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 Amendment/Arguments
1. Amendment to claim 20 overcomes the rejection under 35 U.S.C. 112(b).
2. Applicant’s arguments regarding the rejection under 35 U.S.C. 103 filed on February 12, 2026 have been fully considered but are not persuasive. Applicant asserts that amended claim 1 recites a “specific, multi-tier training architecture” involving (i) training a first model using requirement field values and configuration field values spanning multiple subclasses to identify a subclass, and (ii) training separate subclass-specific models using configuration field values restricted to respective subclasses, and argues that Hearley does not teach or suggest such an architecture, but instead only discloses training individual models using data associated with those models.
This argument is not persuasive because it is based on an interpretation of the claims that require a “multi-tier training architecture,” deliberate partitioning of configuration training data by subclass scope, and a specific relationship between models that is not recited. The claims recite training machine learning models using datasets comprising values associated with requirement fields and configuration fields, generating output identifying a subclass data object, and training additional models using values corresponding to applicable configuration fields. The claims do not require a specific architectural arrangement of models, do not require that a first model be trained for the purpose of selecting or identifying a subclass in the manner asserted by Applicant, and do not require deliberate partitioning of configuration training data by subclass scope beyond using values application to a given data object.
With respect to Applicant’s assertion that Hearley does not disclose training a first model using configuration data that spans multiple subclasses, Hearley teaches training machine learning models using datasets defined in tabular attributes (e.g., “Training Data Set”, “Validation Data Set,” and “Test Data Set”), where data is populated via Data ID linkage to associated individual records (Hearley, page 17). Because these datasets include data associated with multiple individual records, training using such datasets corresponds to training using data spanning multiple data records. Under the broadest reasonable interpretation, each such data record corresponds to a subclass data object.
With respect to Applicant’s assertion that Hearley does not disclose training subclass-specific models using configuration data restricted to fields applicable to a respective subclass, Hearley teaches defining model attributes as inputs, outputs, and not used in the model (Hearley, page 16). Because attributes may be designated as not used, a model is trained using only a subset of data values corresponding to applicable attributes. Accordingly, different models may be trained using different subsets of data values.
Thus, contrary to Applicant’s assertions, Hearley teaches (i) training a model using data associated with multiple data records and (ii) training models using subsets of data values defined by applicable attributes. Under BRI, this corresponds to training using configuration field values spanning a plurality of subclass data objects and training models using configuration field values restricted to those applicable to respective subclass data objects. The “multi-tiered training architecture” asserted by Applicant reflects an interpretation that is not required by the claim language, and the limitations are taught by Hearley.
Applicant asserts that Brady does not teach or suggest the claimed training structure, arguing that Brady does not disclose training machine learning models using configuration parameters or organizing configuration fields by subclass applicability.
This argument is not persuasive. Brady is not relied upon for machine learning model training, but rather for teaching configuration field values and requirement-related information associated with technical devices (e.g., device parameters, network settings, permissions, and operational constraints). As set forth in the Office Action, these correspond to the claimed configuration fields and requirement fields.
Hearley teaches training machine learning models using datasets comprising values associated with data records and defined attributes, including subsets of data values for different models based on attribute applicability. As explained above, this corresponds to training models using configuration fields values and using subsets of such values for different models.
Accordingly, Brady need not disclose machine learning model training or subclass-based organization of configuration data. Rather, Brady provides the claimed configuration and requirement field values, while Hearley teaches training machine learning models using such data and applying subsets of data values for different models. Applicant’s argument is therefore not persuasive.
Applicant further asserts that the cited combination would at most suggest using configuration data as input into a single model and would not suggest the claimed structure involving multiple models. This argument is not persuasive.
As discussed above, Hearley teaches the use of multiple machine learning models, including “two different surrogate models,” which are trained using shared datasets (Hearley, page 12). This demonstrates that Hearley is not limited to a single-model approach, but instead teaches training and utilizing multiple models based on available data. Hearley’s disclosure of multiple models trained from shared datasets further demonstrates that different models may be defined and used based on the same underlying data, which is consistent with the claimed use of multiple models trained using overlapping and subset data values. The claims do not require any particular relationship between the models beyond being trained using defined data values, and therefore Hearley’s teaching of multiple models is applicable.
Further, the combination of Hearley and Brady is proper because both references are directed to the use of structured data to support system functionality. Hearley teaches organizing and applying structured datasets to train machine learning models, while Brady teaches the type of configuration and requirement data associated with technical devices. Brady further discloses the use of machine learning techniques in processing such data, demonstrating that Brady operates within systems that utilize data-driven and machine learning-based approaches. Accordingly, Brady provides the claimed configuration field values and requirement-related data, and Hearley teaches training machine learning models using such data and applying subsets of data values for different models. The combination therefore represents the use of known data types in conjunction with known machine learning techniques, and would have been within the level of ordinary skill in the art.
Applicant asserts that the amended claim 1 recites specific machine learning training architecture involving distinct models trained using differently scoped portions of configuration training data for different purposes, and that such an architecture is not taught or suggested by Hearley in view of Brady. Applicant further asserts that independent claims 14 and 18 are patentable for analogous reasons, the dependent claims 2-13, 15-17, and 20 are allowable by virtue of their dependence, and that any additional art does not cure the alleged deficiencies of Hearley and Brady.
These arguments are not persuasive for the reasons set forth above. As explained, the claims do not require the specific “training architecture” asserted by Applicant, but instead recite training machine learning models using datasets comprising values associated with requirement fields and configuration fields, including training using datasets that comprise values associated with a plurality of subclass data objects and using subsets of such values for different models. Hearley teaches training models using datasets populated via Data ID linkage to multiple data records and further teaches defining model inputs based on applicable attributes. Under the broadest reasonable interpretation, the values associated with such data records correspond to values associated with a plurality of subclass data objects, and the use of applicable attributes corresponds to using subsets of such values for different models. Brady provides the claimed configuration field values and requirement-related data.
Accordingly, the combination of Hearley and Brady teaches the limitations of claim 1. Independent claims 14 and 18 recite limitations similar in scope to claim 1 and are rejected for similar reasons. The dependent claims do not recite additional limitations that would render them patentable for the reasons previously set forth. Therefore, the rejections under 35 U.S.C. 103 are maintained.
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.
Claims 1-6, 8-13 are rejected under the 35 U.S.C. 103 as being unpatentable over Hearley et al., (NPL: “A Robust Machine Learning Schema for Developing, Maintaining, and Disseminating Machine Learning Models” (Published: 2022)) in view of Brady et al., (Pub. No.: US 20200192613 A (Filed: 2019)).
Regarding claim 1, Hearley teaches the following limitations:
receiving first user input defining a data object class (Hearley, page 7 “The Granta MI database platform looks to bridge this gap by allowing users to define not only the data present in the database, but also the attributes for a specific subset of data, denoted as tables in Granta, in each data record, thus allowing users to define both material and model information.” Page 9 “Within Granta, users capture and store specific data by assigning different attributes to be populated within a given record, folder, or table…The highest level of organization in Granta is called a table, where all records within a table have the same layout.” – the user-driven definition of the table’s schema (or blueprint) is the functional act of receiving user input defining a data object class.);
receiving second user input defining a plurality of subclass data objects for the data object class (Hearley, page 9, “Within Granta, users capture and store specific data by assigning different attributes to be populated within a given record, folder, or table... Records can further be organized into folders or generic records. Folders simply contain records and have no other associated data, similar to a computer directory tree. Generic records are a combination of a record and a folder—they have the schema attributes that can be populated for the generic record and can also contain different records within.” – Hearley describes the users can define hierarchical data structures in which records (data objects) are contained within folders or generic records that include additional schema attributes. Such nested organization corresponds to defining a plurality of subclass data objects (e.g., records within generic records) for a parent data object class, thereby satisfying the limitation.);
receiving at least seventh user input to initiate training of machine learning models (Hearley, page 31, “Once the data has been uploaded and the different columns defined in the import tool, the next window presents the user with a table with attributes matching those present in the Project Information and General Model Information layout headings…The next screen presents the user with a table that matches the Data Assembly attribute in the Data Definition layout heading, allowing the user to specify the percentage split of training, validation, and test data… Similar to the Data Assembly attribute, the user also specifies the random seed value or equation used to shuffle the data prior to training… Once the training, validation, and test data split have been defined, the importer will then ask the user to supply a .xlsx or .csv file containing the model test predictions…There are also options for the user to supply .xlsx or .csv files for the training and validation curves, as well as a value for accuracy.”) – Hearley teaches that a user provides inputs via a graphical user interface to define data parameters and prepare datasets prior to model training. These user interactions correspond to receiving user input to initiate training of machine learning models.);
receiving eighth user input to deploy at least the first machine learning model; and deploying the first machine learning model in response to receiving the eighth user input (Hearley, page 34 “The keyword “.compile” indicates the end of the model building, and knowledge of the options supply within the importer with the loss function and learning rate. The code digestion feature for a Models: Machine Learning record has been shown to work well for various architectures but is currently limited to Keras models using TensorFlow. As more users adopt the ML schema presented, additional functionality can be added for different ML programming packages to help incentivize users uploading their data and models to the database. All data uploaded through the import tool GUI is saved to local Python libraries with matching attribute names to the above schema. Using the Python SDK provided by Granta MI, records can then be written to the database automatically by matching the attribute names to the locally populated libraries. – Hearley describes that, after the machine learning model has been built and compiled, the user completes the import process through the graphical user interface. This user action causes the trained machine learning model record to be written to the system database via the Python SDK, thereby making the model available for subsequent use. Accordingly, Hearley teaches receiving user input to deploy the machine learning model and deploying the machine learning model in response to the user input.)
However, Hearley does not teach, but Hearley in view of Brady teaches the limitations:
A computing system comprising: at least one hardware processor; at least one memory coupled to the at least one hardware processor; and one or more computer-readable storage media comprising computer-executable instructions that, when executed, cause the computing system to perform operations comprising (Brady, paragraph [0046] “The mobile device 101 includes one or more processors or controllers 101-2. The one or more processors or controllers 101-2 include one or more central processing units (CPUs), which include microprocessors (e.g., a single core microprocessor, a multi-core microprocessor) or other electronic circuitry. The one or more processors 101-2 are configured to read and perform computer-executable instructions, such as instructions that are stored in the read only memory (ROM) 101-20, the main memory 101-18, or the storage 101-22.”):
receiving third user input defining a set of requirements fields for the data object class (Hearley, page 14, “The second layout header, Data, stores the actual data values. The “Data ID” attribute is used as a linking value to enable the viewing of the data in subsequent Model: Machine Learning records. Because data can take on many forms, including single point values, time-series values, or images, each data label that would be used as either an input or output for a surrogate ML model is stored in a generic long text attribute (“Attribute X”) or generic image attribute (“Image X”). The attributes themselves are defined using the “Data Labels” tabular attribute, in which the user enters the attribute or image number, its associated label, and the unit of that label.” – Hearley describes users entering attribute labels, numbers, and units that define how input or output data for the model are represented. The user-defined labels constitute requirements fields that specify the input parameters and data characteristics for the model. Brady paragraph [0027] “Interaction with an image processing device (e.g., print device) may require a user to perform various steps in order to execute a task. This interaction may occur using a display menu of the print device, using driver software, or using a print application. This requires a user to follow multiple steps that make use of technical information and/or obtain administrative approvals to have the job output from the print device.” – Brady describes user interactions that include determining access permissions, compatibility requirements, and feasibility for connecting mobile devices to a print device. These correspond to requirement fields that define conditions and operational constraints preceding configuration.);
receiving fourth user input defining a set of configuration fields for subclass data objects of the plurality of subclass data objects, wherein at least one configuration field applicable to a first subclass data object differs from configuration fields applicable to a second subclass data object (Hearley, page 16 “Because each attribute is viewed from the Reference Data: Machine Learning record, column names for the three data set attributes take on the generic names described above. Therefore, there is an additional attribute in the model record (under this fourth heading), “Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model (see Table V).” Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information (e.g., Active Directory domain, department ID), default print settings (e.g., duplex, color mode), or print policies (e.g., whether user is allowed to print color, duplex).” – Hearley teaches that a user defines attributes for a model record, including specifying variable names and their roles (input/output), which reasonably corresponds to receiving user input defining configuration fields for data objects. Each model record is associated with particular data and parameters, which under BRI reasonably corresponds to subclass data objects. Brady teaches that configuration information is associated with a particular object and includes various parameters (e.g., Wi-Fi settings, IP address, duplex, color mode). Because such parameters depend on the type of object, different objects are associated with different configuration fields. Under BRI, such objects correspond to subclass data objects. Accordingly, at least one configuration field applicable to a first subclass data object differs from configuration fields applicable to a second subclass data object.);
storing a data model comprising the data object class and the set of requirements fields and the plurality of subclass data objects comprising the set of configuration fields (Hearley, page 34 “the tabular attribute that would allow an ML data set to be stored in a single record without sacrificing the flexibility of the existing schema. In the proposed change to the tabular attribute, when linking from one tabular attribute to another, the user would specify both a linking value and a linked row…For the model record, the Training Data, Validation Data, and Test Data attributes would be defined as linked tabular attributes…The Granta MI software would then search for all records in the Data Table that contain the linking value in the “Data ID” attribute (see gray box Figure 23) and would display all values found for the specified column name (i.e., Value 1, Value 2) in the specified linked row.” – Hearley describes storing a structured data model in which data attributes (requirements) are linked to model records (which include configuration information), allowing both types of data to be maintained within the same schema. Brady [0027] “Interaction with an image processing device (e.g., print device) may require a user…to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job” – Brady describes requirement-related information (e.g., access permissions, compatibility, feasibility) that defines operational conditions before configuration. Paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.);
receiving fifth user input identifying at least a first training data set comprising requirements field values (Hearley, page 17 “The “Data Assembly” tabular attribute…is used to define how the data is split into training, validation, and test data. Each column is defined in Table VI….The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record, within the generic record, in the Reference Data: Machine Learning table…giving the user the ability to easily identify the limitations associated with training the model.” – Hearley describes that a user identifies and links data records for use as training data by entering the Data ID attribute, thereby associating the model with a specific training dataset. The linked data include the model inputs (requirement fields values) defined in the Reference Data: Machine Learning table. Brady [0027] “Interaction with an image processing device (e.g., print device) may require a user…to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job” – Brady describes requirement-related information (e.g., access permissions, compatibility, feasibility) that defines operational conditions before configuration.);
receiving sixth user input identifying at least a second training data set comprising configuration field values (Hearley, page 17 “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record, within the generic record, in the Reference Data: Machine Learning table – Hearley teaches that users identify and link datasets (e.g., training, validation, and test datasets) by entering Data ID values, where each dataset contains model inputs and outputs. This corresponds to receiving user input identifying a second training data set. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady teaches configuration information associated with an object, including parameters such as Wi-Fi settings, IP addresses, and print settings. Under BRI, such configuration information reasonably corresponds to configuration field values.);
training a first machine learning algorithm with the at least the at least a first training data set and at least a first portion of the at least a second training data set, the first portion of the at least a second training data set comprising configuration field values associated with the plurality of subclass data objects including configuration fields that are not applicable to at least one of the subclass data objects, to provide a first machine learning model, wherein the first machine learning model is defined across the plurality of subclass data objects and is trained to generate output identifying a subclass data object from the plurality of subclass data objects (Hearley, [page 16] “Because each attribute is viewed from the Reference Data: Machine Learning record, column names for the three data set attributes take on the generic names described above. Therefore, there is an additional attribute in the model record (under this fourth heading), “Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model (see Table V).” [page 31] “The next screen presents the user with a table that matches the Data Assembly attribute in the Data Definition layout heading, allowing the user to specify the percentage split of training, validation, and test data (Figure 20). Similar to the Data Assembly attribute, the user also specifies the random seed value or equation used to shuffle the data prior to training.” [page 33] “The final window in the import tool asks the user to supply the ML code used to train the model (Figure 22(a)). An additional feature built into the import tool is the ability to digest ML code as static text and automatically extract the libraries used and model architecture by reading keywords expected for a given ML package.” – As set forth above, Hearley teaches receiving user input defining requirement fields and configuration fields, and Brady illustrates such configuration fields corresponding to configuration information associated with an object. Hearley further teaches preparing and supplying training datasets for machine learning model training, including specifying training, validation, and test data splits and parameters prior to training, and supplying machine learning code used to train the model. Because Hearley defines attributes as inputs, outputs, or not used, the training data inherently includes attributes not applied to certain data records while being applied to others. Under BRI, such model records correspond to subclass data objects, and the differing use of attributes reasonably corresponds to configuration field values that are not applicable to at least one subclass data object. Healey further teaches training a model using the defined attributes and associated datasets. Because the training uses multiple data records with defined attributes, the trained model produces outputs based on those attributes across the data records. Under BRI, the different data records correspond to different subclass data objects, and the outputs produced based on those attributes correspond to identifying a subclass data object from among a plurality of subclass data objects based on the requirement field values and configuration field values.)
training the first machine learning algorithm or a second machine learning algorithm with the first training data set and respective portions of the second training data set to provide a plurality of second machine learning models, each second machine learning model being defined for and trained with respect to a respective subclass data object of the plurality of subclass data objects and trained using configuration field values corresponding to the configuration fields applicable to that respective subclass data object (Hearley, [page 16] “Because each attribute is viewed from the Reference Data: Machine Learning record, column names for the three data set attributes take on the generic names described above. Therefore, there is an additional attribute in the model record (under this fourth heading), “Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model (see Table V).” [page 18] “In this example, two different ML surrogate models are created…Two different surrogate models were trained from the same data” and “Both the ANN and RNN are trained from the same data set…” [page 31] “The next screen presents the user with a table that matches the Data Assembly attribute in the Data Definition layout heading, allowing the user to specify the percentage split of training, validation, and test data (Figure 20). Similar to the Data Assembly attribute, the user also specifies the random seed value or equation used to shuffle the data prior to training.” – Hearley teaches creating and training a plurality of machine learning models, as evidence by “two different ML surrogate models are created,” including an ANN and RNN. Hearley further teaches that both models are trained from the same dataset. Hearley also teaches partitioning the dataset into training, validation, and test portions, as shown by specifying “the percentage of split training, validation, and test data,” which corresponds to respective portions of the dataset used for training. Hearley further teaches defining attributes for the models through Data Variables Summary, where attributes are designated as inputs, outputs, or not used, thereby defining how data values are used by the models during training. Because the attributes define which data values are used in training a model, the attributes establish the configuration and scope of the data applied to that model. Under BRI, models trained using defined sets of attributes and associated data reasonably correspond to models defined for respective subclass data objects, where each model is trained using the data applicable to that data grouping. Training the models using the defined attributes corresponds to training using configuration field values applicable to the respective subclass data objects. As set forth above, Brady teaches configuration fields and requirement fields for technical products. It would have been obvious to use such configuration and requirements data as attributes in the models of Hearley, thereby applying the modeling framework to configurable technical products.)
Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, having Hearley and Brady before them, to integrate the structured machine learning data management framework of Hearley with the configuration management system of Brady. One would have been motivated to make such a combination in order to create a unified system capable of capturing user or operational requirements and using them to define, train, and apply models that generate or select appropriate configuration parameters for complex product or systems. This combination leverages Hearley’s schema-based data traceability and model-training infrastructure to organize and process requirement information, together with Brady’s explicit examples of configuration and policy data, to enable a system that streamlines setup and ensures that resulting configurations satisfy defined operational needs.
Regarding claim 2, Hearley in view of Brady, as outlined above, all the elements of claim 1, therefore is rejected for the same reasons as those presented for claim 1, mutatis mutandis. Hearley in view of Brady further teaches:
creating the at least a first training data set by selecting from a data set values for configuration fields of the set of configuration fields (Hearley, page 30, “The next screen asks the user to upload the file containing all data used to train, validate, and test the model. The importer assumes that each row is a given example in the data set and that the first row in the file contains the data labels… By default, all parameters are assumed to be inputs in the final column, so the user must also change any rows that are outputs, or predictions, from the model in the GUI using the drop-down menu in each cell.” – Hearley describes that the user uploads a data file and interacts with the importer GUI to identify which parameters (fields) are to be treated as inputs or outputs for model training. This selection process constitutes creating a training data set by selecting from the upload data values corresponding to particular configuration fields. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.).
Regarding claim 3, Hearley in view of Brady, as outlined above, all the elements of claim 1, therefore is rejected for the same reasons as those presented for claim 1, mutatis mutandis. Hearley in view of Brady further teaches:
receiving nineth user input selecting a subclass data object to provide a selected subclass data object (Hearley, page 9, “Within Granta, users capture and store specific data by assigning different attributes to be populated within a given record, folder, or table… Records can further be organized into folders or generic records. Folders simply contain records and have no other associated data, similar to a computer directory tree. Generic records are a combination of a record and a folder—they have the schema attributes that can be populated for the generic record and can also contain different records within.” - Hearley explains that the Granta MI interface allows users to navigate and select among multiple records or generic records organized under a higher-level data structure. The act of a user selecting one of these records or generic records corresponds to receiving user input selecting a subclass data object to provide a selected subclass data object.);
retrieving configuration fields defined for the selected subclass data object (Hearley, page 16 “Because each attribute is viewed from the Reference Data: Machine Learning record, column names for the three data set attributes take on the generic names described above. Therefore, there is an additional attribute in the model record (under this fourth heading), “Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model (see Table V). – Hearley describes that once a particular model record is selected, the system automatically accesses and displays the configuration-related attributes (e.g., inputs, outputs, units) associated with that record. This corresponds to retrieving the configuration fields defined for the selected subclass data object. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.); and
displaying the configuration fields defined for the selected subclass data object to a user (Hearley, page 11 “Along with viewing direct links to different records in Granta, users can view data in another record within the same database, even if the two records are in different tables, through a tabular attribute type… When viewing data in a tabular attribute, records are linked together by the “Linking Value,”… As a result, when viewing the record on the front end, or the window the user sees when viewing the populated record (Figure 4(c)), the attributes in Record 1 are automatically pulled and displayed in the viewed record. The linked tabular attribute in Granta MI allows the user to view data from multiple records in one location” – Hearley describes that when a record (i.e., subclass data object) is selected, the linked configuration attributes are automatically pulled and displayed to the user through the graphical user interface. This corresponds to displaying the configurations fields defined for the selected subclass data object. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.);
wherein the fourth user input comprises one or more of the configuration fields defined for the selected subclass data object (Hearley, page 16 “Because each attribute is viewed from the Reference Data: Machine Learning record, column names for the three data set attributes take on the generic names described above. Therefore, there is an additional attribute in the model record (under this fourth heading), “Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model” – Hearley explains that users define and populate configuration-related parameters (e.g., inputs, outputs, usage status) for each model record. Thus, the fourth user input (defining configuration fields) directly corresponds to the editable configuration attributes associated with the selected subclass data object. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.).
Regarding claim 4, Hearley in view of Brady, as outlined above, all the elements of claim 3, therefore is rejected for the same reasons as those presented for claim 3, mutatis mutandis. Hearley in view of Brady further teaches:
creating the at least a first training data set at least in part by selecting from a data set values for the one or more configuration fields defined for the selected subclass data object (Hearley, page 17 “The “Data Assembly” tabular attribute (in Figure 6 under the Data Definition header) is used to define how the data is split into training, validation, and test data. Each column is defined in Table VI… The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories… Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record, within the generic record, in the Reference Data: Machine Learning table…giving the user the ability to easily identify the limitations associated with training the model.” – Hearley teaches that users select and link particular data records using the “Data ID” to populate training data sets with model parameters and attributes, corresponding to selecting values for the configuration fields defined for the selected subclass data object. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.).
Regarding claim 5, Hearley in view of Brady, as outlined above, all the elements of claim 1, therefore is rejected for the same reasons as those presented for claim 1, mutatis mutandis. Hearley in view of Brady further teaches:
creating the at least a first training data set at least in part by selecting from a data set values for one or more of the requirements fields defined for the data object class (Hearley, page 14 “The “Data ID” attribute is used as a linking value to enable the viewing of the data in subsequent Model: Machine Learning records. Because data can take on many forms, including single point values, time-series values, or images, each data label that would be used as either an input or output for a surrogate ML model is stored in a generic long text attribute (“Attribute X”) or generic image attribute (“Image X”). The attributes themselves are defined using the “Data Labels” tabular attribute, in which the user enters the attribute or image number, its associated label, and the unit of that label.” – Hearley teaches that the user defines and selects specific data attributes, labels, and units that correspond to model inputs. These represent requirements field values that describe the conditions or input parameters for model training. Brady [0027] “Interaction with an image processing device (e.g., print device) may require a user…to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job” – Brady describes requirement-related information (e.g., access permissions, compatibility, feasibility) that defines operational conditions before configuration.).
Regarding claim 6, Hearley in view of Brady, as outlined above, all the elements of claim 5, therefore is rejected for the same reasons as those presented for claim 5, mutatis mutandis. Hearley in view of Brady further teaches:
extracting the data set values by parsing unstructured data (Hearley, page 33-34 “The final window in the import tool asks the user to supply the ML code used to train the model (Figure 22(a))… All data uploaded through the import tool GUI is saved to local Python libraries with matching attribute names to the above schema. Using the Python SDK provided by Granta MI, records can then be written to the database automatically by matching the attribute names to the locally populated libraries. The import tool gives users the means to upload their large data sets to the database in a time-efficient manner, creating the individual data records necessary for best database management practices that would otherwise be infeasible to do manually. It also provides some additional utility features, such as the automatic code digestion and ability to save and import programmatic information, in an attempt to reduce the burden on the user and increase the likelihood of user adoption.” – Hearley teaches that the import tool automatically processes and converts heterogeneous data – including programmatic files and text-based inputs – into structured schema-compliant records. This inherently involves parsing unstructured data (such as .csv, .xlsx, or text-based ML code) to extract and populate the structured data set values used in model training.).
Regarding claim 8, Hearley in view of Brady, as outlined above, all the elements of claim 1, therefore is rejected for the same reasons as those presented for claim 1, mutatis mutandis. Hearley in view of Brady further teaches:
displaying model performance results for the first machine learning model and the second machine learning model (Hearley, page 12 “In the first (left) example, a ML model is trained from a combination of virtual (simulation) and real (experimental) data. The separation of data and model parameters allows data for a model to come from various sources albeit real or virtual, which enables greater flexibility when defining a model in Granta MI without repeating any preexisting data in the database. In the second (right) example, the same virtual data set is used to define two different surrogate models… Not only does this help maintain the traceability of each model, but it also reduces the storage space necessary for capturing the full model pedigree of the two surrogates…” Page 17, “The fifth layout heading, Model Validation, allows for the definition of the model performance. Because predictions are specific to each model, the values in the tabular attributes “Validation Predictions” and “Test Predictions” are defined in the model record and are not linked to another table in the database.” – Hearley teaches that the system supports multiple trained models (two surrogate models) and maintains distinct validation and test prediction results for each model record. These results are stored and displayed within the model’s interface to evaluate and compare performance. Accordingly, Hearley discloses displaying model performance results for the first and second machine-learning models, each corresponding to a separate subclass data object.).
Regarding claim 9, Hearley in view of Brady, as outlined above, all the elements of claim 1, therefore is rejected for the same reasons as those presented for claim 1, mutatis mutandis. Hearley in view of Brady further teaches:
wherein, for training a respective second machine learning model, the respective port of at least a second training data set corresponds to configuration fields of a single subclass data object (Hearley, page 17 “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories. Each tabular attribute has the same column structure, as defined in Table VII. Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record, within the generic record, in the Reference Data: Machine Learning table, and each row represents a different example in that set. Additionally, the user can enter whether the data in that specific row was virtual or experimental in the “Data Source Type” column, giving the user the ability to easily identify the limitations associated with training the model.” – Hearley teaches that training data for a model is defined using tabular attributes (e.g., Training Data Set, Validation Data Set, Test Data Set) who corresponds to a specific data record identified by a Data ID and includes associated input and output values for that record. Because each row represents a specific data record, the training data used for the model includes portions corresponding to individual data records. Under BRI, each such data record corresponds to a subclass data object, and the associated output values correspond to configuration field values. Accordingly, the respective portion of the training data set used for training a model corresponds to configuration fields of a single subclass data object. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.).
Regarding claim 10, Hearley in view of Brady, as outlined above, all the elements of claim 9, therefore is rejected for the same reasons as those presented for claim 9, mutatis mutandis. Hearley in view of Brady further teaches:
training the first machine learning algorithm with at the at least a first portion of the at least a second training data set, the at least a first training data set, and at least a portion of a third training data set comprising configuration field values for configuration fields defined by the single subclass data object to provide a second machine learning model (Hearley, [page 12] “Figure 5 shows two different example cases that illustrate this division. In the first (left) example, a ML model is trained from a combination of virtual (simulation) and real (experimental) data. The separation of data and model parameters allows data for a model to come from various sources albeit real or virtual, which enables greater flexibility when defining a model in Granta MI without repeating any preexisting data in the database.” [Page 17] “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record…and each row represents a different example in that set.” – Hearley teaches training a machine learning model using multiple datasets, including different types of data (e.g., real and simulated), as shown by training “from a combination of virtual…and real…data.” Hearley further teaches that the training data is organized through tabular attributes (e.g., Training Data Set, Validation Data Set, Test Data Set), where each row corresponds to a specific data record identified by a Data ID and includes associated input and output values. Because training data is linked through Data IDs to individual data records, the training of a model uses data corresponding to specific data records. Under BRI, each such data record corresponds to a single subclass data object. Accordingly, the portion of the dataset used for training corresponds to a respective portion of data for the single subclass data object. Hearley also teaches that the training data includes input and output values associated with a model. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.).
Regarding claim 11, Hearley in view of Brady, as outlined above, all the elements of claim 1, therefore is rejected for the same reasons as those presented for claim 1, mutatis mutandis. Hearley in view of Brady further teaches:
training a second machine learning algorithm with the at least a first training data set and portions of the at least a second training data set corresponding to at least a plurality of configuration fields of the set of configuration fields for the plurality of subclass data objects, wherein the second machine learning algorithm is the first machine learning algorithm or is a machine learning algorithm other than the first machine learning algorithm (Hearley, page 12, “Figure 5 shows two different example cases…In the second (right) example, the same virtual data set is used to define two different surrogate models. In this case, the separation of data and model parameters prevents the virtual data from being repeated…help maintain the traceability of each model…” Page 17 “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…” – Hearley describes the multiple surrogate models can be trained from overlapping and distinct data sets, with linked training, validation, and test data defining each model’s respective inputs and outputs. Thus, Hearley teaches training a second ML algorithm – either the same or a different one – using the first training data and additional portions of the second training data set that correspond to configuration-related parameters across subclass data objects. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.).
Regarding claim 12, Hearley in view of Brady, as outlined above, all the elements of claim 1, therefore is rejected for the same reasons as those presented for claim 1, mutatis mutandis. Hearley in view of Brady further teaches:
receiving nineth user input selecting at least one configurable data object for a subclass data object of the plurality of subclass data objects (Brady paragraph [0027] “ Interaction with an image processing device (e.g., print device) may require a user to perform various steps in order to execute a task. This interaction may occur using a display menu of the print device, using driver software, or using a print application. This requires a user to follow multiple steps that make use of technical information and/or obtain administrative approvals to have the job output from the print device. For example, in case where a user is using a print application and a mobile device, a user may need to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job” Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object. Hearley, page 9 “Within Granta, users capture and store specific data by assigning different attributes to be populated within a given record, folder, or table… Records can further be organized into folders or generic records. Folders simply contain records and have no other associated data, similar to a computer directory tree. Generic records are a combination of a record and a folder—they have the schema attributes that can be populated for the generic record and can also contain different records within.” – Hearley supports that the system enables user-driven selection of records or linked data objects within a database schema. When combined with Brady, Hearley’s schema provides the structure for managing subclass data objects, while Brady provides the interactive selection of configurable objects and their associated configuration fields.)
retrieving at least a portion of fields defined for the at least one configurable data object (Brady paragraph [0017] “the print application determines the associated device ID such as model number, and recognizes the configuration information associated with print device in the image frame captured. Once the print application determines the configuration information associated with the print device captured in the image frame of the mobile device. The print application in block B809 performs the configuration information of the print device to the mobile device and initiates the communication connection between the print device and the mobile device.” – Brady describes that, upon recognizing the device, the application retrieves configuration information (e.g., network credentials, access settings, policies) associated with that identified object from storage. This corresponds to retrieving at least a portion of fields defined for a selected configuration data object.)and
displaying the at least a portion of fields defined for the at least one configurable data object to a user (Hearley, page 11 “When viewing data in a tabular attribute, records are linked together by the “Linking Value,” which is the value of a specified attribute that Granta MI uses to populate the viewed table…when viewing the record on the front end… the attributes in Record 1 are automatically pulled and displayed in the viewed record. The linked tabular attribute in Granta MI allows the user to view data from multiple records in one location… giving users the ability to view large data sets in one record with guaranteed traceability.” – Hearley describes automatically displaying linked attribute data to the user through graphical user interface, thereby showing the retrieved fields of a selected record or data object. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.);
wherein the fourth user input comprises a selection of one or more fields of the at least a portion of fields defined for the at least one configurable data object (Hearley, page 16 “Because each attribute is viewed from the Reference Data: Machine Learning record, column names for the three data set attributes take on the generic names described above. Therefore, there is an additional attribute in the model record (under this fourth heading), “Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model” – Hearley supports that users actively define and select data attributes (fields) for use within a model record. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.).
Regarding claim 13, Hearley in view of Brady, as outlined above, all the elements of claim 1, therefore is rejected for the same reasons as those presented for claim 1, mutatis mutandis. Hearley in view of Brady further teaches:
wherein a first plurality of configurable data objects are assigned to a first subclass data object of the plurality of subclass data objects and a second plurality of configurable data objects are assigned to a second subclass data object of the plurality of subclass data objects, a first subset of the set of configuration fields is defined for the first subclass data object and a second subset of the set of configuration fields is defined for the second subclass data object, where at least one element of the first subset differs from elements defined for the second subset (Hearley, page 12 “The separation of the data and model parameters into two tables allows one to follow the best practice that data only be defined in one location within the database. Figure 5 shows two different example cases that illustrate this division. In the first (left) example, a ML model is trained from a combination of virtual (simulation) and real (experimental) data. The separation of data and model parameters allows data for a model to come from various sources albeit real or virtual, which enables greater flexibility when defining a model in Granta MI without repeating any preexisting data in the database. In the second (right) example, the same virtual data set is used to define two different surrogate models. In this case, the separation of data and model parameters prevents the virtual data from being repeated in the database in multiple locations. Not only does this help maintain the traceability of each model, but it also reduces the storage space necessary for capturing the full model pedigree of the two surrogates,” – Hearley describes a schema in which multiple surrogate models (each representing a subclass data object) are separately defined within the same data structure. Each model record is linked to its own associated training and validation data (configurable data objects), along with specific parameters and architecture attributes. Accordingly, Hearley teaches that individual subclass data objects maintain distinct sets of data linkages and model parameters, which collectively form separate subsets of configuration-related fields. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.).
Claims 14-20 are rejected under the 35 U.S.C. 103 as being unpatentable over Hearley et al., (NPL: “A Robust Machine Learning Schema for Developing, Maintaining, and Disseminating Machine Learning Models” (Published: 2022)) in view of Brady et al., (Pub. No.: US 20200192613 A (Filed: 2019)) further in view of Kuhn et al., (Pub. No.: US 20220366474 A1 (Filed: 2022)).
Regarding claim 14, Hearley teaches the following limitations:
receiving first user input defining a data object class (Hearley, page 7 “The Granta MI database platform looks to bridge this gap by allowing users to define not only the data present in the database, but also the attributes for a specific subset of data, denoted as tables in Granta, in each data record, thus allowing users to define both material and model information.” Page 9 “Within Granta, users capture and store specific data by assigning different attributes to be populated within a given record, folder, or table…The highest level of organization in Granta is called a table, where all records within a table have the same layout.” – the user-driven definition of the table’s schema (or blueprint) is the functional act of receiving user input defining a data object class.);
receiving second user input defining a plurality of subclass data objects for the data object class (Hearley, page 9, “Within Granta, users capture and store specific data by assigning different attributes to be populated within a given record, folder, or table... Records can further be organized into folders or generic records. Folders simply contain records and have no other associated data, similar to a computer directory tree. Generic records are a combination of a record and a folder—they have the schema attributes that can be populated for the generic record and can also contain different records within.” – Hearley describes the users can define hierarchical data structures in which records (data objects) are contained within folders or generic records that include additional schema attributes. Such nested organization corresponds to defining a plurality of subclass data objects (e.g., records within generic records) for a parent data object class, thereby satisfying the limitation.);
However, Hearley does not teach, but Hearley in view of Brady teaches the following limitations:
A method, implemented in a computing system comprising at least one hardware processor and at least one memory coupled to the at least one hardware processor, the method comprising(Brady, paragraph [0046] “The mobile device 101 includes one or more processors or controllers 101-2. The one or more processors or controllers 101-2 include one or more central processing units (CPUs), which include microprocessors (e.g., a single core microprocessor, a multi-core microprocessor) or other electronic circuitry. The one or more processors 101-2 are configured to read and perform computer-executable instructions, such as instructions that are stored in the read only memory (ROM) 101-20, the main memory 101-18, or the storage 101-22.”):
receiving third user input defining a set of requirements fields for the data object class (Hearley, page 14, “The second layout header, Data, stores the actual data values. The “Data ID” attribute is used as a linking value to enable the viewing of the data in subsequent Model: Machine Learning records. Because data can take on many forms, including single point values, time-series values, or images, each data label that would be used as either an input or output for a surrogate ML model is stored in a generic long text attribute (“Attribute X”) or generic image attribute (“Image X”). The attributes themselves are defined using the “Data Labels” tabular attribute, in which the user enters the attribute or image number, its associated label, and the unit of that label.” – Hearley describes users entering attribute labels, numbers, and units that define how input or output data for the model are represented. The user-defined labels constitute requirements fields that specify the input parameters and data characteristics for the model. Brady paragraph [0027] “Interaction with an image processing device (e.g., print device) may require a user to perform various steps in order to execute a task. This interaction may occur using a display menu of the print device, using driver software, or using a print application. This requires a user to follow multiple steps that make use of technical information and/or obtain administrative approvals to have the job output from the print device.” – Brady describes user interactions that include determining access permissions, compatibility requirements, and feasibility for connecting mobile devices to a print device. These correspond to requirement fields that define conditions and operational constraints preceding configuration.);
receiving fourth user input defining a set of configuration fields for subclass data objects of the plurality of subclass data objects (Hearley, page 16 “Because each attribute is viewed from the Reference Data: Machine Learning record, column names for the three data set attributes take on the generic names described above. Therefore, there is an additional attribute in the model record (under this fourth heading), “Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model (see Table V).” – Hearley describes that, for each model record, users define the Data Variables Summary attribute, which specifies variable names, their corresponding roles (input/output), and associated units. This activity represents user input establishing configuration fields – the model’s parameterization – within each subclass data object (i.e., each model instance) under the overarching data object class. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information (e.g., Active Directory domain, department ID), default print settings (e.g., duplex, color mode), or print policies (e.g., whether user is allowed to print color, duplex).” – Hearley provides the mechanism for defining the fields, and Brady provides the context of what those fields present in a configured product – i.e., specific parameters that describe how the product is set up for operation.);
storing a data model comprising the data object class and the set of requirements fields and the plurality of subclass data objects comprising the set of configuration fields (Hearley, page 34 “the tabular attribute that would allow an ML data set to be stored in a single record without sacrificing the flexibility of the existing schema. In the proposed change to the tabular attribute, when linking from one tabular attribute to another, the user would specify both a linking value and a linked row…For the model record, the Training Data, Validation Data, and Test Data attributes would be defined as linked tabular attributes…The Granta MI software would then search for all records in the Data Table that contain the linking value in the “Data ID” attribute (see gray box Figure 23) and would display all values found for the specified column name (i.e., Value 1, Value 2) in the specified linked row.” – Hearley describes storing a structured data model in which data attributes (requirements) are linked to model records (which include configuration information), allowing both types of data to be maintained within the same schema. Brady [0027] “Interaction with an image processing device (e.g., print device) may require a user…to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job” – Brady describes requirement-related information (e.g., access permissions, compatibility, feasibility) that defines operational conditions before configuration. Paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.);
However, Hearley in view of Brady does teach, but Hearley in view of Brady further in view of Kuhn teaches the limitation:
based at least in part on the data model, providing a response to a recommendation request by using a first machine learning model trained using values for the set of requirements fields and values for configuration fields associated with a plurality of subclass data objects to generate output identifying a subclass data object of the plurality of subclass data objects (Kuhn, paragraph [0078] “Recommendation engine 402, in some examples, can include an online processing sub-engine 428 that generates real-time subscription product recommendations in response to requests received “ Hearley, [page 16] ““Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model (see Table V).” [Page 17] “Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record, within the generic record, in the Reference Data: Machine Learning table, and each row represents a different example in that set.” [Page 18] “In this example, two different ML surrogate models are created…Both the ANN and RNN are trained from the same data set” Brady [0027] “ Interaction with an image processing device (e.g., print device) may require a user to perform various steps in order to execute a task…a user may need to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job, and determine the configuration information of the identified print device to connect with the user's mobile device.” [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information (e.g., Active Directory domain, department ID), default print settings (e.g., duplex, color mode), or print policies (e.g., whether user is allowed to print color, duplex).” – Hearley teaches defining machine learning models using attributes, where attributes are designated as inputs, outputs, or not used in the model. Brady teaches that such attributes correspond to requirement fields (e.g., access, compatibility) and configurations fields (e.g., device parameters and settings). Accordingly, Hearley in view of Brady teaches a first machine learning model trained using values for requirement fields and configuration fields. Hearley further teaches that data used by the models is liked through Data IDs to associated individual data records. Because the model is trained using data associated with multiple such data records, the model operates across a plurality of data records. Under BRI, each data corresponds to a subclass data object. Because the training uses multiple data records with defined attributes, the trained model produces outputs corresponding to those data records. Under BRI, outputs corresponding to those records reasonably correspond to those data records identifying a subclass data object from among a plurality of subclass data objects. Accordingly, the reference teaches using a first machine learning model trained using values for requirements fields and configuration fields associated with a plurality of subclass data objects as part of providing a response to a recommendation request based at least in part on a data model.); and
using a second machine learning model defined for the identified subclass data object and trained using values for configuration fields applicable to the identified subclass data object to generate the response to the recommendation request (Hearley, [page 16] ““Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model” [page 18] “In this example, two different ML surrogate models are created…Both the ANN and RNN are trained from the same data set” [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information (e.g., Active Directory domain, department ID), default print settings (e.g., duplex, color mode), or print policies (e.g., whether user is allowed to print color, duplex).” – Hearley teaches a plurality of machine learning models, as shown by the creation of “two different ML surrogate models,” including an ANN and an RNN. Hearley further teaches defining model attributes through the “Data Variable Summary,” where attributes are designated inputs, outputs, or not used in the model. Because attributes may be designated as not used, each model is defined using only the attributes applicable to that model. Brady teaches configuration fields for technical objects, including parameters such as network settings and device-specific configurations. Accordingly, the attributes used by the models correspond to configuration field values. Because model attributes determine which data are used by a model, a model defined using attributes associated with a particular data record is defined with respect to that data record. Under the broadest reasonable interpretation, each such data record corresponds to a subclass data object. Accordingly, a model defined using attributes applicable to a particular data record corresponds to a second machine learning model defined for the identified subclass data object and trained using values for configuration fields applicable to that subclass data object. Using such a model to generate output corresponds to generating the response to the recommendation request.).
Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, having a combination of Hearley, Brady, and Kuhn before them, to incorporate the recommendation engine of Kuhn into a structured data and configuration framework of Hearley and Brady. One would have been motivated to make such a combination in order to provide intelligent response to recommendation requests based on machine learning models trained using the defined requirement and configuration parameters of a product or system. This integration would enable a user-driven platform capable of producing automated, data-informed recommendations that reflect both operational constraints and technical configurations, thereby improving the precision and adaptability of model-based recommendations.
Regarding claim 15, Hearley in view of Brady further in view of Kuhn, as outlined above, all the elements of claim 14, therefore is rejected for the same reasons as those presented for claim 14, mutatis mutandis. Hearley in view of Brady further teaches:
receiving fifth user input identifying at least a first training data set comprising requirements field values (Hearley, page 17 “The “Data Assembly” tabular attribute…is used to define how the data is split into training, validation, and test data. Each column is defined in Table VI….The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record, within the generic record, in the Reference Data: Machine Learning table…giving the user the ability to easily identify the limitations associated with training the model.” – Hearley describes that a user identifies and links data records for use as training data by entering the Data ID attribute, thereby associating the model with a specific training dataset. The linked data include the model inputs (requirement fields values) defined in the Reference Data: Machine Learning table. Brady [0027] “Interaction with an image processing device (e.g., print device) may require a user…to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job” – Brady describes requirement-related information (e.g., access permissions, compatibility, feasibility) that defines operational conditions before configuration.);
receiving sixth user input identifying at least a second training data set comprising configuration fields values (Hearley, page 17 “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record, within the generic record, in the Reference Data: Machine Learning table” – Hearley explains that users link multiple datasets - specifically, the Training Data Set, Validation Data Set, and Test Data Set – by entering distinct Data ID values. The Validation Set and Test Data Set serve as additional data sets associated with the model, each containing both input and output (configuration) information used to train or validate the model. This corresponds to receiving user input identifying a second training data set comprising configuration field values. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object.);
Regarding claim 16, Hearley in view of Brady further in view of Kuhn, as outlined above, all the elements of claim 15, therefore is rejected for the same reasons as those presented for claim 15, mutatis mutandis. Hearley in view of Brady further teaches:
receiving at least seventh user input to train a machine learning model using at least the at least a first training data set and the at least a second training data set (Hearley, page 31, “Once the data has been uploaded and the different columns defined in the import tool, the next window presents the user with a table with attributes matching those present in the Project Information and General Model Information layout headings…The next screen presents the user with a table that matches the Data Assembly attribute in the Data Definition layout heading, allowing the user to specify the percentage split of training, validation, and test data… Similar to the Data Assembly attribute, the user also specifies the random seed value or equation used to shuffle the data prior to training… Once the training, validation, and test data split have been defined, the importer will then ask the user to supply a .xlsx or .csv file containing the model test predictions…There are also options for the user to supply .xlsx or .csv files for the training and validation curves, as well as a value for accuracy.”) – Hearley describes a graphical user interface through which the user provides a series of inputs to define data splits, set randomization parameters, and initiate model training using the identified training, validation, and test datasets. These user-defined actions collectively constitute receiving user input to train a machine learning model using both the first training data set (requirements/input values) and the second training data set (configuration/output values);
training a first machine learning algorithm with the at least the at least a first training data set and at least a first portion of the at least a second training data set to provide a first machine learning model trained to generate output identifying a subclass data object (Hearley, [page 17] “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record” [page 31] “The next screen presents the user with a table that matches the Data Assembly attribute in the Data Definition layout heading, allowing the user to specify the percentage split of training, validation, and test data” [page 33] “The final window in the import tool asks the user to supply the ML code used to train the model… An additional feature built into the import tool is the ability to digest ML code as static text and automatically extract the libraries used and model architecture by reading keywords expected for a given ML package… the import tool supports reading Python codes that use Keras with TensorFlow for ML… By first finding “.Sequential,” the code knows to look for “.add” for a new addition to the model… the code can recognize layers, hidden units, and activation functions for the model.” – Hearley teaches training a machine learning algorithm using defined datasets, including training, validation, and test data. Hearley further teaches that these datasets are populated using data associated with individual data records via Data ID linkage. Because the datasets are defined and partitioned into training, validation, and test portions, the training process uses at least a first training data set and at least a portion of another dataset, corresponding to the claimed use of a first training data set and a first portion of a second training data set. Hearley also teaches supplying machine learning code (e.g., Keras/TensorFlow) to train the model, thereby defining the first machine learning algorithm to produce the trained model. Because the training uses data associated with multiple individual records, the trained model produces outputs corresponding to those data records. Under the broadest reasonable interpretation, each data record corresponds to a subclass data object. Accordingly, a model trained on such data is trained to generate output identifying a subclass data object.);
receiving eighth user input to deploy the first machine learning model; and deploying the first machine learning model in response to receiving the eighth user input (Hearley, page 34 “The keyword “.compile” indicates the end of the model building, and knowledge of the options supply within the importer with the loss function and learning rate. The code digestion feature for a Models: Machine Learning record has been shown to work well for various architectures but is currently limited to Keras models using TensorFlow. As more users adopt the ML schema presented, additional functionality can be added for different ML programming packages to help incentivize users uploading their data and models to the database. All data uploaded through the import tool GUI is saved to local Python libraries with matching attribute names to the above schema. Using the Python SDK provided by Granta MI, records can then be written to the database automatically by matching the attribute names to the locally populated libraries. – Hearley describes that once the ML model has been built and compiled, the user completes the import process through the GUI (eight user input). This final action initiates automatic writing of the trained machine learning model record into the Granta MI via the Python SDK, thereby making the model available for subsequent organizational use. Accordingly, Hearley teaches receiving user input to deploy the ML model and deploying the trained ML model in response.).
Regarding claim 17, Hearley in view of Brady further in view of Kuhn, as outlined above, all the elements of claim 16, therefore is rejected for the same reasons as those presented for claim 16, mutatis mutandis. Hearley in view of Brady further teaches:
training a second machine learning algorithm with the at least a first training data set and portions of the at least a second training data set corresponding to configuration field values corresponding to configuration fields application to a respective subclass data object of the plurality of subclass data objects, wherein the second machine learning algorithm is the first machine learning algorithm or is a machine learning algorithm other than the first machine learning algorithm (Hearley, page 12, “Figure 5 shows two different example cases…In the second (right) example, the same virtual data set is used to define two different surrogate models.” Page 17 “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record,” Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Hearley teaches training multiple machine learning models, including “two different surrogate models,” which may be defined using the same dataset. This teaches training a second machine learning algorithm that may be the same or different from a first machine learning algorithm. Hearley further teaches that model training uses defined datasets, including training, validation, and test datasets, where data is linked via Data ID to individual records. Because datasets include input and outputs associated with these individual records, the training uses data corresponding to multiple records and portions thereof. Brady teaches configuration field values associated with technical objects. Accordingly, the data values associated with attributes used in model training corresponds to configuration field values. Because the datasets are composed of data linked to individual data records, and portions of those datasets are used in training, the training uses portions of data corresponding to configuration field values applicable to particular data records. Under BRI, each such data record corresponds to a subclass data object.).
Regarding claim 18, Hearley teaches the following limitations:
One or more non-transitory computer-readable storage media comprising: computer-executable instructions that, when executed by a computing system comprising at least one hardware processor and at least one memory coupled to the at least one hardware processor, cause the computing system to receive first user input defining a data object class (Hearley, page 7 “The Granta MI database platform looks to bridge this gap by allowing users to define not only the data present in the database, but also the attributes for a specific subset of data, denoted as tables in Granta, in each data record, thus allowing users to define both material and model information.” Page 9 “Within Granta, users capture and store specific data by assigning different attributes to be populated within a given record, folder, or table…The highest level of organization in Granta is called a table, where all records within a table have the same layout.” – Hearley describes user interaction with Granta MI software to define data structures (records, folders, or tables) and assign attributes for storage. These user actions correspond to receiving a first user input defining a data object class (i.e., the data schema used to organize stored data). Because the described operations are carried out within a computerized data-management platform, it is inherent that they are performed by computer-executable instructions stored on non-transitory computer-readable media and executed by a computer system comprising at least one processor or memory to receive and process the user input.)
computer-executable instructions that, when executed by the computing system, cause the computing system to receive second user input defining a plurality of subclass data objects for the data object class (Hearley, page 9, “Within Granta, users capture and store specific data by assigning different attributes to be populated within a given record, folder, or table... Records can further be organized into folders or generic records. Folders simply contain records and have no other associated data, similar to a computer directory tree. Generic records are a combination of a record and a folder—they have the schema attributes that can be populated for the generic record and can also contain different records within.” – Hearley describes user input creating hierarchical data structures in which users generate records (e.g., model or data entries) within broader schema tables or organize them into higher-level generic records. This corresponds to defining multiple subclass data objects for a parent data object class. Because these relationships are created and stored within the Grant MI environment, it is inherent that they are executed by computer instructions on a computing system.);
However, Hearley does not teach, but Hearley in view of Brady teaches the following limitations:
computer-executable instructions that, when executed by the computing system, cause the computing system to receive third user input defining a set of requirements fields for the data object class (Hearley, page 14, “The second layout header, Data, stores the actual data values. The “Data ID” attribute is used as a linking value to enable the viewing of the data in subsequent Model: Machine Learning records. Because data can take on many forms, including single point values, time-series values, or images, each data label that would be used as either an input or output for a surrogate ML model is stored in a generic long text attribute (“Attribute X”) or generic image attribute (“Image X”). The attributes themselves are defined using the “Data Labels” tabular attribute, in which the user enters the attribute or image number, its associated label, and the unit of that label.” – Hearley describes users entering attribute labels, numbers, and units that define how input or output data for the model are represented. The user-defined labels constitute requirements fields that specify the input parameters and data characteristics for the model. Brady paragraph [0027] “Interaction with an image processing device (e.g., print device) may require a user to perform various steps in order to execute a task. This interaction may occur using a display menu of the print device, using driver software, or using a print application. This requires a user to follow multiple steps that make use of technical information and/or obtain administrative approvals to have the job output from the print device.” – Brady likewise teaches defining prerequisite information such as access and compatibility conditions, corresponding to requirement fields that establish operational constraints. Because both sets of activities occur within computerized systems, it is inherent that these steps are executed by computer instructions on a computing system.);
computer-executable instructions that, when executed by the computing system, cause the computing system to receive fourth user input defining a set of configuration fields for subclass data objects of the plurality of subclass data objects (Hearley, page 16 “Because each attribute is viewed from the Reference Data: Machine Learning record, column names for the three data set attributes take on the generic names described above. Therefore, there is an additional attribute in the model record (under this fourth heading), “Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model (see Table V).” – Hearley describes that, for each model record, users define the Data Variables Summary attribute, which specifies variable names, their corresponding roles (input/output), and associated units. This activity represents user input establishing configuration fields – the model’s parameterization – within each subclass data object (i.e., each model instance) under the overarching data object class. Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information (e.g., Active Directory domain, department ID), default print settings (e.g., duplex, color mode), or print policies (e.g., whether user is allowed to print color, duplex).” – Hearley provides the mechanism for defining the fields, and Brady provides the context of what those fields present in a configured product – i.e., specific parameters that describe how the product is set up for operation. Together, Hearley describes the computerized definition of configuration fields, while Brady illustrates the substantive parameters being configured. Because these configurations occur within computerized systems, it is inherent that they are executed by computer instructions on a computing system.);
computer-executable instructions that, when executed by the computing system, cause the computing system to store a data model comprising the data object class and the set of requirements fields and the plurality of subclass data objects comprising the set of configuration fields (Hearley, page 34 “the tabular attribute that would allow an ML data set to be stored in a single record without sacrificing the flexibility of the existing schema. In the proposed change to the tabular attribute, when linking from one tabular attribute to another, the user would specify both a linking value and a linked row…For the model record, the Training Data, Validation Data, and Test Data attributes would be defined as linked tabular attributes…The Granta MI software would then search for all records in the Data Table that contain the linking value in the “Data ID” attribute (see gray box Figure 23) and would display all values found for the specified column name (i.e., Value 1, Value 2) in the specified linked row.” – Hearley describes storing a structured data model in which data attributes (requirements) are linked to model records (which include configuration information), allowing both types of data to be maintained within the same schema. Brady [0027] “Interaction with an image processing device (e.g., print device) may require a user…to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job” – Brady describes requirement-related information (e.g., access permissions, compatibility, feasibility) that defines operational conditions before configuration. Paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Brady further describes the configuration information associated with each object. Together, Hearley supplies the computerized schema and storage mechanism, while Brady supplies the content being stored. Because these operations occur within computerized systems, it is inherent that they are executed by computer-executable instructions on a processor with memory.);
However, Hearley in view of Brady does not teach, but Hearley in view of Brady further in view of Kuhn teaches the limitation:
computer-executable instructions that, when executed by the computing system, cause the computing system to, based at least in part on the data model, provide a response to a recommendation request by: using a first machine learning model trained using values for the set of requirements fields and values for configuration fields associated with a plurality of subclass data objects to generate output identifying a subclass data object of the plurality of subclass data objects (Kuhn, paragraph [0078] “Recommendation engine 402, in some examples, can include an online processing sub-engine 428 that generates real-time subscription product recommendations in response to requests received “ Hearley, [page 16] ““Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model (see Table V).” [Page 17] “Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record, within the generic record, in the Reference Data: Machine Learning table, and each row represents a different example in that set.” [Page 18] “In this example, two different ML surrogate models are created…Both the ANN and RNN are trained from the same data set” Brady [0027] “ Interaction with an image processing device (e.g., print device) may require a user to perform various steps in order to execute a task…a user may need to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job, and determine the configuration information of the identified print device to connect with the user's mobile device.” [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information (e.g., Active Directory domain, department ID), default print settings (e.g., duplex, color mode), or print policies (e.g., whether user is allowed to print color, duplex).” – Hearley teaches defining machine learning models using attributes, where attributes are designated as inputs, outputs, or not used in the model. Brady teaches that such attributes correspond to requirement fields (e.g., access, compatibility) and configurations fields (e.g., device parameters and settings). Accordingly, Hearley in view of Brady teaches a first machine learning model trained using values for requirement fields and configuration fields. Hearley further teaches that data used by the models is liked through Data IDs to associated individual data records. Because the model is trained using data associated with multiple such data records, the model operates across a plurality of data records. Under BRI, each data corresponds to a subclass data object. Because the training uses multiple data records with defined attributes, the trained model produces outputs corresponding to those data records. Under BRI, outputs corresponding to those records reasonably correspond to those data records identifying a subclass data object from among a plurality of subclass data objects. Accordingly, the reference teaches using a first machine learning model trained using values for requirements fields and configuration fields associated with a plurality of subclass data objects as part of providing a response to a recommendation request based at least in part on a data model.); and
using a second machine learning model defined for the identified subclass data object and trained using values for configuration fields applicable to the identified subclass data object to generate the response to the recommendation request (Hearley, [page 16] ““Data Variables Summary,” to allow for the definition of each of the attribute names and whether they are inputs, outputs, or not used in the model” [page 18] “In this example, two different ML surrogate models are created…Both the ANN and RNN are trained from the same data set” [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information (e.g., Active Directory domain, department ID), default print settings (e.g., duplex, color mode), or print policies (e.g., whether user is allowed to print color, duplex).” – Hearley teaches a plurality of machine learning models, as shown by the creation of “two different ML surrogate models,” including an ANN and an RNN. Hearley further teaches defining model attributes through the “Data Variable Summary,” where attributes are designated inputs, outputs, or not used in the model. Because attributes may be designated as not used, each model is defined using only the attributes applicable to that model. Brady teaches configuration fields for technical objects, including parameters such as network settings and device-specific configurations. Accordingly, the attributes used by the models correspond to configuration field values. Because model attributes determine which data are used by a model, a model defined using attributes associated with a particular data record is defined with respect to that data record. Under the broadest reasonable interpretation, each such data record corresponds to a subclass data object. Accordingly, a model defined using attributes applicable to a particular data record corresponds to a second machine learning model defined for the identified subclass data object and trained using values for configuration fields applicable to that subclass data object. Using such a model to generate output corresponds to generating the response to the recommendation request.).
Accordingly, it would have been obvious to a person of ordinary skill in the art, before the effective filing date of the claimed invention, having a combination of Hearley, Brady, and Kuhn before them, to incorporate the recommendation engine of Kuhn into a structured data and configuration framework of Hearley and Brady. One would have been motivated to make such a combination in order to provide intelligent response to recommendation requests based on machine learning models trained using the defined requirement and configuration parameters of a product or system. This integration would enable a user-driven platform capable of producing automated, data-informed recommendations that reflect both operational constraints and technical configurations, thereby improving the precision and adaptability of model-based recommendations.
Regarding claim 19, Hearley in view of Brady further in view of Kuhn, as outlined above, all the elements of claim 18, therefore is rejected for the same reasons as those presented for claim 18, mutatis mutandis. Hearley in view of Brady further teaches:
computer-executable instructions that, when executed by the computing system, cause the computing system to train a first machine learning algorithm with at least a first training data set comprising the values for the set of requirements fields and at least a second training data set comprising the values for the set of configuration fields associated with a plurality of subclass data objects to provide the first machine learning model (Hearley, [page 12] “In the first (left) example, a ML model is trained from a combination of virtual (simulation) and real (experimental) data. The separation of data and model parameters allows data for a model to come from various sources” [page 17] “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record” Brady, paragraph [0027] “Interaction with an image processing device (e.g., print device) may require a user…to identify the print device, determine whether the user is allowed to access the identified print device, determine whether the identified print device is compatible with the required print job” Paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Hearley teaches training a machine learning model using multiple datasets, including a combination of different types of data (e.g., simulation and experimental data), thereby teaching training using at least a first training dataset and at least a second training dataset. Hearley further teaches that datasets used in training include inputs and outputs defined in tabular attributes, where data is linked via Data ID to associated individual data records. Because these datasets include values associated with different data attributes, the datasets comprise values used for training the model. Brady teaches requirement-related information (e.g., access permissions, compatibility) and configuration-related information (e.g., device parameters and settings). Accordingly, the values in the datasets correspond to values for requirement fields and configuration fields. Because the datasets are composed of values associated with individual data records, and training is performed using those datasets, the training uses values for requirement fields and values for configuration fields associated with a plurality of individual records. Under the broadest reasonable interpretation, each such data record corresponds to a subclass data object.).
Regarding claim 20, Hearley in view of Brady further in view of Kuhn, as outlined above, all the elements of claim 18, therefore is rejected for the same reasons as those presented for claim 18, mutatis mutandis. Hearley in view of Brady further teaches:
training a second machine learning algorithm with the at least a first training data set and portions of the at least a second training data set corresponding to configuration fields corresponding to configuration fields applicable to a respective subclass data object of the plurality of subclass data object, wherein the second machine learning algorithm is the at least one machine learning algorithm or is a machine learning algorithm other than the at least one machine learning algorithm (Hearley, page 12, “Figure 5 shows two different example cases…In the second (right) example, the same virtual data set is used to define two different surrogate models.” Page 17 “The remaining three tabular attributes (“Training Data Set,” “Validation Data Set,” “Test Data Set”) contain the specific model inputs and outputs for the three respective categories…Each linked column in the three tabular attributes is automatically populated by entering the “Data ID” from the associated individual data record,” Brady paragraph [0036] “Further configuration information concerning the object may include, but is not limited to, a Wi-Fi connection, a local network or internet connection over LAN or WAN, a Wi-Fi network name, encryption type, any passwords, print device IP address or domain name, login setting information…default print settings…or print policies” – Hearley teaches training multiple machine learning models, including “two different surrogate models,” which may be defined using the same dataset. This teaches training a second machine learning algorithm that may be the same or different from the at least one machine learning algorithm. Hearley further teaches that model training uses defined datasets, including training, validation, and test datasets, where data is linked via Data ID to individual records. Because datasets include input and outputs associated with these individual records, the training uses data corresponding to multiple records and portions thereof. Brady teaches configuration field values associated with technical objects. Accordingly, the data values associated with attributes used in model training corresponds to configuration field values. Because the datasets are composed of data linked to individual data records, and portions of those datasets are used in training, the training uses portions of data corresponding to configuration field values applicable to particular data records. Under BRI, each such data record corresponds to a subclass data object.).
Conclusion
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 Daravanh Phakousonh whose telephone number is (571)272-6324. The examiner can normally be reached Mon - Thurs 7 AM - 5 PM, Every other Friday 7 AM - 4PM.
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, Li B Zhen can be reached at 571-272-3768. 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.
/Daravanh Phakousonh/Examiner, Art Unit 2121
/Li B. Zhen/Supervisory Patent Examiner, Art Unit 2121
Prosecution Insights