the effective filing date of the application to have modified the Hoffman disclosures and teachings by sampling the training data for use as taught by Banerjee. Such a person would have been motivated to do so with a reasonable expectation of success to optimize the training in a resource efficient manner.
As to Claim 9, Hoffman and Banerjee teach the elements of claim 1.
Hoffman further teaches: wherein the machine learning platform is operated on a second server, the second server being configured to:
train, using the machine learning platform, the machine learning model with the sampled data and the machine learning model configuration, an output of training the machine learning model comprising a trained machine learning model;
monitor computing resources of the machine learning platform during training of the machine learning model on the second server, an output of monitoring the computing resources comprising resource usage data; and
provide the resource usage data and the trained machine learning model to a client device (Hoffman: pars. 0188-89 notes that clients and servers architecture may exist to implement the system. Examiner asserts it would be design choice to select which components and where to perform the functions).
As to Claim 10, Hoffman and Banerjee teach the elements of claim 1.
Hoffman further teaches: wherein the machine learning training strategy comprises one of a random forest classifier (Hoffman: par. 0030, a random forest model) or a gaussian process regressor.
As to Claim 11, it is rejected for similar reasons as claim 1. Hoffman further teaches a processor and memory (Hoffman: par. 0179, 0184).
As to Claim 19, it is rejected for similar reasons as claim 9.
As to Claim 20, it is rejected for similar reasons as claim 1 and 11.
B.
Claims 2-5 and 12-15 are rejected under 35 USC § 103 as being unpatentable over Hoffman et al. (“Hoffman”), United States Patent Application Publication 2023/0315532 published on Oct. 5, 2023, in view of non-patent literature Banerjee et al. (“Banerjee”), “Budgeted Subset Selection for Fine-tuning Deep Learning Architectures in Resource-Constrained Applications,” published in 2020 in further view of Capelo et al. (“Capelo”), United States Patent 11,379,140 published on July 5, 2022.
As to Claim 2, Hoffman and Banerjee teach the elements of claim 1.
Hoffman further teaches: training, using the machine learning platform at the server, the machine learning model with the sampled data and the machine learning model configuration, an output of training of the machine learning model comprising a trained machine learning model (Horrman: par. 0107-110, the machine learning model is trained based on the training data and the configuration and parameters).
Hoffman and Banerjee may not explicitly teach: monitoring computing resources of the machine learning platform during training of the machine learning model, an output of monitoring computing resources comprising resource usage data.
Capelo teaches in general concepts related to a large-scale machine learning experiment that allows scheduling an a user-specified run (Capelo: Abstract). Specifically, Capelo teaches that experiment parameters may be gathered in a training phase of the model lifecycle and used to determine the computing resources needed (Capelo: Fig. 7; col. 24, lines 63 to col. 25, line 8). The run metrics, including telemetry data of the computing resources may be monitored (Capelo: col. 29, lines 1 to 19, [S700]).
PNG
media_image2.png
537
736
media_image2.png
Greyscale
It would have been obvious to a person having ordinary skill in the art at a time before the effective filing date of the application to have modified the Hoffman-Banerjee disclosures and teachings by monitoring the computing resources for use as taught by Capelo. Such a person would have been motivated to do so with a reasonable expectation of success to optimize the training in a resource efficient manner with an agent to perform actions including monitoring the run progress (Capelo: col. 2, lines 47 to 60).
As to Claim 3, Hoffman, Banerjee and Capelo teach the elements of claim 2.
Hoffman further teaches: generating an updated machine learning model configuration recommendation based on the resource usage data (Hoffman: par. 0111, the training of the model is performed using objective functions and other techniques to meet the objective loss functions as needed, which would include the resource usage optimization).
As to Claim 4, Hoffman, Banerjee and Capelo teach the elements of claim 2.
Hoffman further teaches: receiving, from a client device, the training data, the computing resource limit setting, and parameters of the machine learning model; and
providing the resource usage data and the trained machine learning model to the client device (Hoffman: pars. 0188-89, a client-server system may be implemented for the different aspects of the machine learning system).
As to Claim 5, Hoffman, Banerjee and Capelo teach the elements of claim 2.
Capelo further teaches: wherein the resource usage data indicate a time length and peak memory used during the training of the machine learning model (Capelo: col. 5, lines 45-48, the monitored metrics include run time and memory consumed).
As to Claim 12, it is rejected for similar reasons as claim 2.
As to Claim 13, it is rejected for similar reasons as claim 3.
As to Claim 14, it is rejected for similar reasons as claim 4.
As to Claim 15, it is rejected for similar reasons as claim 5.
C.
Claims 6-7 and 16-17 are rejected under 35 USC § 103 as being unpatentable over Hoffman et al. (“Hoffman”), United States Patent Application Publication 2023/0315532 published on Oct. 5, 2023, in view of non-patent literature Banerjee et al. (“Banerjee”), “Budgeted Subset Selection for Fine-tuning Deep Learning Architectures in Resource-Constrained Applications,” published in 2020 in further view of Capelo et al. (“Capelo”), United States Patent 11,379,140 published on July 5, 2022 and in further view of Ospenica et al. (“Ospenica”), United States Patent 11,640,323 published on May 2, 2023.
As to Claim 6, Hoffman, Banerjee and Capelo teach the elements of claim 2.
Hoffman, Banerjee and Capelo further teaches: wherein forming the machine learning training strategy comprises:
providing a summary of the training data and the computing resource limit setting to a resource estimator and an efficiency accuracy trade-off modeler (Hoffman: Fig. 10, par. 0170, the optimization system (i.e. which include the resource estimator and an efficiency accuracy trade-off modeler) determines parameters for an estimation function that takes in parameters including input model size and input amount of training data (i.e. computing resource limit setting) in [1010]); and
receiving the machine learning training strategy from the resource estimator and the efficiency accuracy trade-off modeler (Hoffman: Fig. 10, the values are received to determine the allocation mapping parameters using the performance estimation function).
PNG
media_image3.png
932
956
media_image3.png
Greyscale
Hoffman, Banerjee and Capelo may not explicitly teach: wherein the resource estimator is configured to estimate a complexity of the machine learning model based on the summary of the training data;
wherein the efficiency accuracy trade-off modeler is configured to model a trade-off between an efficiency of the machine learning model and an accuracy of the machine learning model.
Ospenica teaches in general concepts related to a learning model that utilizes a training algorithm that is adapted so that the amount of available computing resources in the edge cloud is sufficient for computation ins the training algorithm for an industrial process (Ospenica: Abstract). Specifically, Ospenica teaches that whether sufficient resources for a machine learning objective are available and if not complexity is therefore reduced (Ospenica: Fig. 5, col. 9m lines 43 to 67, steps [510 to [512]). Examiner asserts that this is a calculation of complexity and determination of a tradeoff between efficiency and accuracy that is calculated in determining sufficient resources and the machine learning objective at hand.
PNG
media_image4.png
429
535
media_image4.png
Greyscale
It would have been obvious to a person having ordinary skill in the art at a time before the effective filing date of the application to have modified the Hoffman-Banerjee-Capelo disclosures and teachings by including the estimator and modeler as taught by Ospenica. Such a person would have been motivated to do so with a reasonable expectation of success to allow for the reduction of complexity of the models appropriately to achieve the machine learning objectives in an optimized manner (Ospenica: col. 2, lines 28 to 41, “there are typically certain constraints in the above-described edge clouds in terms of resource availability, and the amount of available computing resources in an edge cloud is thus limited.”).
As to Claim 7, Hoffman, Banerjee, Capelo and Ospina teach the elements of claim 6.
Hoffman, Banerjee, Capelo and Ospina as combined further teaches: providing the resource usage data and the machine learning training strategy to the resource estimator and the efficiency accuracy trade-off modeler, wherein the resource estimator and the efficiency accuracy trade-off modeler are configured to generate an updated machine learning training strategy;
forming an updated machine learning model configuration based on the updated machine learning training strategy; and
providing the updated machine learning model configuration and the sampled data to the machine learning platform (Examiner asserts that the models are updated and optimized per the iterative and feedback-loops disclosed in the machine learning lifecycles of each of the references).
As to Claim 16, it is rejected for similar reasons as claim 6.
As to Claim 17, it is rejected for similar reasons as claim 7.
D.
Claims 8 and 18 are rejected under 35 USC § 103 as being unpatentable over Hoffman et al. (“Hoffman”), United States Patent Application Publication 2023/0315532 published on Oct. 5, 2023, in view of non-patent literature Banerjee et al. (“Banerjee”), “Budgeted Subset Selection for Fine-tuning Deep Learning Architectures in Resource-Constrained Applications,” published in 2020 in further view of Bowers et al. (“Bowers”), United States Patent Application Publication 9,996,804 published on June 12, 2018.
As to Claim 8, Hoffman and Banerjee teach the elements of claim 1.
Hoffman further teaches: accessing a performance assessment of the deployed machine learning mode (Hoffman: Fig. 10, par. 0169-70, a performance estimation function is utilized to determine the efficacy of the system); and
generating a performance indicator of the deployed machine learning model based on the performance assessment (Hoffman: par. 0172, a performance measure is calculated);
determining that the performance indicator of the deployed machine learning model transgresses performance threshold (Hoffman: par. 0175, a that optimize the performance estimation function subject to a constraint (i.e. a performance threshold).
Hoffman may not explicitly teach: testing a deployed machine learning model based on the machine learning model configuration;
generating a performance indicator of the deployed machine learning model based on the testing and the performance assessment;
determining that the performance indicator of the deployed machine learning model transgresses a deployed machine learning model performance threshold;
in response to determining that the performance indicator of the deployed machine learning model transgresses the deployed machine learning model performance threshold, updating the deployed machine learning model,
wherein updating the deployed machine learning model further comprises:
updating the machine learning training strategy and the sampled data based on the performance indicator of the machine learning model.
Bowers teaches in general concepts related to tracking one or more machine learning models for one or more application services(Bowers: Abstract). Specifically, Bowers teaches that models are tested against deployed machine learning models (in production) (Bowers: Fig. 4). If a system falls below certain threshold constraints, an update is automatically pushed to replace the production model with a latent model (Bowers: col. 5, lines 23 to 37).
PNG
media_image5.png
740
550
media_image5.png
Greyscale
It would have been obvious to a person having ordinary skill in the art at a time before the effective filing date of the application to have modified the Hoffman-Banerjee disclosures and teachings by updating the model per falling below threshold constraints as taught by Bowers. Such a person would have been motivated to do so with a reasonable expectation of success to allow for the reduction of complexity of the models appropriately to allow for live testing and production update for the optimal model performance and experience (Bowers: col. 5, lines 10 to 22).
As to Claim 18, it is rejected for similar reasons as claim 8.
Conclusion
Relevant prior art not relied upon but made of the record:
Aggarawal, US Patent Application Publication 2025/0094861 (Mar. 20, 2025) (teaching time-bound hyperparameter tuning).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to JAMES T TSAI whose telephone number is (571)270-3916. The examiner can normally be reached M-F 8-5 Eastern.
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, Viker Lamardo can be reached on 571-270-5871. 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./JAMES T TSAI
/JAMES T TSAI/ Primary Examiner, Art Unit 2147
Prosecution Insights