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 .
DETAILED ACTION
This action is responsive to the following communication: Amendment filed Jul. 7, 2026. This action is made non-final.
Claims 1-20 are pending in the case. Claims 1, 11 and 19 are independent claims.
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 1-20 are rejected under 35 U.S.C. 103 as being unpatentable over Chai et al. (hereinafter Chai) U.S. Patent Publication No. 2022/0091837 in view of Kennedy (hereinafter Kennedy) U.S. Patent Publication No. 2020/0311611.
With respect to independent claim 1, Chai teaches a computing system for generating models, comprising: a processor; and memory encoding instructions which, when executed by the processor (see e.g., Para [4][88]), cause the computing system to:
store a library including modules, the modules being programmed to perform tasks of tools of different machine learning models (see e.g., Para [34]-[39]- “a lightweight machine learning library designed specifically for mobile and embedded devices. In one example, the platform can use a conversion tool known as TensorFlow Lite Optimizing Converter (“TOCO”) to convert a standard TensorFlow graph of a model into a TensorFlow Lite graph, where TensorFlow Lite is a lightweight machine learning library designed for mobile applications. … The compression/conversion service can also provide various tools and dashboards to enable the developer to explore and control compression options. For example, the tools can show tradeoffs with quality, enable parameter tuning, and/or other controls that allow granular control of the compression outcome. For example, the developer can use the compression service to generate multiple models with different sizes and corresponding tradeoffs. These multiple models can be used as part of a model targeting scheme, as described further below.”); and
incorporate different ones of the modules into the different machine learning models while the different machine learning models are being built, the different ones of the modules being incorporated into the different machine learning models in response to different calls made via an Application Programming Interface (API), the API being an interface between the library and a computer application configured to receive commands for building the different machine learning models (see e.g., Fig. 10-14 Para [4][35]-[38][221][226]- “The machine intelligence software development kit is configured to store one or more machine-learned models and a machine learning library. The machine intelligence software development kit is configured to communicate with the computer application using an application programming interface to receive input data from the computer application. The machine intelligence software development kit is configured to implement the one or more machine-learned models and machine learning library on-device to produce an inference based at least in part on the input data. “).
Chai does not expressly show the tools including a feature engineering tool, and the tasks including a task of the feature engineering tool that identifies variables from input data that are relevant to determining predicted outcomes by a machine learning model being built. However, Kennedy teaches similar feature (see e.g. para [39]-[70]– “feature generation module 51 may select one or more features from the list of features and apply a transformation operation, a combination operation, or both to generate new synthetic features… feature selection module 55 selects a specific number of features that have the highest feature scores, to define filtered feature list 77 … Wrapper method module 57 may iteratively select various subsets of features from the list of features provided thereto by feature selection module 55 and train a machine learning algorithm using those selected subsets of features.” feature generation, scoring, selection, and wrapper modules correspond to the “feature engineer tools” and these modules receives an input labeled dataset and, via statistical test and wrapped-based selection, identify and select features/variables that are predictive of the target characteristic, providing them to a machine learning tool for training and prediction.). Both Chai and Kennedy are directed to machine learning model training. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Chai and Kennedy in front of them to modify the system of Chai to include the above feature. The motivation to combine Chai and Kennedy comes from Kennedy. Kennedy discloses the motivation to provide tools to analyze input data and identify subset of variables to better predict the outcome (see e.g. para [39]-[70]). This motivation for combination also applies to the remaining claims which depend on this combination.
With respect to dependent claim 2, the modified Chai teaches using the different machine learning models, predict outcomes based on data inputs (see e.g., Para [42]-[45] – “learn from various input sources; cover a wide range of prediction tasks; support a plurality of use-cases; be powerful in terms of learnable machine-learned models and/or architectures supported; be compatible with infrastructure requirements, such as TensorFlow Lite integration, fast training (e.g., through the use of distributed techniques), access by internal and/or external users (e.g., developers), and platform integration for SDK; and be easily used.”).
With respect to dependent claim 3, the modified Chai teaches the tools include a training tool; and wherein the tasks include a task of the training tool that provides input data to the machine learning model being built to train the machine learning model being built (see e.g., Para [34][74]-[78] – “developers can have access to a single SDK for all machine learning services.”” the machine intelligence SDK can further include a dedicated machine learning library that can be implemented by the application to run and/or train the models included in the machine intelligence SDK on-device. “).
With respect to dependent claim 4, the modified Chai teaches the tools include a data filtering tool; and wherein the tasks include a filtering task of the data filtering tool that discards a portion of input data (Chai does not expressly show this feature. However, Chai expressly indicates “developers can have access to a single SDK for all machine learning services” see e.g., Para [34]. Further, Chai discloses data cleaning feature see Para [194]- “Training data creation: The platform can assist developers to augment and manage their training data. Data augmentation can include adding more training samples (e.g., via an image search), leveraging crowd-sourcing platforms to provide labels, and introducing/or transformations of existing samples (e.g., add noises, rotations, perturbations, etc.). The platform can also provide data cleaning and visualization tools.” Therefore, it would have been obvious to include the above feature. Also see Kennedy para [39]-[70]).
With respect to dependent claim 5, the modified Chai teaches the tools include a preprocessing tool; and wherein the tasks include a preprocessing task of the preprocessing tool that converts input data having a data format into another data format that can be processed by a machine learning model being built (Chai does not expressly teach a preprocessing tool. However, However, Chai expressly indicates “developers can have access to a single SDK for all machine learning services” see e.g., Para [34]. Furthermore, Chai discloses that the machine leaning framework learns from various input sources and integrates with TensorFlow - see e.g., Para [43]-[45][194]-[198] -” learn from various input sources; cover a wide range of prediction tasks; support a plurality of use-cases; be powerful in terms of learnable machine-learned models and/or architectures supported; be compatible with infrastructure requirements” Therefore, it would have been obvious to include the above feature. Also see Kennedy para [39]-[70]).
With respect to dependent claim 7, the modified Chai teaches the tools include a scoring tool; and wherein the tasks include a scoring task of the scoring tool that measures a performance of the machine learning model being built by comparing predicted outcomes generated by the machine learning model being built to known data (Chai does not expressly teach a scoring tool. However, Chai expressly indicates “developers can have access to a single SDK for all machine learning services” see e.g., Para [34]. Further, Chai discloses ”[t]he model manager 120 can provide various model management services, including, as examples, a model compression service, a model conversion service, a model evaluation service” Therefore, it would have been obvious to include the above feature. Also see Kennedy para [39]-[70]).
With respect to dependent claim 8, the modified Chai teaches the scoring tool includes a classification tool; and wherein the tasks include a classification task of the classification tool that maps a function of input variables learned by the machine learning model being built to one or more discrete output variables corresponding to the predicted outcomes (Chai does not expressly teach a classification tool. However, Chai expressly indicates “developers can have access to a single SDK for all machine learning services” see e.g., Para [34]. The examiner notes that classification tool is well-known in the art. It would have been obvious to include this feature. Also see Kennedy para [39]-[70]).
With respect to dependent claim 9, the modified Chai teaches the scoring tool includes a regression tool; and wherein the tasks include a regression task of the regression tool that maps a function of input variables learned by the machine learning model being built to a continuous output variable corresponding to the predicted outcomes (Chai does not expressly teach a regression tool. However, Chai expressly indicates “developers can have access to a single SDK for all machine learning services” see e.g., Para [34]. The examiner notes that regression tool is well-known in the art. It would have been obvious to include this feature. Also see Kennedy para [39]-[70]).
With respect to dependent claim 10, the modified Chai teaches a software development kit operable with a plurality of differently configured computer applications each of which is configured to receive commands for building the different machine learning models using tools of the software development kit (see e.g., Para [34][192][208] – “ the present disclosure is directed to an application development platform and associated software development kits (“SDKs”) that provide comprehensive services for generation, deployment, and management of machine-learned models used by computer applications such as, for example, mobile applications executed by a mobile computing device. In particular, the application development platform and SDKs can provide or otherwise leverage a unified, cross-platform application programming interface (“API”) that enables access to all of the different machine learning services needed for full machine learning functionality within the application.”).
Claim 11 is rejected for the similar reason discussed above with respect to claim 1.
Claim 12 is rejected for the similar reason discussed above with respect to claim 2.
Claim 13 is rejected for the similar reason discussed above with respect to claim 3.
Claim 14 is rejected for the similar reason discussed above with respect to claim 4.
Claim 15 is rejected for the similar reason discussed above with respect to claim 5.
Claim 16 is rejected for the similar reason discussed above with respect to claim 6.
Claim 17 is rejected for the similar reason discussed above with respect to claim 7.
Claim 18 is rejected for the similar reason discussed above with respect to claim 10.
With respect to independent claim 19, the Chai teaches a computing system for generating models, comprising: a processor; and memory encoding instructions which, when executed by the processor (see e.g., Para [4] [88] – “The mobile computing device includes one or more processors and one or more non-transitory computer-readable media that collectively store: a computer application; and a machine intelligence software development kit. The machine intelligence software development kit is configured to store one or more machine-learned models and a machine learning library.”), cause the computing system to:
store a library including modules, the modules being programmed to perform tasks of tools of different machine learning models (see e.g., Para [4][74] – “the machine intelligence SDK can further include a dedicated machine learning library that can be implemented by the application to run and/or train the models included in the machine intelligence SDK on-device.”), the library including:
at least one first module configured to filter out and discard portions of input data while the different machine learning models are being built (Chai does not expressly show this feature. However, Chai expressly indicates “developers can have access to a single SDK for all machine learning services” see e.g., Para [34]. Further, Chai discloses data cleaning feature see Para [194]- “Training data creation: The platform can assist developers to augment and manage their training data. Data augmentation can include adding more training samples (e.g., via an image search), leveraging crowd-sourcing platforms to provide labels, and introducing/or transformations of existing samples (e.g., add noises, rotations, perturbations, etc.). The platform can also provide data cleaning and visualization tools.” Therefore, it would have been obvious to include the above feature);
at least one second module configured to convert different sets of input data having data formats into other data formats that can be processed by the different machine learning models while the different machine learning models are being built (Chai does not expressly teach a convert module. However, However, Chai expressly indicates “developers can have access to a single SDK for all machine learning services” see e.g., Para [34]. Furthermore, Chai discloses that the machine leaning framework learns from various input sources and integrates with TensorFlow - see e.g., Para [43]-[45][194]-[198] -” learn from various input sources; cover a wide range of prediction tasks; support a plurality of use-cases; be powerful in terms of learnable machine-learned models and/or architectures supported; be compatible with infrastructure requirements” Therefore, it would have been obvious to include the above feature);
at least one third module configured to identify variables from the different sets of input data that are relevant to determining predicted outcomes by the different machine learning models while the machine learning models are being built (Chai does not expressly teach a variable identification module. However, Chai expressly indicates “developers can have access to a single SDK for all machine learning services … the created schema can include several fields, such as experiment name, features (e.g., name of a field, type of a feature, one or more dimensions of a feature, etc.), hyperparameters (e.g., learning rate, number of steps, optimizer, activation layer, loss weight for a pre-trained model, loss weight for the compact model, cross loss weight, etc.), a model specification of the compact model that contains multiple fields to construct the compact model, a model specification of the pre-trained model that contains multiple fields to construct the pre-trained model.” see e.g., Para [34][58]-[60]. Furthermore, Chai discloses various machine learning architecture and feature generating algorithms - see e.g., Para [118]-[120][235] – ”The model manager 120 can provide various model management services, including, as examples, a model compression service, a model conversion service, a model evaluation service, model hosting/download management services, and/or other model management services including, for example, versioning, compatibility, and/or A/B testing services.” Therefore, it would have been obvious to include the above feature); and
at least one fourth module configured to measure performances of the different machine learning models while the different machine learning models are being built by comparing the predicted outcomes generated by the machine learning models being built to known data (see e.g., Para [63] [77]– “ the training pipeline can include one or more debugging metrics such as various losses, accuracy, confusion matrices for pre-trained machine-learned models and compact machine-learned models. In some implementations, these added metrics are apart from other metrics (e.g., metrics along with Tensorboard integration). In some implementations, the training pipeline can include example implementations of a wrapper for using tf.Estimator API and other plugins. In some implementations, the training pipelines can include integration with TOCO to export the trained compact machine-learned model to TF-Lite format.” “The logging can also enable on-device performance monitoring. For example, in some implementations, the machine intelligence SDK can further perform on-device trained model quality validation (e.g., performance monitoring). These quality statistics can be relayed to the developer via a dashboard offered by the application development platform.”); and
incorporate the modules into the different machine learning models while the different machine learning models are being built, the modules being incorporated into the different machine learning models in response to different calls made via an Application Programming Interface (API), the API being an interface between the library and a computer application configured to receive commands for building the different machine learning models (see e.g., Para [4][34] – “The machine intelligence software development kit is configured to communicate with the computer application using an application programming interface to receive input data from the computer application. The machine intelligence software development kit is configured to implement the one or more machine-learned models and machine learning library on-device to produce an inference based at least in part on the input data. The machine intelligence software development kit is configured to communicate with the computer application using the application programming interface to provide the inference to the computer application.” “the application development platform and SDKs can provide or otherwise leverage a unified, cross-platform application programming interface (“API”) that enables access to all of the different machine learning services needed for full machine learning functionality within the application. In such fashion, developers can have access to a single SDK for all machine learning services. “).
Chai does not expressly show the tools including a feature engineering tool, and the tasks including a task of the feature engineering tool that identifies variables from input data that are relevant to determining predicted outcomes by a machine learning model being built. However, Kennedy teaches similar feature (see e.g. para [39]-[70]– “feature generation module 51 may select one or more features from the list of features and apply a transformation operation, a combination operation, or both to generate new synthetic features… feature selection module 55 selects a specific number of features that have the highest feature scores, to define filtered feature list 77 … Wrapper method module 57 may iteratively select various subsets of features from the list of features provided thereto by feature selection module 55 and train a machine learning algorithm using those selected subsets of features.” feature generation, scoring, selection, and wrapper modules correspond to the “feature engineer tools” and these modules receives an input labeled dataset and, via statistical test and wrapped-based selection, identify and select features/variables that are predictive of the target characteristic, providing them to a machine learning tool for training and prediction.). Both Chai and Kennedy are directed to machine learning model training. Accordingly, it would have been obvious to the skilled artisan before the effective filing date of the claimed invention having Chai and Kennedy in front of them to modify the system of Chai to include the above feature. The motivation to combine Chai and Kennedy comes from Kennedy. Kennedy discloses the motivation to provide tools to analyze input data and identify subset of variables to better predict the outcome (see e.g. para [39]-[70]). This motivation for combination also applies to the remaining claims which depend on this combination.
With respect to dependent claim 20, the modified Chai teaches the API is configured to interface between the library and a plurality of differently configured computer applications each of which is configured to receive commands for building the different machine learning models (see e.g., Para [34][192][208] – “ the present disclosure is directed to an application development platform and associated software development kits (“SDKs”) that provide comprehensive services for generation, deployment, and management of machine-learned models used by computer applications such as, for example, mobile applications executed by a mobile computing device. In particular, the application development platform and SDKs can provide or otherwise leverage a unified, cross-platform application programming interface (“API”) that enables access to all of the different machine learning services needed for full machine learning functionality within the application.”).
Conclusion
Any inquiry concerning this communication or earlier communications from the examiner should be directed to PEIYONG WENG whose telephone number is (571)270-1660. The examiner can normally be reached on Mon.-Fri. 8 am to 5 pm.
If attempts to reach the examiner by telephone are unsuccessful, the examiner's supervisor, Matthew Ell, can be reached on (571) 270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://portal.uspto.gov/external/portal. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free).
/PEI YONG WENG/ Primary Examiner, Art Unit 2141
Prosecution Insights