DETAILED ACTION
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This Office Action is in response to claims filed on 12/30/2025.
Claims 1, 3-19, and 21-22 are pending.
Claim Rejections - 35 USC § 103
The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
Claims 1-10, and 12-14 are rejected under 35 U.S.C. 103 as being unpatentable over Sivakumar (US 2024/0223675 A1) in view of Goldsteen (US 2024/0160965 A1) in view of Arbelo-Gonzalez (US 2024/0281981 A1) in view of Todasco (US 2023/0036623 A1).
With regard to claim 1, Sivakumar teaches:
A method, comprising: performing, by a data materialization platform, a data migration process between a core cloud environment and an edge cloud environment; “The data storage engine 250 of the public cloud (core cloud) 105 sends the fingerprint calculator algorithm 222A (which is stored as fingerprint calculator algorithm 222B) and the map 220A (which is stored as the map 220B) to the edge device 202 such that the traffic management code 150 can use fingerprint calculator algorithm 222B to calculate the deduplication fingerprints for all the participating blocks in the edge cache 210, for eventual comparison with the stored fingerprints in the map 220B (e.g., B-tree)” [Sivakumar ¶ 52].
identifying, by the data materialization platform and during the data migration process, attribute values stored in a data repository of the core cloud environment, “When the data storage engine 250 is running additional deduplication on the incoming data, on every incoming data request, the data storage engine 250 calculates the fingerprints (attributes) of the incoming data using a fingerprint algorithm 222A, checks for a match to the fingerprints of the available track records in a map 220A, and decides to execute the actual write to the disk of the customer data volume 256 based on whether the fingerprints of the incoming data blocks match the stored fingerprints in the map 220A” [Sivakumar ¶ 40, Fig. 2 Examiner notes the customer data volume 256 within public cloud 105]. “This may be accomplished by performing the identification of deduplication fingerprints at the edge device level, by identifying data blocks/chunks that are saved in the cache of the edge device but not saved at the target device (e.g., the cloud). Accordingly, the edge device selectively sends the packets of data chunks, which are not saved at the target device, from the edge device to the target device at the cloud in order to avoid any unnecessary data transfer to the target device” [Sivakumar ¶ 11].
…that is to be executed by the edge cloud environment; “Edge computing upends the traditional architecture by shifting key processing functions away from the core of the network and out to the edge where users are located. Through a combination of edge data centers and IoT devices that can process data for themselves, edge computing can improve network performance and reduce latency” [Sivakumar ¶ 2].
analyzing, by the data materialization platform, the attribute values to identify (duplicates) ranges, associated with a subset of the attribute values, “This may be accomplished by performing the identification of deduplication fingerprints (attributes) at the edge device level, by identifying data blocks/chunks that are saved in the cache of the edge device but not saved at the target device (e.g., the cloud)” [Sivakumar ¶ 11]. “Data deduplication is a technique for eliminating duplicate copies of repeating data. The deduplication process requires comparison of data "chunks" (also known as "byte patterns") which are unique, contiguous blocks of data. These chunks are identified and stored during a process of analysis and compared to other chunks within existing data. Whenever a match occurs, the redundant chunk is replaced with a small reference that points to the stored chunk. Given that the same byte pattern may occur dozens, hundreds, or even thousands of times (the match frequency is dependent on the chunk size), the amount of data that is stored or transferred can be greatly reduced” [Sivakumar ¶ 15].
deduplicating, by the data materialization platform, the subset of the attribute values from the data repository of the core cloud environment to generate deduplicated attribute values “When the traffic management code 150 of the edge device 202 is communicating with a data storage engine 250 of the public cloud 105 and issuing operations (e.g., READ/WRITE operations) on the customer data volume 256 (which can be issued via a data receiver (e.g., IBM Spectrum Virtualize™ for public cloud (SVPC) in the gateway 140), the data storage engine 250 offers many data optimization features, such as, for example, data deduplication and others … When the data storage engine 250 is running additional deduplication on the incoming data, on every incoming data request, the data storage engine 250 calculates the fingerprints (attributes) of the incoming data using a fingerprint algorithm 222A, checks for a match to the fingerprints of the available track records in a map 220A, and decides to execute the actual write to the disk of the customer data volume 256 based on whether the fingerprints of the incoming data blocks match the stored fingerprints in the map 220A” [Sivakumar ¶ 40].
wherein the deduplicating is performed prior to the migrating to reduce a volume of the training data transmitted to and stored in the edge cloud environment, “This may be accomplished by performing the identification of deduplication fingerprints at the edge device level, by identifying data blocks/chunks that are saved in the cache of the edge device but not saved at the target device (e.g., the cloud). Accordingly, the edge device selectively sends the packets of data chunks, which are not saved at the target device, from the edge device to the target device at the cloud in order to avoid any unnecessary data transfer to the target device” [Sivakumar ¶ 11].
Sivakumar fails to teach wherein the attribute values are to be used as inputs for a machine learning model that is to be executed by the edge cloud environment; analyzing, by the data materialization platform, the attribute values to identify ranges, associated with a subset of the attribute values, for which outputs of the machine learning model are estimated to be a same output within a tolerance; that are associated with median range values; adding metadata to the deduplicated attribute values … wherein the deduplicated attribute values are used as training data to train the machine learning model in the edge cloud environment.
However, Goldsteen teaches:
wherein the attribute values are to be used as inputs for a machine learning model “An embodiment includes generating feature explainability data associated with a feature (attribute), wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Predictive modeling refers to a form of predictive analytics that typically uses a machine learning (ML) algorithm to build a predictive model (also referred to herein as a ML model or ML predictive model)” [Goldsteen ¶ 17].
that is to be executed by the edge cloud environment; “Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers” [Goldsteen ¶ 159].
analyzing, by the data materialization platform, the attribute values to identify ranges, associated with a subset of the attribute values, “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy. In some embodiments, a generalized version of numerical feature data includes a list of ranges, and a generalized version of categorical feature data is one or more groups of values” [Goldsteen ¶ 20].
for which outputs of the machine learning model are estimated to be a same output within a tolerance; “As an example, data records that provide for feature data that is representative of people's ages may be used by a predictive ML model to predict heart attacks with about 90% accuracy. Data minimization that involves feature generalization may include replacing certain ages with an age range that still allows the ML model to maintain an accuracy of about 90%” [Goldsteen ¶ 21].
that are associated with median range values; “In some embodiments, where the generalization is a numerical range, the representative value is a median value of the numerical range (e.g., the first representative value for the 1-20 range is 10.5)” [Goldsteen ¶ 129].
adding metadata to the deduplicated attribute values “An embodiment includes generating feature explainability data associated with a feature (attribute), wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20].
wherein the deduplicated attribute values are used as training data to train the machine learning model in the edge cloud environment, “An embodiment includes generating feature explainability data associated with a feature (attribute), wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Predictive modeling refers to a form of predictive analytics that typically uses a machine learning (ML) algorithm to build a predictive model (also referred to herein as a ML model or ML predictive model)” [Goldsteen ¶ 17]. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20]. “For example, in some embodiments, the ML model is a classification model trained to make class predictions from tabular input data, which may include features having numerical, categorical, and/or continuous data domains” [Goldsteen ¶ 70]. “The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers” [Goldsteen ¶ 159].
Goldsteen is considered to be analogous to the claimed invention because it is in the same field of machine learning. The method of range calculation for machine learning inputs of Goldsteen can be combined with the core cloud and edge could system of Sivakumar such that the fingerprints (attributes) of Sivakumar can be deduplicated using the method of Goldsteen. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar to incorporate the teachings of Goldsteen and include that the attribute values are to be used as inputs for a machine learning model that is to be executed by the edge cloud environment; analyzing, by the data materialization platform, the attribute values to identify ranges, associated with a subset of the attribute values, for which outputs of the machine learning model are estimated to be a same output within a tolerance; that are associated with median range values; adding metadata to the deduplicated attribute values … wherein the deduplicated attribute values are used as training data to train the machine learning model in the edge cloud environment. Doing so would allow for the use of the system to make predictions within a certain level of accuracy. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20].
Sivakumar in view of Goldsteen fails to teach adding metadata to the deduplicated attribute values to indicate a quantity of data records associated with each of the median range values; and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model.
However, Arbelo-Gonzalez teaches:
adding metadata to (machine learning training data) the deduplicated attribute values to indicate a quantity of data records associated with each of the (classes) median range values; “wherein training the machine learning model further comprises determining the first set of weights based on a set of frequencies associated with the set of classes within the set of n-dimensional images” [Arbelo-Gonzalez Claim 5]. “The class weight for a given class could be computed by dividing the number of occurrences of the most frequent class in training 3D images 240 (or a given set of training 3D patches 258 extracted from training 3D images) by the number of occurrences of the given class in training 3D images 240 (or the set of training 3D patches 258). Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model. “More specifically, the loss function includes different weights for different classes of objects with which training data for the machine learning model is labeled” [Arbelo-Gonzalez ¶ 17]. “The class weight for a given class could be computed by dividing the number of occurrences of the most frequent class in training 3D images 240 (or a given set of training 3D patches 258 extracted from training 3D images) by the number of occurrences of the given class in training 3D images 240 (or the set of training 3D patches 258). Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
Arbelo-Gonzalez is considered to be analogous to the claimed invention because it is in the same field of machine learning. The metadata of the quantity of data records of Arbelo-Gonzalez can be combined with the deduplicated attribute values of Sivakumar in view of Goldsteen such that this metadata can be used to determine weights for the data records as taught by the method of Arbelo-Gonzalez. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen to incorporate the teachings of Arbelo-Gonzalez and include adding metadata to the deduplicated attribute values to indicate a quantity of data records associated with each of the median range values; and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model. Doing so would allow for bias correction within the machine learning model training process. “Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez fails to explicitly teach and migrating the deduplicated attribute values from the core cloud environment to the edge cloud environment, … data transmitted to and stored in the edge cloud environment.
and migrating the deduplicated attribute values from the core cloud environment to the edge cloud environment, “At step 422, data for the user is accessed from the edge computing storage. The data may be migrated or moved to the edge computing storage from a central storage, such as a cloud computing storage utilized by the user. In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
data transmitted to and stored in the edge cloud environment, “At step 422, data for the user is accessed from the edge computing storage. The data may be migrated or moved to the edge computing storage from a central storage, such as a cloud computing storage utilized by the user. In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
Todasco is considered to be analogous to the claimed invention because it is in the same field of indexing schemes relating to priority. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez to incorporate the teachings of Todasco and include and migrating the deduplicated attribute values from the core cloud environment to the edge cloud environment, … data transmitted to and stored in the edge cloud environment. Doing so would allow for useful data to be more quickly accessed. “In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
With regard to claim 3, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the method of claim 1, as referenced above. Sivakumar further teaches further comprising: storing the deduplicated attribute values on the edge cloud environment. “The data storage engine 250 of the public cloud 105 sends the fingerprint calculator algorithm 222A (which is stored as fingerprint calculator algorithm 222B) and the map (deduplicated attribute values) 220A (which is stored as the map 220B) to the edge device 202 such that the traffic management code 150 can use fingerprint calculator algorithm 222B to calculate the deduplication fingerprints for all the participating blocks in the edge cache 210, for eventual comparison with the stored fingerprints in the map 220B (e.g., B-tree)” [Sivakumar ¶ 52, fig. 2 Examiner notes map 220B where the deduplicated attribute values are stored is located at edge device 202].
With regard to claim 4, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the method of claim 1, as referenced above. Sivakumar fails to teach modifying the attribute values based on generated rules to generate modified attribute values.
However, Goldsteen teaches:
further comprising: modifying the attribute values “The generalization process replaces the feature (or features) associated with the selected slice with an alternative feature (or alternative features) that is a generalization of the selected feature. As an example, the selected feature may be representative of age based on the generalization process determining that the age feature has a relatively low feature importance value and the generalization process identifying groups of ages that are good generalization candidates. The generalization process then replaces the age feature, which includes any integer from 1 to 120 as a feature value, with an alternative feature representative of age ranges, for example [1-20], [21-25], [26-50], [51-120].” [Goldsteen ¶ 35 Examiner notes the replacement of features with alternative features is in accordance with the descriptions of modifying the attribute values in paragraphs 54 and 60 of the instant specification].
based on generated rules to generate modified attribute values. “The embodiment also includes extracting feature value data from the input data, where the feature value data is representative of a feature value of the feature for the sample. The embodiment also includes constructing a generalization group comprising the feature of the sample, where the constructing of the generalization group comprises detecting that the feature value of the feature and the explainability value of the feature satisfy a predetermined condition (rule)” [Goldsteen ¶ 3]. “In an exemplary embodiment, groups of samples (referred to as slices) are identified that satisfy certain conditions (rules) related to feature values and local explainability values, and therefore can be generalized together. In various embodiments, the conditions may be based on distances between feature/explainability values, ranges of feature/ explainability values, variance of feature/explainability values, size of the slice, etc.” [Goldsteen ¶ 29].
With regard to claim 5, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the method of claim 4, as referenced above. Sivakumar further teaches further comprising: migrating the modified attribute values to the data repository of the edge cloud environment. “When the data storage engine 250 is running additional deduplication on the incoming data, on every incoming data request, the data storage engine 250 calculates the fingerprints of the incoming data using a fingerprint algorithm 222A, checks for a match to the fingerprints of the available track records in a map 220A, and decides to execute the actual write to the disk of the customer data volume 256 based on whether the fingerprints of the incoming data blocks match the stored fingerprints (deduplicated attribute values) in the map 220A” [Sivakumar ¶ 40 Examiner notes the duplicated attribute values are considered the modified attribute values]. “The data storage engine 250 of the public cloud 105 sends the fingerprint calculator algorithm 222A (which is stored as fingerprint calculator algorithm 222B) and the map (modified attribute values) 220A (which is stored as the map 220B) to the edge device 202 such that the traffic management code 150 can use fingerprint calculator algorithm 222B to calculate the deduplication fingerprints for all the participating blocks in the edge cache 210, for eventual comparison with the stored fingerprints in the map 220B (e.g., B-tree)” [Sivakumar ¶ 52].
With regard to claim 6, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the method of claim 1, as referenced above. Sivakumar further teaches:
further comprising: validating one or more operations associated with identifying the attribute values; “When deduplication is enabled for the customer data volumes 256, the data storage engine 250 opens the communication tunnel 208 with the edge device 202 in preparation for the edge device to validate the deduplication fingerprints of the customer data volume 256 against the blocks of data in the edge cache 210” [Sivakumar ¶ 51].
and providing, based on validating the one or more operations, the one or more operations to (storage) the machine learning model via an edge data migrator. “At block 318, in response to receiving the message (e.g., I/O WRITE) from the traffic management code 150, the data storage engine (edge data migrator) 250 already knows that the these are new fingerprints that were pre-validated by the edge device 202, and accordingly, the data storage engine 250 bypasses the fingerprint calculation on incoming I/O requests of the blocks of data 272. The data storage engine 250 writes the data to the new location in the customer data volume 256, sends the acknowledgement (ACK) to the edge device 202, and marks (WRITE) complete” [Sivakumar ¶ 54].
Sivakumar fails to teach and providing … the one or more operations to the machine learning model.
However, Goldsteen teaches:
and providing … the one or more operations to the machine learning model “The data preparation 502 performs data preparation on the incoming data as necessary to prepare the data for processing by the machine learning model 504. The data preparation 502 then provides the prepared data to the machine learning model 504” [Goldsteen ¶ 110-111].
With regard to claim 7, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the method of claim 1, as referenced above. Sivakumar fails to teach receiving an information request associated with the edge cloud environment; obtaining, based on receiving the information request, information associated with one or more pre-derived boundary values from one or more machine learning models that include the machine learning model; and providing the information associated with the one or more pre-derived boundary values to satisfy the information request.
However, Goldsteen teaches:
further comprising: receiving an information request associated with the edge cloud environment; “In some embodiments, the dynamic display data provides instructions to update the user interface to replace a user input field configured to request or receive the original feature values with an alternative input field configured to request or receive the alternative generalized feature values” [Goldsteen ¶ 41].
obtaining, based on receiving the information request, information associated with one or more pre-derived boundary values from one or more machine learning models that include the machine learning model; “For example, an age feature, which includes any integer from 1 to 120 as a feature value, may be generalized to an alternative feature representative of age ranges (boundary values), for example [1-20], [21-25], [26-50], [51-120]. In this example, the dynamic display data instructs the user interface to replace a user input field that requests an age with a user input field that requests selection of one of the age ranges. The user interface will then output a representative value that will be used as input to the ML runtime module for any value in the selected range” [Goldsteen ¶ 41].
and providing the information associated with the one or more pre-derived boundary values to satisfy the information request. “In this example, the dynamic display data instructs the user interface to replace a user input field that requests an age with a user input field that requests selection of one of the age ranges. The user interface will then output a representative value that will be used as input to the ML runtime module for any value in the selected range” [Goldsteen ¶ 41].
With regard to claim 8, Sivakumar teaches:
A data materialization platform, comprising: one or more memories;
“These computer readable program instructions are stored in various types of computer readable storage media, such as cache 121 and the other storage media discussed below” [Sivakumar ¶ 22].
and one or more processors, communicatively coupled to the one or more memories, configured to: “Computer readable program instructions are typically loaded onto computer 101 to cause a series of operational steps to be performed by processor set 110 of computer 101 and thereby effect a computer-implemented method, such that the instructions thus executed will instantiate the methods specified in flowcharts and/or narrative descriptions of computer-implemented methods included in this document (collectively referred to as "the inventive methods")” [Sivakumar ¶ 22].
perform a data migration process between a core cloud environment and an edge cloud environment; “The data storage engine 250 of the public cloud (core cloud) 105 sends the fingerprint calculator algorithm 222A (which is stored as fingerprint calculator algorithm 222B) and the map 220A (which is stored as the map 220B) to the edge device 202 such that the traffic management code 150 can use fingerprint calculator algorithm 222B to calculate the deduplication fingerprints for all the participating blocks in the edge cache 210, for eventual comparison with the stored fingerprints in the map 220B (e.g., B-tree)” [Sivakumar ¶ 52].
identify, during the data migration process, attribute values stored in a data repository of the core cloud environment, “When the data storage engine 250 is running additional deduplication on the incoming data, on every incoming data request, the data storage engine 250 calculates the fingerprints (attributes) of the incoming data using a fingerprint algorithm 222A, checks for a match to the fingerprints of the available track records in a map 220A, and decides to execute the actual write to the disk of the customer data volume 256 based on whether the fingerprints of the incoming data blocks match the stored fingerprints in the map 220A” [Sivakumar ¶ 40, Fig. 2 Examiner notes the customer data volume 256 within public cloud 105]. “This may be accomplished by performing the identification of deduplication fingerprints at the edge device level, by identifying data blocks/chunks that are saved in the cache of the edge device but not saved at the target device (e.g., the cloud). Accordingly, the edge device selectively sends the packets of data chunks, which are not saved at the target device, from the edge device to the target device at the cloud in order to avoid any unnecessary data transfer to the target device” [Sivakumar ¶ 11].
…that is to be executed by the edge cloud environment; “Edge computing upends the traditional architecture by shifting key processing functions away from the core of the network and out to the edge where users are located. Through a combination of edge data centers and IoT devices that can process data for themselves, edge computing can improve network performance and reduce latency” [Sivakumar ¶ 2].
analyze the attribute values to identify a subset of the attribute values “This may be accomplished by performing the identification of deduplication fingerprints (attributes) at the edge device level, by identifying data blocks/chunks that are saved in the cache of the edge device but not saved at the target device (e.g., the cloud)” [Sivakumar ¶ 11]. “Data deduplication is a technique for eliminating duplicate copies of repeating data. The deduplication process requires comparison of data "chunks" (also known as "byte patterns") which are unique, contiguous blocks of data. These chunks are identified and stored during a process of analysis and compared to other chunks within existing data. Whenever a match occurs, the redundant chunk is replaced with a small reference that points to the stored chunk. Given that the same byte pattern may occur dozens, hundreds, or even thousands of times (the match frequency is dependent on the chunk size), the amount of data that is stored or transferred can be greatly reduced” [Sivakumar ¶ 15].
deduplicate the subset of the attribute values to generate deduplicated attribute values “When the traffic management code 150 of the edge device 202 is communicating with a data storage engine 250 of the public cloud 105 and issuing operations (e.g., READ/WRITE operations) on the customer data volume 256 (which can be issued via a data receiver (e.g., IBM Spectrum Virtualize™ for public cloud (SVPC) in the gateway 140), the data storage engine 250 offers many data optimization features, such as, for example, data deduplication and others … When the data storage engine 250 is running additional deduplication on the incoming data, on every incoming data request, the data storage engine 250 calculates the fingerprints (attributes) of the incoming data using a fingerprint algorithm 222A, checks for a match to the fingerprints of the available track records in a map 220A, and decides to execute the actual write to the disk of the customer data volume 256 based on whether the fingerprints of the incoming data blocks match the stored fingerprints in the map 220A” [Sivakumar ¶ 40].
wherein deduplicating is performed prior to migrating to reduce a volume of the training data transmitted to and stored in the edge cloud environment, “This may be accomplished by performing the identification of deduplication fingerprints at the edge device level, by identifying data blocks/chunks that are saved in the cache of the edge device but not saved at the target device (e.g., the cloud). Accordingly, the edge device selectively sends the packets of data chunks, which are not saved at the target device, from the edge device to the target device at the cloud in order to avoid any unnecessary data transfer to the target device” [Sivakumar ¶ 11].
Sivakumar fails to teach wherein the attribute values are to be used as training data to train for a machine learning model that is to be executed by the edge cloud environment; analyze the attribute values to identify a subset of the attribute values for which outputs of the machine learning model are estimated to be a same output within a tolerance; deduplicate the subset of the attribute values to generate deduplicated attribute values that are associated with median range values; add metadata to the deduplicated attribute values … wherein the deduplicated attribute values are used as the training data to train the machine learning model in the edge cloud environment.
However, Goldsteen teaches:
wherein the attribute values are to be used as training data to train for a machine learning model “An embodiment includes generating feature explainability data associated with a feature (attribute), wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Predictive modeling refers to a form of predictive analytics that typically uses a machine learning (ML) algorithm to build a predictive model (also referred to herein as a ML model or ML predictive model)” [Goldsteen ¶ 17]. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20]. “For example, in some embodiments, the ML model is a classification model trained to make class predictions from tabular input data, which may include features having numerical, categorical, and/or continuous data domains” [Goldsteen ¶ 70].
that is to be executed by the edge cloud environment; “Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers” [Goldsteen ¶ 159].
analyze the attribute values to identify a subset of the attribute values for which outputs of the machine learning model are estimated to be a same output within a tolerance; “As an example, data records that provide for feature data that is representative of people's ages may be used by a predictive ML model to predict heart attacks with about 90% accuracy. Data minimization that involves feature generalization may include replacing certain ages with an age range that still allows the ML model to maintain an accuracy of about 90%” [Goldsteen ¶ 21].
deduplicate the subset of the attribute values to generate deduplicated attribute values that are associated with median range values; “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy. In some embodiments, a generalized version of numerical feature data includes a list of ranges, and a generalized version of categorical feature data is one or more groups of values” [Goldsteen ¶ 20]. “In some embodiments, where the generalization is a numerical range, the representative value is a median value of the numerical range (e.g., the first representative value for the 1-20 range is 10.5)” [Goldsteen ¶ 129].
add metadata to the deduplicated attribute values “An embodiment includes generating feature explainability data associated with a feature (attribute), wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20].
wherein the deduplicated attribute values are used as the training data to train the machine learning model in the edge cloud environment, “An embodiment includes generating feature explainability data associated with a feature (attribute), wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Predictive modeling refers to a form of predictive analytics that typically uses a machine learning (ML) algorithm to build a predictive model (also referred to herein as a ML model or ML predictive model)” [Goldsteen ¶ 17]. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20]. “For example, in some embodiments, the ML model is a classification model trained to make class predictions from tabular input data, which may include features having numerical, categorical, and/or continuous data domains” [Goldsteen ¶ 70]. “The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers” [Goldsteen ¶ 159].
The method of range calculation for machine learning inputs of Goldsteen can be combined with the core cloud and edge could system of Sivakumar such that the fingerprints (attributes) of Sivakumar can be deduplicated using the method of Goldsteen. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar to incorporate the teachings of Goldsteen and include that wherein the attribute values are to be used as training data to train for a machine learning model that is to be executed by the edge cloud environment; analyze the attribute values to identify a subset of the attribute values for which outputs of the machine learning model are estimated to be a same output within a tolerance; deduplicate the subset of the attribute values to generate deduplicated attribute values that are associated with median range values; add metadata to the deduplicated attribute values … wherein the deduplicated attribute values are used as the training data to train the machine learning model in the edge cloud environment. Doing so would allow for the use of the system to make predictions within a certain level of accuracy. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20].
Sivakumar in view of Goldsteen fails to teach add metadata to the deduplicated attribute values to indicate a quantity of data records associated with each of the median range values; and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model.
However, Arbelo-Gonzalez teaches:
add metadata to (machine learning training data) the deduplicated attribute values to indicate a quantity of data records associated with each of the (classes) median range values; “wherein training the machine learning model further comprises determining the first set of weights based on a set of frequencies associated with the set of classes within the set of n-dimensional images” [Arbelo-Gonzalez Claim 5]. “The class weight for a given class could be computed by dividing the number of occurrences of the most frequent class in training 3D images 240 (or a given set of training 3D patches 258 extracted from training 3D images) by the number of occurrences of the given class in training 3D images 240 (or the set of training 3D patches 258). Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model. “More specifically, the loss function includes different weights for different classes of objects with which training data for the machine learning model is labeled” [Arbelo-Gonzalez ¶ 17]. “The class weight for a given class could be computed by dividing the number of occurrences of the most frequent class in training 3D images 240 (or a given set of training 3D patches 258 extracted from training 3D images) by the number of occurrences of the given class in training 3D images 240 (or the set of training 3D patches 258). Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
Arbelo-Gonzalez is considered to be analogous to the claimed invention because it is in the same field of machine learning. The metadata of the quantity of data records of Arbelo-Gonzalez can be combined with the deduplicated attribute values of Sivakumar in view of Goldsteen such that this metadata can be used to determine weights for the data records as taught by the method of Arbelo-Gonzalez. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen to incorporate the teachings of Arbelo-Gonzalez and include add metadata to the deduplicated attribute values to indicate a quantity of data records associated with each of the median range values; and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model. Doing so would allow for bias correction within the machine learning model training process. “Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez fails to explicitly teach and migrate the deduplicated attribute values, from the core cloud environment to the edge cloud environment, … data transmitted to and stored in the edge cloud environment.
and migrate the deduplicated attribute values, from the core cloud environment to the edge cloud environment, “At step 422, data for the user is accessed from the edge computing storage. The data may be migrated or moved to the edge computing storage from a central storage, such as a cloud computing storage utilized by the user. In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
data transmitted to and stored in the edge cloud environment, “At step 422, data for the user is accessed from the edge computing storage. The data may be migrated or moved to the edge computing storage from a central storage, such as a cloud computing storage utilized by the user. In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
Todasco is considered to be analogous to the claimed invention because it is in the same field of indexing schemes relating to priority. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez to incorporate the teachings of Todasco and include and migrate the deduplicated attribute values, from the core cloud environment to the edge cloud environment, … data transmitted to and stored in the edge cloud environment. Doing so would allow for useful data to be more quickly accessed. “In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
With regard to claim 9, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the data materialization platform of claim 8, as referenced above. Sivakumar fails to teach wherein the one or more processors, to deduplicate the subset of the attribute values, are configured to: identify a first attribute value and a second attribute value, from among the subset of the attribute values, for which outputs of the machine learning model are estimated to be the same output within the tolerance; determine a median range value, of the median range values, based on a first range value associated with the first attribute value and a second range value associated with the second range value; and associate the median range value with a deduplicated attribute value of the deduplicated attribute values.
However, Goldsteen teaches:
wherein the one or more processors, to deduplicate the subset of the attribute values, are configured to: identify a first attribute value and a second attribute value, from among the subset of the attribute values, “The summary plot 900 illustrates the relationship between the value of a feature and the impact on the prediction. This information allows the disclosed generalization process to identify feature values for generalization. For example, regions 902 and 904 present good potential candidates for generalization because they are groups of similar feature values (region 902 are all low feature values; region 904 are all high feature values) with similar SHAP values, and the associated feature (exer_angina) has a low importance value” [Goldsteen ¶ 138, fig. 9 Examiner notes the plurality of data points identified in region 902 and the plurality of data points identified in region 904; any two of these data points from one region can be considered the first and second attribute values in this case].
for which outputs of the machine learning model are estimated to be the same output within the tolerance; “As an example, data records that provide for feature data that is representative of people's ages may be used by a predictive ML model to predict heart attacks with about 90% accuracy. Data minimization that involves feature generalization may include replacing certain ages with an age range that still allows the ML model to maintain an accuracy of about 90%” [Goldsteen ¶ 21].
determine a median range value, of the median range values, based on a first range value associated with the first attribute value and a second range value associated with the second range value; “In some embodiments, where the generalization is a numerical range, the representative value is a median value of the numerical range (e.g., the first representative value for the 1-20 range is 10.5)” [Goldsteen ¶ 129]. “… wherein the constructing of the generalization group comprises identifying a continuous range of numerical feature values having explainability values that satisfy the predetermined condition … computing the generalized feature value to be a median value of the continuous range of numerical feature values” [Goldsteen Claims 5-6].
and associate the median range value with a deduplicated attribute value of the deduplicated attribute values. “… computing the generalized feature value to be a median value of the continuous range of numerical feature values” [Goldsteen Claim 6].
With regard to claim 10, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the data materialization platform of claim 9, as referenced above.
While Sivakumar teaches that identified duplicate attribute values are not to be saved: “If there is a match between the fingerprints, the incoming data is not saved because the data blocks are already present in the customer data volume 256” [Sivakumar ¶ 40], it does not explicitly teach remove the first attribute value and the second attribute value from the attribute values. However, Goldsteen teaches wherein the one or more processors, to deduplicate the subset of the attribute values, are configured to: remove the first attribute value and the second attribute value from the attribute values; “The generalization module 316 then replaces the age feature, which includes any integer from 1 to 120 as a feature value, with an alternative feature representative of age ranges, for example [1-20], [21-25], [26-50], [51-120]. Thus, the granularity of the selected feature is reduced with the alternative feature, from a domain of 120 separate values, to a generalized domain of four separate values” [Goldsteen ¶ 92 examiner notes replacing the integers 1 to 120 with alternative features representative of ranges is considered a removal of the attribute values 1-120].
Sivakumar fails to explicitly teach and replace the first attribute value and the second attribute value, in the attribute values, with the deduplicated attribute value.
However, Goldsteen teaches and replace the first attribute value and the second attribute value, in the attribute values, with the deduplicated attribute value. “For example, in some embodiments, the generalization process identifies a slice for feature generalization processing according to a prioritization determined by the generalization process. The generalization process replaces the feature (or features) associated with the selected slice with an alternative feature (or alternative features) that is a generalization of the selected feature” [Goldsteen ¶ 35].
With regard to claim 12, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the data materialization platform of claim 8, as referenced above. Sivakumar further teaches wherein the one or more processors are further configured to: label the subset of the attribute values as repetitive attribute values. “Data deduplication is a technique for eliminating duplicate copies of repeating data. The deduplication process requires comparison of data "chunks" Data deduplication is a technique for eliminating duplicate copies of repeating data. The deduplication process requires comparison of data "chunks" (also known as "byte patterns") which are unique, contiguous blocks of data. These chunks are identified and stored during a process of analysis and compared to other chunks within existing data. Whenever a match occurs, the redundant chunk is replaced with a small reference that points to the stored chunk” [Sivakumar ¶ 17].
With regard to claim 13, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the data materialization platform of claim 8, as referenced above. Sivakumar fails to teach wherein the one or more processors, to analyze the attribute values to identify the subset of the attribute values, are configured to: determine that the subset of the attribute values are within an attribute value range for which outputs of the machine learning model are estimated to be the same output within the tolerance.
However, Goldsteen teaches:
wherein the one or more processors, to analyze the attribute values to identify the subset of the attribute values, are configured to: determine that the subset of the attribute values are within an attribute value range “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy. In some embodiments, a generalized version of numerical feature data includes a list of ranges, and a generalized version of categorical feature data is one or more groups of values” [Goldsteen ¶ 20]. “In some embodiments, data mapping instructions include instructions to map feature values in a generalized range or group of values to a representative value for the generalized range or group. For example, if the generalization process has replaced the age feature, which includes any integer from 1 to 120 as a feature value, with an alternative feature representative of age ranges, for example [1-20], [21-25], [26-50], [51-120], the value mapping data will instruct the value mapping module to map any age from 1 to 20 to a first representative value for the first range, map any age from 21 to 25 to a second representative value for the second range, map any age from 26 to 50 to a third representative value for the third range, and map any age from 51 to 120 to a fourth representative value for the fourth range” [Goldsteen ¶ 43].
for which outputs of the machine learning model are estimated to be the same output within the tolerance. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy. In some embodiments, a generalized version of numerical feature data includes a list of ranges, and a generalized version of categorical feature data is one or more groups of values” [Goldsteen ¶ 20].
With regard to claim 14, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the data materialization platform of claim 13, as referenced above. Sivakumar fails to teach wherein the one or more processors are configured to: identify a median range value, of the median range values, based on the median value range being associated with the attribute value range.
However, Goldsteen teaches wherein the one or more processors are configured to: identify a median range value, of the median range values, based on the median value range being associated with the attribute value range. “In some embodiments, data mapping instructions include instructions to map feature values in a generalized range or group of values to a representative value for the generalized range or group. For example, if the generalization process has replaced the age feature, which includes any integer from 1 to 120 as a feature value, with an alternative feature representative of age ranges, for example [1-20], [21-25], [26-50], [51-120], the value mapping data will instruct the value mapping module to map any age from 1 to 20 to a first representative value for the first range, map any age from 21 to 25 to a second representative value for the second range, map any age from 26 to 50 to a third representative value for the third range, and map any age from 51 to 120 to a fourth representative value for the fourth range” [Goldsteen ¶ 43].
Claims 11 is rejected under 35 U.S.C. 103 as being unpatentable over Sivakumar (US 2024/0223675 A1) in view of Goldsteen (US 2024/0160965 A1) in view of Arbelo-Gonzalez (US 2024/0281981 A1) in view of Todasco (US 2023/0036623 A1) in view of Laliberté (US 2024/0220887 A1).
With regard to claim 11, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the data materialization platform of claim 8, as referenced above. Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco fails to explicitly teach wherein the one or more processors, to analyze the attribute values, are configured to: analyze, using another machine learning model, the attribute values to identify ranges associated with the attribute values.
However, Laliberté teaches wherein the one or more processors, to analyze the attribute values, are configured to: analyze, using another machine learning model, the attribute values to identify ranges associated with the attribute values. “In some embodiments, a method for determining project attribute range values for at least one project attribute, such as a project cost and/or a project schedule, of at least one new project includes receiving historical data related to at least one previous performance of a same or similar project as the at least one new project, the historical data including historical project attribute values of the at least one project attribute, generating multiple respective machine learning models using different sets of training data determined from the received historical data, each of the different sets of the training data being used to train a respective one of the machine learning models, and determining a range of values for the at least one project attribute of the at least one new project by applying the multiple respective machine learning models to the at least one project attribute of the new project” [Laliberté ¶ 4].
Laliberté is considered to be analogous to the claimed invention because it is in the same field of machine learning. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco to incorporate the teachings of Laliberté and include that the one or more processors, to analyze the attribute values, are configured to: analyze, using another machine learning model, the attribute values to identify ranges associated with the attribute values. Doing so would allow for more accurate determinations of ranges for the attribute values. “Embodiments of PAE system in accordance with the present principles create a more accurate prediction of project attribute value ranges as described above” [Laliberté ¶ 81].
Claims 15-19 and 21-22 are rejected under 35 U.S.C. 103 as being unpatentable over Sivakumar (US 2024/0223675 A1) in view of Goldsteen (US 2024/0160965 A1) in view of Arbelo-Gonzalez (US 2024/0281981 A1) in view of Todasco (US 2023/0036623 A1) in view of Bonafe (US 2024/0220914 A1).
With regard to claim 15, Sivakumar teaches:
A non-transitory computer-readable medium storing a set of instructions, the set of instructions comprising: “computer program product embodiment ("CPP embodiment" or "CPP") is a term used in the present disclosure to describe any set of one, or more, storage media (also called "mediums") collectively included in a set of one, or more, storage devices that collectively include machine readable code corresponding to instructions and/or data for performing computer operations specified in a given CPP claim. A "storage device" is any tangible device that can retain and store instructions for use by a computer processor” [Sivakumar ¶ 18].
one or more instructions that, when executed by one or more processors of a data materialization platform, cause the data materialization platform to: “Computer readable program instructions are typically loaded onto computer 101 to cause a series of operational steps to be performed by processor set 110 of computer 101 and thereby effect a computer-implemented method, such that the instructions thus executed will instantiate the methods specified in flowcharts and/or narrative descriptions of computer-implemented methods included in this document (collectively referred to as "the inventive methods")” [Sivakumar ¶ 22].
perform a data migration process between a core cloud environment and an edge cloud environment; “The data storage engine 250 of the public cloud (core cloud) 105 sends the fingerprint calculator algorithm 222A (which is stored as fingerprint calculator algorithm 222B) and the map 220A (which is stored as the map 220B) to the edge device 202 such that the traffic management code 150 can use fingerprint calculator algorithm 222B to calculate the deduplication fingerprints for all the participating blocks in the edge cache 210, for eventual comparison with the stored fingerprints in the map 220B (e.g., B-tree)” [Sivakumar ¶ 52].
identify, during the data migration process, a plurality of feature-set values stored in a data repository of the core cloud environment, “When the data storage engine 250 is running additional deduplication on the incoming data, on every incoming data request, the data storage engine 250 calculates the fingerprints (feature-set values) of the incoming data using a fingerprint algorithm 222A, checks for a match to the fingerprints of the available track records in a map 220A, and decides to execute the actual write to the disk of the customer data volume 256 based on whether the fingerprints of the incoming data blocks match the stored fingerprints in the map 220A” [Sivakumar ¶ 40, Fig. 2 Examiner notes the customer data volume 256 within public cloud 105]. “This may be accomplished by performing the identification of deduplication fingerprints at the edge device level, by identifying data blocks/chunks that are saved in the cache of the edge device but not saved at the target device (e.g., the cloud). Accordingly, the edge device selectively sends the packets of data chunks, which are not saved at the target device, from the edge device to the target device at the cloud in order to avoid any unnecessary data transfer to the target device” [Sivakumar ¶ 11].
…that is to be executed by the edge cloud environment; “Edge computing upends the traditional architecture by shifting key processing functions away from the core of the network and out to the edge where users are located. Through a combination of edge data centers and IoT devices that can process data for themselves, edge computing can improve network performance and reduce latency” [Sivakumar ¶ 2].
deduplicate the subset of feature-set values from the data repository to generate deduplicated feature-set values “Data deduplication is a technique for eliminating duplicate copies of repeating data. The deduplication process requires comparison of data "chunks" (also known as "byte patterns") which are unique, contiguous blocks of data. These chunks are identified and stored during a process of analysis and compared to other chunks within existing data. Whenever a match occurs, the redundant chunk is replaced with a small reference that points to the stored chunk. Given that the same byte pattern may occur dozens, hundreds, or even thousands of times (the match frequency is dependent on the chunk size), the amount of data that is stored or transferred can be greatly reduced” [Sivakumar ¶ 15]. “When the traffic management code 150 of the edge device 202 is communicating with a data storage engine 250 of the public cloud 105 and issuing operations (e.g., READ/WRITE operations) on the customer data volume 256 (which can be issued via a data receiver (e.g., IBM Spectrum Virtualize™ for public cloud (SVPC) in the gateway 140), the data storage engine 250 offers many data optimization features, such as, for example, data deduplication and others … When the data storage engine 250 is running additional deduplication on the incoming data, on every incoming data request, the data storage engine 250 calculates the fingerprints (feature-set values) of the incoming data using a fingerprint algorithm 222A, checks for a match to the fingerprints of the available track records in a map 220A, and decides to execute the actual write to the disk of the customer data volume 256 based on whether the fingerprints of the incoming data blocks match the stored fingerprints in the map 220A” [Sivakumar ¶ 40].
wherein the deduplicating is performed prior to the migrating to reduce a volume of the training data transmitted to and stored in the edge cloud environment, “This may be accomplished by performing the identification of deduplication fingerprints at the edge device level, by identifying data blocks/chunks that are saved in the cache of the edge device but not saved at the target device (e.g., the cloud). Accordingly, the edge device selectively sends the packets of data chunks, which are not saved at the target device, from the edge device to the target device at the cloud in order to avoid any unnecessary data transfer to the target device” [Sivakumar ¶ 11].
Sivakumar fails to teach wherein the plurality of feature-set values each includes a dataset value and a vector, and wherein the dataset value and the vector of each of the plurality of feature-set values are to be used as training data to train a machine learning model that is to be executed by the edge cloud environment; analyze, using one or more input rules, the dataset value of each of the plurality of feature-set values to identify a subset of dataset values, associated with a subset of feature-set values of the plurality of feature-set values, for which outputs of the machine learning model are estimated to be a same output within a tolerance; and deduplicate the subset of feature-set values from the data repository to generate deduplicated feature-set values that are associated with a first median range value for the subset of dataset values … a subset of vectors that are associated with the deduplicated feature-set values; add metadata to the deduplicated feature-set values … wherein the deduplicated feature-set values are used as the training data to train the machine learning model in the edge cloud environment.
However, Goldsteen teaches:
wherein the plurality of feature-set values each includes a dataset value and a vector, “This can be achieved by computing an overall explainability (for example SHAP) value for all samples sharing the same (or similar for continuous features) feature value with sign (directionality), for example, by computing the sum and/or mean and/or variance of SHAP values on sub-groups of samples that share a same feature value. In some embodiments, the process compares a sum of SHAP values of samples having a particular feature value to SHAP values associated with other feature values in an effort to locate other feature values of the same feature that have similar SHAP and direction values” [Goldsteen ¶ 30].
and wherein the dataset value and the vector of each of the plurality of feature-set values are to be used as training data to train “In some embodiments, the evaluation also considers the SHAP direction, so the SHAP values and directions are compared for different feature values to identify feature values that are good generalization candidates” [Goldsteen ¶ 30]. “In some embodiments, the LEA includes a SHapley Additive exPlanations (SHAP) analysis that results in Shapley values for each of the features in each of the input datasets (e.g., training instances). Each Shapley value represents the contribution of a given feature towards the prediction of the model for a given input dataset. The explanation presents how strongly, and which way, a given feature affects the prediction of the predictive model” [Goldsteen ¶ 140]. “An embodiment includes generating feature explainability data associated with a feature, wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Predictive modeling refers to a form of predictive analytics that typically uses a machine learning (ML) algorithm to build a predictive model (also referred to herein as a ML model or ML predictive model)” [Goldsteen ¶ 17]. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20]. “For example, in some embodiments, the ML model is a classification model trained to make class predictions from tabular input data, which may include features having numerical, categorical, and/or continuous data domains” [Goldsteen ¶ 70].
a machine learning model “An embodiment includes generating feature explainability data associated with a feature, wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Predictive modeling refers to a form of predictive analytics that typically uses a machine learning (ML) algorithm to build a predictive model (also referred to herein as a ML model or ML predictive model)” [Goldsteen ¶ 17].
that is to be executed by the edge cloud environment; “Computer readable program instructions described herein can be downloaded to respective computing/processing devices from a computer readable storage medium or to an external computer or external storage device via a network, for example, the Internet, a local area network, a wide area network and/or a wireless network. The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers” [Goldsteen ¶ 159].
analyze, using one or more input rules, “The embodiment also includes extracting feature value data from the input data, where the feature value data is representative of a feature value of the feature for the sample. The embodiment also includes constructing a generalization group comprising the feature of the sample, where the constructing of the generalization group comprises detecting that the feature value of the feature and the explainability value of the feature satisfy a predetermined condition (rule)” [Goldsteen ¶ 3]. “In an exemplary embodiment, groups of samples (referred to as slices) are identified that satisfy certain conditions (rules) related to feature values and local explainability values, and therefore can be generalized together. In various embodiments, the conditions may be based on distances between feature/explainability values, ranges of feature/ explainability values, variance of feature/explainability values, size of the slice, etc.” [Goldsteen ¶ 29].
the dataset value of each of the plurality of feature-set values to identify a subset of dataset values, associated with a subset of feature-set values of the plurality of feature-set values, for which outputs of the machine learning model are estimated to be a same output within a tolerance; “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy. In some embodiments, a generalized version of numerical feature data includes a list of ranges, and a generalized version of categorical feature data is one or more groups of values” [Goldsteen ¶ 20]. “As an example, data records that provide for feature data that is representative of people's ages may be used by a predictive ML model to predict heart attacks with about 90% accuracy. Data minimization that involves feature generalization may include replacing certain ages with an age range that still allows the ML model to maintain an accuracy of about 90%” [Goldsteen ¶ 21].
and deduplicate the subset of feature-set values from the data repository to generate deduplicated feature-set values that are associated with a first median range value for the subset of dataset values “The generalization process replaces the feature (or features) associated with the selected slice with an alternative feature (or alternative features) that is a generalization of the selected feature. As an example, the selected feature may be representative of age based on the generalization process determining that the age feature has a relatively low feature importance value and the generalization process identifying groups of ages that are good generalization candidates. The generalization process then replaces the age feature, which includes any integer from 1 to 120 as a feature value, with an alternative feature representative of age ranges, for example [1-20], [21-25], [26-50], [51-120].” [Goldsteen ¶ 35]. “In some embodiments, where the generalization is a numerical range, the representative value is a median value of the numerical range (e.g., the first representative value for the 1-20 range is 10.5)” [Goldsteen ¶ 129].
a subset of vectors that are associated with the deduplicated feature-set values; “This can be achieved by computing an overall explainability (for example SHAP) value for all samples sharing the same (or similar for continuous features) feature value with sign (directionality), for example, by computing the sum and/or mean and/or variance of SHAP values on sub-groups of samples that share a same feature value. In some embodiments, the process compares a sum of SHAP values of samples having a particular feature value to SHAP values associated with other feature values in an effort to locate other feature values of the same feature that have similar SHAP and direction values” [Goldsteen ¶ 30].
add metadata to the deduplicated feature-set values “An embodiment includes generating feature explainability data associated with a feature (attribute), wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20].
wherein the deduplicated feature-set values are used as the training data to train the machine learning model in the edge cloud environment, “An embodiment includes generating feature explainability data associated with a feature (attribute), wherein the feature is in a set of features of a sample represented by input data for a predictive model, wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3]. “Predictive modeling refers to a form of predictive analytics that typically uses a machine learning (ML) algorithm to build a predictive model (also referred to herein as a ML model or ML predictive model)” [Goldsteen ¶ 17]. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20]. “For example, in some embodiments, the ML model is a classification model trained to make class predictions from tabular input data, which may include features having numerical, categorical, and/or continuous data domains” [Goldsteen ¶ 70]. “The network may comprise copper transmission cables, optical transmission fibers, wireless transmission, routers, firewalls, switches, gateway computers and/or edge servers” [Goldsteen ¶ 159].
Goldsteen is considered to be analogous to the claimed invention because it is in the same field of machine learning. The method of range calculation for machine learning inputs of Goldsteen can be combined with the core cloud and edge could system of Sivakumar such that the fingerprints (attributes) of Sivakumar can be deduplicated using the method of Goldsteen. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar to incorporate the teachings of Goldsteen and include that the plurality of feature-set values each includes a dataset value and a vector, and wherein the dataset value and the vector of each of the plurality of feature-set values are to be used as training data to train a machine learning model that is to be executed by the edge cloud environment; analyze, using one or more input rules, the dataset value of each of the plurality of feature-set values to identify a subset of dataset values, associated with a subset of feature-set values of the plurality of feature-set values, for which outputs of the machine learning model are estimated to be a same output within a tolerance; and deduplicate the subset of feature-set values from the data repository to generate deduplicated feature-set values that are associated with a first median range value for the subset of dataset values … a subset of vectors that are associated with the deduplicated feature-set values; add metadata to the deduplicated feature-set values … wherein the deduplicated feature-set values are used as the training data to train the machine learning model in the edge cloud environment. Doing so would allow for the use of the system to make predictions within a certain level of accuracy. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy” [Goldsteen ¶ 20].
Sivakumar in view of Goldsteen fails to teach add metadata to the deduplicated feature-set values to indicate a quantity of data records associated with each of the first median range value and the second median range value; and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model.
However, Arbelo-Gonzalez teaches:
add metadata to (machine learning training data) the deduplicated feature-set values to indicate a quantity of data records associated with each of the (classes) first median range value and the second median range value; “wherein training the machine learning model further comprises determining the first set of weights based on a set of frequencies associated with the set of classes within the set of n-dimensional images” [Arbelo-Gonzalez Claim 5]. “The class weight for a given class could be computed by dividing the number of occurrences of the most frequent class in training 3D images 240 (or a given set of training 3D patches 258 extracted from training 3D images) by the number of occurrences of the given class in training 3D images 240 (or the set of training 3D patches 258). Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model. “More specifically, the loss function includes different weights for different classes of objects with which training data for the machine learning model is labeled” [Arbelo-Gonzalez ¶ 17]. “The class weight for a given class could be computed by dividing the number of occurrences of the most frequent class in training 3D images 240 (or a given set of training 3D patches 258 extracted from training 3D images) by the number of occurrences of the given class in training 3D images 240 (or the set of training 3D patches 258). Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
Arbelo-Gonzalez is considered to be analogous to the claimed invention because it is in the same field of machine learning. The metadata of the quantity of data records of Arbelo-Gonzalez can be combined with the deduplicated feature-set values of Sivakumar in view of Goldsteen such that this metadata can be used to determine weights for the data records as taught by the method of Arbelo-Gonzalez. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen to incorporate the teachings of Arbelo-Gonzalez and include add metadata to the deduplicated feature-set values to indicate a quantity of data records associated with each of the first median range value and the second median range value; and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model. Doing so would allow for bias correction within the machine learning model training process. “Because the class weight for a given class is inversely proportional to the frequency of the class, losses 216 that incorporate class weights 214 can be used to correct for the bias of 3D segmentation model 228 toward classes that occur more frequently in training 3D images 240 (or training 3D patches 258 extracted from training 3D images 240)” [Arbelo-Gonzalez ¶ 49].
Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez fails to explicitly teach and migrate the deduplicated feature-set values from the core cloud environment to the edge cloud environment, … data transmitted to and stored in the edge cloud environment.
and migrate the deduplicated feature-set values from the core cloud environment to the edge cloud environment, “At step 422, data for the user is accessed from the edge computing storage. The data may be migrated or moved to the edge computing storage from a central storage, such as a cloud computing storage utilized by the user. In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
data transmitted to and stored in the edge cloud environment, “At step 422, data for the user is accessed from the edge computing storage. The data may be migrated or moved to the edge computing storage from a central storage, such as a cloud computing storage utilized by the user. In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
Todasco is considered to be analogous to the claimed invention because it is in the same field of indexing schemes relating to priority. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez to incorporate the teachings of Todasco and include migrate the deduplicated feature-set values from the core cloud environment to the edge cloud environment, … data transmitted to and stored in the edge cloud environment. Doing so would allow for useful data to be more quickly accessed. “In one or more embodiments, the data may be determined or predicted as useful or relevant to the user at the location, and therefore stored or migrated to the edge computing storage so it may be more quickly accessed (e.g., with lower load times and/or latency)” [Todasco ¶ 79].
Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco fails to explicitly teach and wherein the dataset value and the vector of each of the plurality of feature-set values are to be used as inputs and a second median range value for a subset of vectors.
However, Bonafe teaches:
and wherein the dataset value and the vector of each of the plurality of feature-set values are to be used as inputs “In one or more embodiments, the neural network 705 learns features from the training data input(s) 715 and responsively applies weights to them during training” [Bonafe ¶ 121].
and a second median range value for a subset of vectors “One or more embodiments determine one or more feature vectors representing the input(s) 615 in vector space by aggregating (for example, mean/median or dot product) the feature vector values to arrive at a particular point in feature space” [Bonafe ¶ 120].
Bonafe is considered to be analogous to the claimed invention because it is in the same field of machine learning. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco to incorporate the teachings of Bonafe and include that the dataset value and the vector of each of the plurality of feature-set values are to be used as inputs and a second median range value for a subset of vectors. Doing so would allow for the use of feature vectors in the detection of similar features. “In another illustrative example of training, one or more embodiments learn an embedding of feature vectors based on learning (for example, deep learning) to detect similar features between training data input(s) 715 in feature space using distance measures, such as cosine (or Euclidian) distance.” [Bonafe ¶ 122].
With regard to claim 16, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco in view of Bonafe teaches the non-transitory computer-readable medium of claim 15, as referenced above. Sivakumar fails to teach wherein the one or more instructions, that cause the one or more processors to analyze the dataset value of each of the plurality of feature-set values, cause the data materialization platform to: determine that the subset of dataset values are within a dataset value range, wherein the dataset value range includes a range of dataset values for which outputs of the machine learning model are estimated to be the same output within a tolerance.
However, Goldsteen teaches:
wherein the one or more instructions, that cause the one or more processors to analyze the dataset value of each of the plurality of feature-set values, cause the data materialization platform to: determine that the subset of dataset values are within a dataset value range, “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy. In some embodiments, a generalized version of numerical feature data includes a list of ranges, and a generalized version of categorical feature data is one or more groups of values” [Goldsteen ¶ 20]. “In some embodiments, data mapping instructions include instructions to map feature values in a generalized range or group of values to a representative value for the generalized range or group. For example, if the generalization process has replaced the age feature, which includes any integer from 1 to 120 as a feature value, with an alternative feature representative of age ranges, for example [1-20], [21-25], [26-50], [51-120], the value mapping data will instruct the value mapping module to map any age from 1 to 20 to a first representative value for the first range, map any age from 21 to 25 to a second representative value for the second range, map any age from 26 to 50 to a third representative value for the third range, and map any age from 51 to 120 to a fourth representative value for the fourth range” [Goldsteen ¶ 43].
wherein the dataset value range includes a range of dataset values for which outputs of the machine learning model are estimated to be the same output within a tolerance. “Exemplary embodiments disclosed herein provide for a determination of a generalized version of the feature data that is input to an ML model and still enables the model to make predictions above a threshold level of accuracy. In some embodiments, a generalized version of numerical feature data includes a list of ranges, and a generalized version of categorical feature data is one or more groups of values” [Goldsteen ¶ 20].
With regard to claim 17, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco in view of Bonafe teaches the non-transitory computer-readable medium of claim 16, as referenced above. Sivakumar fails to teach wherein the first median range value is associated with the dataset value range.
However, Goldsteen teaches wherein the first median range value is associated with the dataset value range. “In some embodiments, where the generalization is a numerical range, the representative value is a median value of the numerical range (e.g., the first representative value for the 1-20 range is 10.5)” [Goldsteen ¶ 129].
With regard to claim 18, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco in view of Bonafe teaches the non-transitory computer-readable medium of claim 15, as referenced above. Sivakumar fails to teach wherein the one or more instructions, when executed by the one or more processors, further cause the data materialization platform to: identify the first median range value based on a first input rule of the one or more input rules; and identify the second median range value based on a second input rule of the one or more input rules.
However, Goldsteen teaches:
wherein the one or more instructions, when executed by the one or more processors, further cause the data materialization platform to: identify the first median range value based on a first input rule of the one or more input rules; “wherein the predetermined condition comprises at least one of a first threshold minimum difference between feature values (first input rule) and a second threshold minimum difference between explainability values” [Goldsteen Claim 2]. “The embodiment also includes constructing a generalization group comprising the feature of the sample, where the constructing of the generalization group comprises detecting that the feature value of the feature and the explainability value of the feature satisfy a predetermined condition” [Goldsteen ¶ 3]. “In some embodiments, where the generalization is a numerical range, the representative value is a median value of the numerical range (e.g., the first representative value for the 1-20 range is 10.5)” [Goldsteen ¶ 129].
and identify the second median range value based on a second input rule of the one or more input rules. “The computer-implemented method of claim 1, wherein the predetermined condition comprises at least one of a first threshold minimum difference between feature values and a second threshold minimum difference between explainability values (second input rule)” [Goldsteen Claim 2]. “The embodiment also includes constructing a generalization group comprising the feature of the sample, where the constructing of the generalization group comprises detecting that the feature value of the feature and the explainability value of the feature satisfy a predetermined condition” [Goldsteen ¶ 3]. “In some embodiments, where the generalization is a numerical range, the representative value is a median value of the numerical range (e.g., the first representative value for the 1-20 range is 10.5)” [Goldsteen ¶ 129].
Sivakumar in view of Goldsteen fails to explicitly teach identify the second median range value.
However, Bonafe teaches identify the second median range value “One or more embodiments determine one or more feature vectors representing the input(s) 615 in vector space by aggregating (for example, mean/median or dot product) the feature vector values to arrive at a particular point in feature space” [Bonafe ¶ 120].
With regard to claim 19, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco in view of Bonafe teaches the non-transitory computer-readable medium of claim 18, as referenced above. Sivakumar fails to teach wherein the second input rule associates the second median range value with the first median range value.
However, Goldsteen teaches wherein the second input rule associates the second median range value with the first median range value. “Next, at block 1004, the process identifies groups of samples (referred to as slices) that satisfy certain conditions related to feature values and local explainability values, and therefore can be generalized together” [Goldsteen 141]. “wherein the feature explainability data is representative of an explainability value of the feature, wherein the explainability value is based at least in part on a significance of the feature for an output of the predictive model for the sample” [Goldsteen ¶ 3].
With regard to claim 21, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the method of claim 1, as referenced above. Sivakumar fails to teach wherein each of the deduplicated attribute values includes a dataset value.
However, Goldsteen teaches wherein each of the deduplicated attribute values includes a dataset value. “This can be achieved by computing an overall explainability (for example SHAP) value for all samples sharing the same (or similar for continuous features) feature value with sign (directionality), for example, by computing the sum and/or mean and/or variance of SHAP values on sub-groups of samples that share a same feature value. In some embodiments, the process compares a sum of SHAP values of samples having a particular feature value to SHAP values associated with other feature values in an effort to locate other feature values of the same feature that have similar SHAP and direction values” [Goldsteen ¶ 30].
Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco fails to explicitly teach wherein each of the deduplicated attribute values includes a dataset value and a vector.
However, Bonafe teaches:
wherein each of the deduplicated attribute values includes a dataset value and a vector. “In one or more embodiments, the neural network 705 learns features from the training data input(s) 715 and responsively applies weights to them during training” [Bonafe ¶ 121]. “One or more embodiments determine one or more feature vectors representing the input(s) 615 in vector space by aggregating (for example, mean/median or dot product) the feature vector values to arrive at a particular point in feature space” [Bonafe ¶ 120].
Bonafe is considered to be analogous to the claimed invention because it is in the same field of machine learning. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco to incorporate the teachings of Bonafe and include that wherein each of the deduplicated attribute values includes a dataset value and a vector. Doing so would allow for the use of feature vectors in the detection of similar features. “In another illustrative example of training, one or more embodiments learn an embedding of feature vectors based on learning (for example, deep learning) to detect similar features between training data input(s) 715 in feature space using distance measures, such as cosine (or Euclidian) distance.” [Bonafe ¶ 122].
With regard to claim 22, Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco teaches the data materialization platform of claim 8, as referenced above. Sivakumar fails to teach wherein each of the deduplicated attribute values includes a dataset value.
However, Goldsteen teaches wherein each of the deduplicated attribute values includes a dataset value. “This can be achieved by computing an overall explainability (for example SHAP) value for all samples sharing the same (or similar for continuous features) feature value with sign (directionality), for example, by computing the sum and/or mean and/or variance of SHAP values on sub-groups of samples that share a same feature value. In some embodiments, the process compares a sum of SHAP values of samples having a particular feature value to SHAP values associated with other feature values in an effort to locate other feature values of the same feature that have similar SHAP and direction values” [Goldsteen ¶ 30].
Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco fails to explicitly teach wherein each of the deduplicated attribute values includes a dataset value and a vector.
However, Bonafe teaches:
wherein each of the deduplicated attribute values includes a dataset value and a vector. “In one or more embodiments, the neural network 705 learns features from the training data input(s) 715 and responsively applies weights to them during training” [Bonafe ¶ 121]. “One or more embodiments determine one or more feature vectors representing the input(s) 615 in vector space by aggregating (for example, mean/median or dot product) the feature vector values to arrive at a particular point in feature space” [Bonafe ¶ 120].
Bonafe is considered to be analogous to the claimed invention because it is in the same field of machine learning. Therefore, it would be obvious to someone of ordinary skill in the art before the effective filing date of the claimed invention to have modified Sivakumar in view of Goldsteen in view of Arbelo-Gonzalez in view of Todasco to incorporate the teachings of Bonafe and include that wherein each of the deduplicated attribute values includes a dataset value and a vector. Doing so would allow for the use of feature vectors in the detection of similar features. “In another illustrative example of training, one or more embodiments learn an embedding of feature vectors based on learning (for example, deep learning) to detect similar features between training data input(s) 715 in feature space using distance measures, such as cosine (or Euclidian) distance.” [Bonafe ¶ 122].
Response to Arguments
Applicant's arguments filed 12/30/2025 have been fully considered but they are not persuasive. Applicant argues in substance:
I. Without acquiescing in the Examiner's rejection, the cited sections of the applied references, whether taken alone or in any reasonable combination, do not disclose at least "adding metadata to the deduplicated attribute values to indicate a quantity of data records associated with each of the median range values; and migrating the deduplicated attribute values from the core cloud environment to the edge cloud environment, wherein the deduplicated attribute values are used as training data to train the machine learning model in the edge cloud environment ... and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model," as recited in claim 1, as amended. Independent claims 8, as amended, recite similar features. Therefore, independent claims 1 and 8, and the claims that depend thereon, are patentable over the cited sections of the applied references.
Accordingly, Applicant respectfully requests that the Examiner reconsider and withdraw the rejection of claims 1-10 and 12-14 under 35 U.S.C. § 103 based on SIVAKUMAR and GOLDSTEEN.
Without acquiescing in the Examiner's rejection, the cited sections of the applied references, whether taken alone or in any reasonable combination, do not disclose at least causing the data materialization platform to "add metadata to the deduplicated feature-set values to indicate a quantity of data records associated with each of the first median range value and the second median range value; and migrate the deduplicated feature-set values from the core cloud environment to the edge cloud environment, wherein the deduplicated feature-set values are used as the training data to train the machine learning model in the edge cloud environment . . . and wherein the machine learning model uses the metadata to determine weights for the data records during training of the machine learning model," as recited in claim 15, as amended. Therefore, independent claim 15, and the claims that depend thereon, are patentable over the cited sections of the applied references.
Accordingly, Applicant respectfully requests that the Examiner reconsider and withdraw the rejection of claims 15-20 under 35 U.S.C. § 103 based on SIVAKUMAR, GOLDSTEEN, and BONAFE.
a) Examiner respectfully disagrees. As detailed in the rejection above, the combination of Sivakumar in view of Goldsteen teaches wherein the deduplicated attribute values are used as training data to train the machine learning model in the edge cloud environment [Goldsteen ¶ 3, 17, 20, 70, 159].
Applicant’s further arguments with respect to claim(s) 1, 3-19, and 21-22 have been considered but are moot because the new ground of rejection does not rely on any reference applied in the prior rejection of record for any teaching or matter specifically challenged in the argument.
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.
Examiner respectfully requests, in response to this Office action, support be shown for language added to any original claims on amendment and any new claims. That is, indicate support for newly added claim language by specifically pointing to page(s) and line number(s) in the specification and/or drawing figure(s). This will assist Examiner in prosecuting the application.
When responding to this Office Action, Applicant is advised to clearly point out the patentable novelty which he or she thinks the claims present, in view of the state of the art disclosed by the references cited or the objections made. He or she must also show how the amendments avoid such references or objections. See 37 CFR 1.111(c).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to ARI F RIGGINS whose telephone number is (571)272-2772. The examiner can normally be reached Monday-Friday 7:00AM-4:30PM.
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, Bradley Teets can be reached at (571) 272-3338. 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.
/A.F.R./Examiner, Art Unit 2197
/BRADLEY A TEETS/Supervisory Patent Examiner, Art Unit 2197
Prosecution Insights