DETAILED ACTION
Notice of Pre-AIA or AIA Status
The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
This office action is in response to Applicant’s communication filed on 01/26/2026. Claims 1-2,4-12,14-22 have been examined. Claims 3, 13 are cancelled. Claims 21 -22 are new.
Response to Arguments
Applicant’s arguments, see Remarks – Pages 12-13 , filed on 01/26/2026, with respect to the rejections of claims 1, 11, 17 under 103 have been fully considered and are persuasive. Therefore, the rejection has been withdrawn. However, upon further consideration, a new grounds of rejection is made in view of Song.
With regards to 101 rejection.
Applicant relied on his argument is that claim 1 does not recite mental process, a method of organizing human activity or any mathematical relationship, formula or calculation and that the claim is directed to a specific improvement in the functioning of a machine learning model not to an abstract idea– See Remarks pages – 9 -13
Examiner respectfully disagrees. Claim 1 recites (a) initialize a local model…; b)cause the local model to train based on… ; c)calculate the accuracy metric….. ; d) label the local mode as one of valid or invalid e) transmit parameters corresponding to the local model and a flag .. to the server .
The limitation (b) encompasses a person looking at the data collected and performing mathematical calculation on the data (See specification – ¶0030 - training with specified ML algorithm (e.g., linear regression (LR), support vector machine (SVN), stochastic gradient descent (SGD), etc.). this limitation falls within of mental process” and “mathematical concepts” groupings of abstract ideas. The limitation(c) encompasses mathematical concepts (See Specification - ¶Para 0024 - calculating accuracy value using accuracy function). Thus, this limitation recites a concept that falls within the mathematical concept groupings of abstract ideas. The limitation (d) fall within the mental process and organizing human activities groupings of abstract ideas because it covers concepts performed in the human mind or using pen and paper, including evaluation, judgment, and opinion. The limitation (a) & (e ) are mere data gathering and outputting recited at high level of generality , and thus are insignificant extra solution activity. See MPEP 2106.05 (g) ) (“whether the limitation is significant”). In addition, all uses of the recited judicial exceptions require such data gathering and outputting, and, as such, these limitations do not impose any meaningful limits on the claim. These limitations amount to necessary data gathering and outputting. See MPEP 2106.05.
Applicant also relied on his argument is that the claimed apparatus is structured to implement a practical solution to the problems associated with resources consumed when sending, storing and processing model data from many networks by limiting the model data that is communicated and processed to that which is labeled as valid. The Applicant relied on paragraphs 0017-0018,0064 to provide the solutions to the problems. The examiner notes that this information is not shown in claim.
Therefore, the 101 rejection is maintained.
Claim Rejections - 35 USC § 101
35 U.S.C. 101 reads as follows:
Whoever invents or discovers any new and useful process, machine, manufacture, or composition of matter, or any new and useful improvement thereof, may obtain a patent therefor, subject to the conditions and requirements of this title.
Claims 1-22 are rejected under 35 U.S.C. 101 because the claimed invention is directed to an abstract idea without significantly more.
The claims 1, 11,17 recite (a) initialize a local model…; b)cause the local model to train based on… ; c)calculate the accuracy metric….. ; d) label the local mode as one of valid or invalid e) transmit parameters corresponding to the local model and a flag .. to the server
The limitation (b) encompasses a person looking at the data collected and performing mathematical calculation on the data (See specification – ¶0030 - training with specified ML algorithm (e.g., linear regression (LR), support vector machine (SVN), stochastic gradient descent (SGD), etc.). this limitation falls within of mental process” and “mathematical concepts” groupings of abstract ideas.
The limitation(c) encompasses mathematical concepts (See Specification - ¶Para 0024 - calculating accuracy value using accuracy function). Thus, this limitation recites a concept that falls within the mathematical concept groupings of abstract ideas,
The limitation (d) fall within the mental process and organizing human activities groupings of abstract ideas because it covers concepts performed in the human mind or using pen and paper, including evaluation, judgment, and opinion.
The limitation (a) & (e ) are mere data gathering and output recited at high level of generality , and thus are insignificant extra solution activity. See MPEP 2106.05 (g) ) (“whether the limitation is significant”). In addition, all uses of the recited judicial exceptions require such data gathering and outputting, and, as such, these limitations do not impose any meaningful limits on the claim. These limitations amount to necessary data gathering and outputting. See MPEP 2106.05.
This judicial exception is not integrated into a practical application. In particular, the claims 1,11,17 only recite additional elements–an apparatus comprising interface circuit, machine readable instructions and at least one processor circuit as recited in claim 1, and “non transitory medium” and processor circuitry as recited in claim 11.
The interface circuit, processor circuit and medium are recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using interface circuit, processor circuit and medium to perform the limitations (a) - (d) amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claims are not patent eligible.
With regards to claims 2, 12,18, The claims recite “tokenize the client telemetry data” which encompass a person replacing sensitive data with non sensitive data , mathematically generating token. This limitation falls within the mental process and mathematical concepts groupings of abstract ideas.
This judicial exception is not integrated into a practical application. In particular, the claims 2,12 only recite additional elements–processor circuitry.
The processor circuitry is recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using processor circuitry to perform the limitation amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claims are not patent eligible.
With regards to claims 4, 14 ,19, the claims recite “transmit an indication of model invalidity…” The limitation of transmitting is mere data gathering recited at high level of generality and thus are insignificant extra solution activity.
This judicial exception is not integrated into a practical application. In particular, the claims 4,14 only recite additional elements–processor circuit,
The processor circuit is recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using processor circuitry to perform the limitation amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claims are not patent eligible.
With regards to claims 5,15 the claim recites “ discard configuration parameters …and enforcing configuration parameters…”this limitation encompasses a person looking at the data collected and making decision either to discard parameters or enforcing parameters which falls within the mental process and organizing human activity groupings of abstract ideas. Note: the claim does not specify how these parameters are enforced.
This judicial exception is not integrated into a practical application. In particular, the claim 5 only recite additional elements–processor circuit.
The processor circuit is recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using processor circuitry amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claims are not patent eligible.
With regards to claim 6, the claim recites |”…wherein one or more of the at least one processor circuit is to: calculate a second accuracy metric of the local model based on the client-side ground truth data in response to enforcing the configuration parameters; and label the local model as one of valid or invalid based on a comparison between the second accuracy metric and the accuracy threshold”. The calculating limitation encompasses mathematical concepts (calculating accuracy value using accuracy function - See Specification - ¶0024). Thus, this limitation recite a concept that falls within the mathematical concepts grouping of abstract ideas, the enforcing and labeling limitation fall within “the mental process groupings of abstract ideas” and “organizing human activities” grouping of abstract ideas because it covers concepts performed in the human mind or using pen and paper, including evaluation, judgment, and opinion.
This judicial exception is not integrated into a practical application. In particular, the claim 6 only recite additional elements–processor circuit.
The processor circuit is recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using processor circuitry amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
With regards to claim 7 , the claim recites” transmit client telemetry data …” The limitation of transmitting is mere data gathering recited at high level of generality and thus are insignificant extra solution activity.
This judicial exception is not integrated into a practical application. In particular, the claim 7 only recite additional elements–processor circuit,
The processor circuit is recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using processor circuitry amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
With regards to claims 8 ,16, the claims recite” update a quantity of neural network layer …” This limitation encompasses a process based on mathematical calculation. . this limitation falls within the mathematical concept and organizing human activity groupings of abstract ideas
This judicial exception is not integrated into a practical application. In particular, the claims 8,16 only recite additional elements–processor circuit,
The processor circuit is recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using processor circuitry to perform the limitation amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claims are not patent eligible.
With regards to claim 9 , the claim recites” wherein one or more of the at least one processor circuit causes the local model to train with the client telemetry data for a threshold quantity of iteration” the training limitation encompasses a person looking at the data collected and performing mathematical calculation on the data (See specification – ¶0030 - training with specified ML algorithm (e.g., linear regression (LR), support vector machine (SVN), stochastic gradient descent (SGD), etc.). this limitation falls within mental process and mathematical concept groupings of abstract ideas.
This judicial exception is not integrated into a practical application. In particular, the claim 9 only recite additional elements–processor circuitry,
The processor circuit is recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using processor circuitry amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claim is not patent eligible.
With regards to claim 10 , the claim recites “the first modelling plan includes …. “ the limitation just describing what the plan includes.
With regards to claim 20, the claim recites “ discard configuration parameters … enforcing the configuration parameter …. calculating …. and labeling…” the discarding/enforcing limitation encompasses a person looking at the data collected and making decision either to discard parameters or enforcing parameters which falls within the “groupings of mental processor concept” and “grouping of organizing human activity” . The calculating limitation encompasses mathematical concepts (calculating accuracy value using accuracy function -See Specification -¶ 0024). Thus, this limitation recite a concept that falls within “ the mathematical concept groupings of abstract ideas, the labeling limitation fall within “the mental process groupings of abstract ideas” and “organizing human activities” grouping of abstract ideas because it covers concepts performed in the human mind or using pen and paper, including evaluation, judgment, and opinion.
This judicial exception is not integrated into a practical application. In particular, the claim does not recite any additional element.
The claim does not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application. The claim is not patent eligible.
With regards to claims 21-22, the claim recites “initializing the local model with an activation function…”. This limitation is mere data gathering which gathers the data with activation function to perform calculation. The activation function is just mathematical formula, all uses of the recited judicial exceptions require such data gathering to perform the calculation, and, as such, this limitation does not impose any meaningful limits on the claim. This limitation amount to necessary data gathering .
This judicial exception is not integrated into a practical application. In particular, the claims 1,11,17 only recite additional elements–processor circuit as recited in claim 21, and “non transitory medium” and processor circuitry as recited in claim 22.
These additional elements are recited at a high-level of generality such that it amounts no more than mere instructions to apply the exception using a generic computer components.
The claims do not include additional elements that are sufficient to amount to significantly more than the judicial exception. As discussed above with respect to integration of the abstract idea into a practical application, the additional elements of using interface circuit, processor circuit and medium to perform the limitations amounts to no more than mere instructions to apply the exception using a generic computer component. Mere instructions to apply an exception using a generic computer component cannot provide an inventive concept. The claims are not patent eligible.
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 1,4-6,11,14-15,17,19-20 are rejected under 35 U.S.C. 103 as being unpatentable over Fenichel et al. Publication No.US 10,979,422 B1 ( Fenichel hereinafter) in view of Estrada et al. Publication No. US 2017/0220407 A1 ( Estrada hereinafter) further in view of Song et al. Publication No. US 2023/0196121 A1 (Song hereinafter).
Regarding claim 1,
Fenichel teaches an apparatus (Fig. 1), comprising:
interface circuitry; machine-readable instructions; and at least one processor circuit to be programmed by the machine-readable instructions to: initialize a local model with tokenized parameters associated with server telemetry data, the tokenized parameters included in a first modeling plan retrieved from a server (Col.11, lines 55-70,Col.12,lines 1-10 - client system 110 transmits command to verification system 102, the command comprising an instruction to generate a verification model. Receiving tokenized request data may include retrieving tokenized request data from a data storage in response to the command - verification system 102 may receive result data corresponding to tokenized request data, consistent with disclosed embodiments. Result data may indicate whether a tokenized request was properly or improperly granted (or denied) – Col.11, lines 40-55- verification system 102 may receive tokenized request data, consistent with disclosed embodiments. 50 Tokenized request data may include data relating to historical tokenized requests (e.g., verification training data). Tokenized request data may include data relating to a single request (e.g., data that may be used to update an existing verification model - Substantive request data may include information relating to a tokenized request such At step 308, verification system 102 may provide a trained model, consistent with disclosed embodiments. Providing a model may include transmitting a model to client system 110 and/or storing a model in a data storage (e.g., data storage 231, verification database 108, and/or other data storage). Substantive request data
may include information relating to a tokenized request such as a transaction amount, a location, a course, a medical diagnosis and/or any other information.. ).
cause the local model to train based on trigger parameters from the first modeling plan, the local model to train with (a) the tokenized parameters associated with the server telemetry data and (b) client telemetry data (Col.12, lines 10-35 verification system 102 may train a verification model to verify tokenized requests using tokenized request data and result data, For example, trainer 236 may optimize a verification model using request data and result data - a verification model may be trained to generate verification model output. Verification model output may include a binary result associated with the tokenized request (e.g., information specifying whether to grant or deny a tokenization request). Verification model output may include a likelihood or score reflecting a confidence or probability that a tokenization request is illegitimate (i.e., unauthorized – Col.16, lines 1 -25 - verification system 102 may receive verification performance data, Verification performance data may include verification challenges and/or verification. verification performance data may include a statistical result of an analysis of historical request data of a tokenized request and corresponding result data associated with a verification model verification performance data may include a statistical result of an analysis of a plurality of implementations of process 400 (FIG. 4). A statistical result may include any known statistical result, such as a regression result of request and result data, an average, a correlation measure of request data and result data, and/or any other statistical result );
However, Fenichel does not explicitly teach
calculate an accuracy metric of the local model based on client-side ground truth data; and label the local model as one of valid or invalid based on a comparison between the accuracy metric and an accuracy threshold
Estrada teaches
calculate an accuracy metric of the local model based on client-side ground truth data; and label the local model as one of valid or invalid based on a comparison between the accuracy metric and an accuracy threshold (¶0019 - ¶ 0019 -the training phase 100 takes telemetry data ( data collection and transformation operation 106) and uses it to generate a model (model induction operation 108) – ¶ 0024 - Each model takes time series data and predicts the performance of the system. The best model is selected at operation 208. In an embodiment, the simplest model with the highest accuracy is the model that is selected. Model selection may be performed using various methods, If the "best model" available is not good enough, then the flow iterates back to the beginning of the data collection and transformation phase 106, where additional telemetry data may be obtained and analyzed to determine alternative models and select from the alternative models. Once a model is selected, the model, metrics involved, and parameters are stored in a model database 210 – ¶ 0025 - If the training accuracy, measured in percentage, is over a certain predefined threshold, for example, 90% of accuracy, then the model may be declared one that is "good enough" for further evaluation – ¶ 0028 - the model is evaluated against new data and recalibrated when needed. In particular, a data feed is used to obtain new data. When performance data is available, then the model is verified and updated ( operation 306). The difference between the observed value from the data feed (e.g., ground truth or labeled data) and the prediction is computed and used to validate the model's performance A model may be determined as being invalid using a similar or same threshold as used to determine whether a model is "good enough" (e.g., 90% accuracy threshold)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Fenichel in view of Estrada does not explicitly teach
transmit parameters corresponding to the local model and a flag indicative of the local model being valid to the server when the local model is labeled valid.
However, Song teaches
transmit parameters corresponding to the local model and a flag indicative of the local model being valid to the server when the local model is labeled valid (¶ 0137 - Step 104. The first client sends the training result of the current round of training and alarm information to the server – ¶ 0139 - The alarm information existing when the first value of the parameter meets the first condition meets the first condition indicates that the first value of the parameter meets the first condition – ¶ 0140 - There may also be a plurality of forms of the alarm information. This is not specifically limited in this embodiment of this application. For example, the alarm information may indicate a detection result of the first value of the parameter by using 0 and 1. when the value of the alarm information is 1, it indicates that the first value of the parameter does not meet the first condition. When the value of the alarm information is 0, it indicates that the first value of the parameter meets the first condition – Note: the client sends the training results to the server with the alarm information value (0) to flag that the parameter successfully meet the required conditions).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel in view of Estrada to include the teachings of Song. The motivation for doing so is to allow the system to improve robustness of the federated learning system. (Abstract– Song).
Regarding claim 4,
Fenichel does not explicitly teach
wherein one or more of the at least one processor circuit is to transmit an indication of model invalidity when the accuracy threshold is not satisfied, the indication of model invalidity to cause the server to send a second modeling plan.
However, Estrada teaches
wherein one or more of the at least one processor circuit is to transmit an indication of model invalidity when the accuracy threshold is not satisfied, the indication of model invalidity to cause the server to send a second modeling plan (¶ 0039 In an embodiment, to implement the automatic verification operation, the model manager 502 is to compare an observed value from the telemetry data to a predicted value from the performance model and declare the performance model invalid when the observed value deviates from the predicted value by more than a threshold amount. ¶ 0040 - In an embodiment, to initiate the remedial action, the event processor is to monitor for the alert state, retrieve an actuator mechanism when the alert state occurs, and trigger an adaptation at the operational node using the actuator mechanism. the actuator mechanism is a code injection and the adaptation includes injecting executable instructions into an application executing on the operational node. In a related embodiment, the actuator mechanism is a command to a node manager, the command initiating the node manager to conduct the remedial action on the operational node).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Regarding claim 5,
Fenichel does not explicitly teach
wherein one or more of the at least one processor circuit is to discard configuration parameters corresponding to the first modeling plan and enforce configuration parameters corresponding to the second modeling plan.
However, Estrada teaches
wherein one or more of the at least one processor circuit is to discard configuration parameters corresponding to the first modeling plan and enforce configuration parameters corresponding to the second modeling plan (¶ 0032 - Once the actuator mechanism is selected, the command control mechanism is pushed to the application or orchestrator (operation 404) and the system is updated (operation 406). The adaptation phase 104 may take place in-band (e.g., injected into the application) or out-of-band (e.g., via a node manager or orchestrator – Note when updating parameters, the system will discard old parameter and enforce new ones).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Regarding claim 6,
Fenichel does not explicitly teach
wherein the accuracy metric is a first accuracy metric, and wherein one or more of the at least one processor circuit is to: calculate a second accuracy metric of the local model based on the client-side ground truth data in response to enforcing the configuration parameters corresponding to the second modeling plan; and label the local model as one of valid or invalid based on a comparison between the second accuracy metric and the accuracy threshold.
However, Estrada teaches
wherein the accuracy metric is a first accuracy metric, and wherein one or more of the at least one processor circuit is to: calculate a second accuracy metric of the local model based on the client-side ground truth data in response to enforcing the configuration parameters corresponding to the second modeling plan; and label the local model as one of valid or invalid based on a comparison between the second accuracy metric and the accuracy threshold (¶ 0025 - If the training accuracy, measured in percentage, is over a certain predefined threshold, for example, 90% of accuracy, then the model may be declared one that is "good enough" for further evaluation – ¶ 0028 - the model is evaluated against new data and recalibrated when needed. In particular, a data feed is used to obtain new data. When performance data is available, then the model is verified and updated ( operation 306). The difference between the observed value from the data feed (e.g., ground truth or labeled data) and the prediction is computed and used to validate the model's performance A model may be determined as being invalid using a similar or same threshold as used to determine whether a model is "good enough" (e.g., 90% accuracy threshold - ¶ 0039 In an embodiment, to implement the automatic verification operation, the model manager 502 is to compare an observed value from the telemetry data to a predicted value from the performance model and declare the performance model invalid when the observed value deviates from the predicted value by more than a threshold amount. ¶ 0040 - In an embodiment, to initiate the remedial action, the event processor is to monitor for the alert state, retrieve an actuator mechanism when the alert state occurs, and trigger an adaptation at the operational node using the actuator mechanism. the actuator mechanism is a code injection and the adaptation includes injecting executable instructions into an application executing on the operational node. In a related embodiment, the actuator mechanism is a command to a node manager, the command initiating the node manager to conduct the remedial action on the operational node – See Also ¶ 0032)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Regarding claim 11,
Fenichel teaches at least one non-transitory machine readable storage medium comprising instructions that cause processor circuitry (Fig. 1), to at least
initialize a local model with tokenized parameters associated with server telemetry data, the tokenized parameters included in a first modeling plan retrieved from a server (Col.11, lines 55-70,Col.12,lines 1-10 - client system 110 transmits command to verification system 102, the command comprising an instruction to generate a verification model. Receiving tokenized request data may include retrieving tokenized request data from a data storage in response to the command - verification system 102 may receive result data corresponding to tokenized request data, consistent with disclosed embodiments. Result data may indicate whether a tokenized request was properly or improperly granted (or denied) – Col.11, lines 40-55- verification system 102 may receive tokenized request data, consistent with disclosed embodiments. 50 Tokenized request data may include data relating to historical tokenized requests (e.g., verification training data). Tokenized request data may include data relating to a single request (e.g., data that may be used to update an existing verification model - Substantive request data may include information relating to a tokenized request such At step 308, verification system 102 may provide a trained model, consistent with disclosed embodiments. Providing a model may include transmitting a model to client system 110 and/or storing a model in a data storage (e.g., data storage 231, verification database 108, and/or other data storage). Substantive request data may include information relating to a tokenized request such At step 308, verification system 102 may provide a trained model, consistent with disclosed embodiments. Providing a model may include transmitting a model to client system 110 and/or storing a model in a data storage (e.g., data storage 231, verification database 108, and/or other data storage). Substantive request data may include information relating to a tokenized request such as a transaction amount, a location, a course, a medical diagnosis and/or any other information..).
cause the local model to train based on trigger parameters from the first modeling plan, the local model to train with (a) the tokenized parameters associated with the server telemetry data and (b) client telemetry data (Col.12, lines 10-35 verification system 102 may train a verification model to verify tokenized requests using tokenized request data and result data, For example, trainer 236 may optimize a verification model using request data and result data - a verification model may be trained to generate verification model output. Verification model output may include a binary result associated with the tokenized request (e.g., information specifying whether to grant or deny a tokenization request). Verification model output may include a likelihood or score reflecting a confidence or probability that a tokenization request is illegitimate (i.e., unauthorized – Col.16, lines 1 -25 - verification system 102 may receive verification performance data, Verification performance data may include verification challenges and/or verification. verification performance data may include a statistical result of an analysis of historical request data of a tokenized request and corresponding result data associated with a verification model verification performance data may include a statistical result of an analysis of a plurality of implementations of process 400 (FIG. 4). A statistical result may include any known statistical result, such as a regression result of request and result data, an average, a correlation measure of request data and result data, and/or any other statistical result ).;
Fenichel does not explicitly teach
calculate an accuracy metric of the local model based on client-side ground truth data; and label the local model as one of valid or invalid based on a comparison between the accuracy metric and an accuracy threshold
Estrada teaches
calculate an accuracy metric of the local model based on client-side ground truth data; and label the local model as one of valid or invalid based on a comparison between the accuracy metric and an accuracy threshold (¶ 0019 - ¶ 0019 -the training phase 100 takes telemetry data ( data collection and transformation operation 106) and uses it to generate a model (model induction operation 108) – ¶ 0024 - Each model takes time series data and predicts the performance of the system. The best model is selected at operation 208. In an embodiment, the simplest model with the highest accuracy is the model that is selected. Model selection may be performed using various methods, If the "best model" available is not good enough, then the flow iterates back to the beginning of the data collection and transformation phase 106, where additional telemetry data may be obtained and analyzed to determine alternative models and select from the alternative models. Once a model is selected, the model, metrics involved, and parameters are stored in a model database 210 – ¶ 0025 - If the training accuracy, measured in percentage, is over a certain predefined threshold, for example, 90% of accuracy, then the model may be declared one that is "good enough" for further evaluation – ¶ 0028 - the model is evaluated against new data and recalibrated when needed. In particular, a data feed is used to obtain new data. When performance data is available, then the model is verified and updated ( operation 306). The difference between the observed value from the data feed (e.g., ground truth or labeled data) and the prediction is computed and used to validate the model's performance A model may be determined as being invalid using a similar or same threshold as used to determine whether a model is "good enough" (e.g., 90% accuracy threshold)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Fenichel in view of Estrada does not explicitly teach
transmit parameters corresponding to the local model and a flag indicative of the local model being valid to the server when the local model is labeled valid.
However, Song teaches
transmit parameters corresponding to the local model and a flag indicative of the local model being valid to the server when the local model is labeled valid (¶ 0137 - Step 104. The first client sends the training result of the current round of training and alarm information to the server – ¶ 0139 - The alarm information existing when the first value of the parameter meets the first condition meets the first condition indicates that the first value of the parameter meets the first condition – ¶ 0140 - There may also be a plurality of forms of the alarm information. This is not specifically limited in this embodiment of this application. For example, the alarm information may indicate a detection result of the first value of the parameter by using 0 and 1. when the value of the alarm information is 1, it indicates that the first value of the parameter does not meet the first condition. When the value of the alarm information is 0, it indicates that the first value of the parameter meets the first condition – Note: the client sends the training results to the server with the alarm information value (0) to flag that the parameter successfully meet the required conditions).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel in view of Estrada to include the teachings of Song. The motivation for doing so is to allow the system to improve robustness of the federated learning system. (Abstract– Song).
Regarding claim 14,
Fenichel does not explicitly teach
wherein the instructions, when executed, cause the processor circuitry to transmit an indication of model invalidity when the accuracy threshold is not satisfied, the indication of model invalidity to cause the server to send a second modeling plan.
However, Estrada teaches
wherein the instructions, when executed, cause the processor circuitry to transmit an indication of model invalidity when the accuracy threshold is not satisfied, the indication of model invalidity to cause the server to send a second modeling plan (¶ 0039 In an embodiment, to implement the automatic verification operation, the model manager 502 is to compare an observed value from the telemetry data to a predicted value from the performance model and declare the performance model invalid when the observed value deviates from the predicted value by more than a threshold amount. ¶ 0040 - In an embodiment, to initiate the remedial action, the event processor is to monitor for the alert state, retrieve an actuator mechanism when the alert state occurs, and trigger an adaptation at the operational node using the actuator mechanism. the actuator mechanism is a code injection and the adaptation includes injecting executable instructions into an application executing on the operational node. In a related embodiment, the actuator mechanism is a command to a node manager, the command initiating the node manager to conduct the remedial action on the operational node).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Regarding claim 15,
Fenichel does not explicitly teach
wherein the instructions, when executed, cause the processor circuitry to discard configuration parameters corresponding to the first modeling plan and enforce configuration parameters corresponding to the second modeling plan.
However, Estrada teaches
wherein the instructions, when executed, cause the processor circuitry to discard configuration parameters corresponding to the first modeling plan and enforce configuration parameters corresponding to the second modeling plan (¶ 0032 - Once the actuator mechanism is selected, the command control mechanism is pushed to the application or orchestrator (operation 404) and the system is updated (operation 406). The adaptation phase 104 may take place in-band (e.g., injected into the application) or out-of-band (e.g., via a node manager or orchestrator).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Regarding claim 17,
Fenichel teaches a method to update a global model, comprising
initializing a local model with tokenized parameters associated with server telemetry data, the tokenized parameters included in a first modeling plan retrieved from a server (Col.11, lines 55-70,Col.12,lines 1-10 - client system 110 transmits command to verification system 102, the command comprising an instruction to generate a verification model. Receiving tokenized request data may include retrieving tokenized request data from a data storage in response to the command - verification system 102 may receive result data corresponding to tokenized request data, consistent with disclosed embodiments. Result data may indicate whether a tokenized request was properly or improperly granted (or denied) – Col.11, lines 40-55- verification system 102 may receive tokenized request data, consistent with disclosed embodiments. 50 Tokenized request data may include data relating to historical tokenized requests (e.g., verification training data). Tokenized request data may include data relating to a single request (e.g., data that may be used to update an existing verification model - Substantive request data may include information relating to a tokenized request such At step 308, verification system 102 may provide a trained model, consistent with disclosed embodiments. Providing a model may include transmitting a model to client system 110 and/or storing a model in a data storage (e.g., data storage 231, verification database 108, and/or other data storage). as a transaction amount, a location, a course, a medical diagnosis and/or any other information. ).
causing the local model to train based on trigger parameters from the first modeling plan, the local model to train with (a) the tokenized parameters associated with the server telemetry data and (b) client telemetry data (Col.12, lines 10-35 verification system 102 may train a verification model to verify tokenized requests using tokenized request data and result data, For example, trainer 236 may optimize a verification model using request data and result data - a verification model may be trained to generate verification model output. Verification model output may include a binary result associated with the tokenized request (e.g., information specifying whether to grant or deny a tokenization request). Verification model output may include a likelihood or score reflecting a confidence or probability that a tokenization request is illegitimate (i.e., unauthorized – Col.16, lines 1 -25 - verification system 102 may receive verification performance data, Verification performance data may include verification challenges and/or verification. verification performance data may include a statistical result of an analysis of historical request data of a tokenized request and corresponding result data associated with a verification model verification performance data may include a statistical result of an analysis of a plurality of implementations of process 400 (FIG. 4). A statistical result may include any known statistical result, such as a regression result of request and result data, an average, a correlation measure of request data and result data, and/or any other statistical result ).;
However, Fenichel does not explicitly teach
calculating an accuracy metric of the local model based on client-side ground truth data; and labeling the local model as one of valid or invalid based on a comparison between the accuracy metric and an accuracy threshold
Estrada teaches
calculating an accuracy metric of the local model based on client-side ground truth data; and labeling the local model as one of valid or invalid based on a comparison between the accuracy metric and an accuracy threshold (¶ 0019 - ¶ 0019 -the training phase 100 takes telemetry data ( data collection and transformation operation 106) and uses it to generate a model (model induction operation 108) – ¶ 0024 - Each model takes time series data and predicts the performance of the system. The best model is selected at operation 208. In an embodiment, the simplest model with the highest accuracy is the model that is selected. Model selection may be performed using various methods, If the "best model" available is not good enough, then the flow iterates back to the beginning of the data collection and transformation phase 106, where additional telemetry data may be obtained and analyzed to determine alternative models and select from the alternative models. Once a model is selected, the model, metrics involved, and parameters are stored in a model database 210 – ¶ 0025 - If the training accuracy, measured in percentage, is over a certain predefined threshold, for example, 90% of accuracy, then the model may be declared one that is "good enough" for further evaluation – ¶ 0028 - the model is evaluated against new data and recalibrated when needed. In particular, a data feed is used to obtain new data. When performance data is available, then the model is verified and updated ( operation 306). The difference between the observed value from the data feed (e.g., ground truth or labeled data) and the prediction is computed and used to validate the model's performance A model may be determined as being invalid using a similar or same threshold as used to determine whether a model is "good enough" (e.g., 90% accuracy threshold).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Fenichel in view of Estrada does not explicitly teach
transmit parameters corresponding to the local model and a flag indicative of the local model being valid to the server when the local model is labeled valid.
However, Song teaches
transmit parameters corresponding to the local model and a flag indicative of the local model being valid to the server when the local model is labeled valid (¶ 0137 - Step 104. The first client sends the training result of the current round of training and alarm information to the server – ¶ 0139 - The alarm information existing when the first value of the parameter meets the first condition meets the first condition indicates that the first value of the parameter meets the first condition – ¶ 0140 - There may also be a plurality of forms of the alarm information. This is not specifically limited in this embodiment of this application. For example, the alarm information may indicate a detection result of the first value of the parameter by using 0 and 1. when the value of the alarm information is 1, it indicates that the first value of the parameter does not meet the first condition. When the value of the alarm information is 0, it indicates that the first value of the parameter meets the first condition – Note: the client sends the training results to the server with the alarm information value (0) to flag that the parameter successfully meet the required conditions).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel in view of Estrada to include the teachings of Song. The motivation for doing so is to allow the system to improve robustness of the federated learning system. (Abstract– Song).
Regarding claim 19,
Fenichel does not explicitly teach
transmit an indication of model invalidity when the accuracy threshold is not satisfied, the indication of model invalidity to cause the server to send a second modeling plan.
However, Estrada teaches
wherein one or more of the at least one processor circuit is to transmit an indication of model invalidity when the accuracy threshold is not satisfied, the indication of model invalidity to cause the server to send a second modeling plan (¶ 0039 In an embodiment, to implement the automatic verification operation, the model manager 502 is to compare an observed value from the telemetry data to a predicted value from the performance model and declare the performance model invalid when the observed value deviates from the predicted value by more than a threshold amount. ¶ 0040 - In an embodiment, to initiate the remedial action, the event processor is to monitor for the alert state, retrieve an actuator mechanism when the alert state occurs, and trigger an adaptation at the operational node using the actuator mechanism. the actuator mechanism is a code injection and the adaptation includes injecting executable instructions into an application executing on the operational node. In a related embodiment, the actuator mechanism is a command to a node manager, the command initiating the node manager to conduct the remedial action on the operational node).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Regarding claim 20,
Fenichel does not explicitly teach
discarding configuration parameters corresponding to the first modeling plan; enforcing configuration parameters corresponding to the second modeling plan; calculating a second accuracy metric of the local model based on the client-side ground truth data in response to enforcing the configuration parameters corresponding to the second modeling plan; and labeling the local model as one of valid or invalid based on a comparison between the second accuracy metric and the accuracy threshold.
However, Estrada teaches
discarding configuration parameters corresponding to the first modeling plan; enforcing configuration parameters corresponding to the second modeling plan; calculating a second accuracy metric of the local model based on the client-side ground truth data in response to enforcing the configuration parameters corresponding to the second modeling plan; and labeling the local model as one of valid or invalid based on a comparison between the second accuracy metric and the accuracy threshold (¶ 0032 - Once the actuator mechanism is selected, the command control mechanism is pushed to the application or orchestrator (operation 404) and the system is updated (operation 406). The adaptation phase 104 may take place in-band (e.g., injected into the application) or out-of-band (e.g., via a node manager or orchestrator).¶ 0025 - If the training accuracy, measured in percentage, is over a certain predefined threshold, for example, 90% of accuracy, then the model may be declared one that is "good enough" for further evaluation – ¶ 0028 - the model is evaluated against new data and recalibrated when needed. In particular, a data feed is used to obtain new data. When performance data is available, then the model is verified and updated ( operation 306). The difference between the observed value from the data feed (e.g., ground truth or labeled data) and the prediction is computed and used to validate the model's performance A model may be determined as being invalid using a similar or same threshold as used to determine whether a model is "good enough" (e.g., 90% accuracy threshold - ¶ 0039 In an embodiment, to implement the automatic verification operation, the model manager 502 is to compare an observed value from the telemetry data to a predicted value from the performance model and declare the performance model invalid when the observed value deviates from the predicted value by more than a threshold amount. ¶ 0040 - In an embodiment, to initiate the remedial action, the event processor is to monitor for the alert state, retrieve an actuator mechanism when the alert state occurs, and trigger an adaptation at the operational node using the actuator mechanism. the actuator mechanism is a code injection and the adaptation includes injecting executable instructions into an application executing on the operational node. In a related embodiment, the actuator mechanism is a command to a node manager, the command initiating the node manager to conduct the remedial action on the operational node – See Also ¶ 0032)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Estrada. The motivation for doing so is to allow the system to keep track of performance issues and provide a way to retrain the model when the prediction accuracy deteriorates (¶ 0016 – Estrada).
Claims 2,12,18 are rejected under 35 U.S.C. 103 as being unpatentable over Fenichel in view of Estrada further in view of Song further in view of Fiumara et al. Publication No. US 2022/0150132 A1 ( Fiumara hereinafter).
Regarding claim 2,
Fenichel does not explicitly teach
wherein one or more of the at least one processor circuit is to tokenize the client telemetry data corresponding to the local model in response to labeling the local model as valid.
However, Fiumara teaches
wherein one or more of the at least one processor circuit is to tokenize the client telemetry data corresponding to the local model in response to labeling the local model as valid (Abstract - The device may select an optimum machine learning model based on several indicators, such as the accuracy, precision, and/or the like. The device may provide the optimum machine learning model to the monitoring device associated with the telecommunications network. The optimum machine learning model may cause the monitoring device to process real time telecommunications data of the telecommunications network, with the optimum machine learning model, to determine a customer care action, and may cause the customer care action to be implemented in the telecommunications network ¶ 0029 – ¶ 0030 - the monitoring device processes the telecommunications data based on a natural language processing (NLP) technique. the monitoring device anonymizes the historical telecommunications data. The customer care system may anonymize the historical telecommunications
data to remove data identifying a telecommunications network provider associated with the telecommunications network).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Fiumara. The motivation for doing so is to allow the system to tokenize/anonymize data in order to protect privacy.
Regarding claim 12,
Fenichel does not explicitly teach
wherein the instructions, when executed, cause the processor circuitry to tokenize the client telemetry data corresponding to the local model in response to labeling the local model as valid.
However, Fiumara teaches
wherein the instructions, when executed, cause the processor circuitry to tokenize the client telemetry data corresponding to the local model in response to labeling the local model as valid. (Abstract - The device may select an optimum machine learning model based on several indicators, such as the accuracy, precision, and/or the like. The device may provide the optimum machine learning model to the monitoring device associated with the telecommunications network. The optimum machine learning model may cause the monitoring device to process real time telecommunications data of the telecommunications network, with the optimum machine learning model, to determine a customer care action, and may cause the customer care action to be implemented in the telecommunications network ¶ 0029 – ¶ 0030 - the monitoring device processes the telecommunications data based on a natural language processing (NLP) technique. the monitoring device anonymizes the historical telecommunications data. The customer care system may anonymize the historical telecommunications data to remove data identifying a telecommunications network provider associated with the telecommunications network).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Fiumara. The motivation for doing so is to allow the system to tokenize/anonymize data in order to protect privacy.
Regarding claim 18,
Fenichel does not explicitly teach
to tokenize the client telemetry data corresponding to the local model in response to labeling the local model as valid.
However, Fiumara teaches
to tokenize the client telemetry data corresponding to the local model in response to labeling the local model as valid (Abstract - The device may select an optimum machine learning model based on several indicators, such as the accuracy, precision, and/or the like. The device may provide the optimum machine learning model to the monitoring device associated with the telecommunications network. The optimum machine learning model may cause the monitoring device to process real time telecommunications data of the telecommunications network, with the optimum machine learning model, to determine a customer care action, and may cause the customer care action to be implemented in the telecommunications network ¶ 0029 – ¶ 0030 - the monitoring device processes the telecommunications data based on a natural language processing (NLP) technique. the monitoring device anonymizes the historical telecommunications data. The customer care system may anonymize the historical telecommunications
data to remove data identifying a telecommunications network provider associated with the telecommunications network).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Fiumara. The motivation for doing so is to allow the system to tokenize/anonymize data in order to protect privacy.
Claim 7 is rejected under 35 U.S.C. 103 as being unpatentable over Fenichel in view of Estrada further in view of Song further in view of Reinecke et al Publication No. US 2018/0004958 A1 ( Reinecke hereinafter)
Regarding claim 7,
Fenichel does not explicitly teach
wherein one or more of the at least one processor circuit is to transmit the client telemetry data to the server with a label indicating that the client telemetry data is malicious when the accuracy threshold is not satisfied, and wherein the server utilizes the client telemetry data for model training.
However, Reinecke teaches
wherein one or more of the at least one processor circuit is to transmit the client telemetry data to the server with a label indicating that the client telemetry data is malicious when the accuracy threshold is not satisfied, and wherein the server utilizes the client telemetry data for model training (¶ 0025 – ¶ 0026 The hardware processor 120 executes instructions 136 to update the first set of attack models based on the performance data 152.. the computing device 110 provides the attack model storage 140 device with an updated Set S' 144. In some implementations, the set of attack models is updated by adding, removing, or changing an attack model in the first set. For example, if one of the three example data exfiltration attack models performs significantly worse than the other two, e.g., takes longer, doesn't often successfully identify attacks – ¶ 0028 - The attack model management device 410 obtains performance data 414 from the attack model performance data 430. As noted above, the performance data 414 may include a variety of performance related information for each of the attack models in the set S 412. The example data flow 400 depicts performance data 414 as including information such as attack model execution time, processor usage storage usage, and successful attack detection statistics. For attack model A, these are represented by T(A), P(A), S(A), and a binary set that indicates whether the attack model was successful (1) or not (0) for each use of the attack model in detecting an attack on a computing system – ¶ 0039 - The attack model management device 410 updates the attack model set S 412 using the performance data 414. The resulting attack model set S').
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Reinecke. The motivation for doing so is to allow the system to determine whether the particular attack occurred on the computing system and update the first set of attack models based on the performance data (Abstract – Reinecke).
Claims 8,16 are rejected under 35 U.S.C. 103 as being unpatentable over Fenichel in view of Estrada further in view of Song further in view of Walters et al. Publication No.US 2020/0012584 A1 ( Walters hereinafter).
Regarding claim 8,
Fenichel does not explicitly teach
wherein one or more of the at least one processor circuit is to update a quantity of neural network layers in the local model based on the modeling plan retrieved from the server.
However, Walters teaches
wherein one or more of the at least one processor circuit is to update a quantity of neural network layers in the local model based on the modeling plan retrieved from the server ( ¶ 0044, ¶ 0193 – ¶ 0194 -model optimizer 107 receives a command from interface 113. In some embodiments, the command is based on user inputs (i.e., debugging may be a supervised process). The command may comprise a training condition to be used for machine learning. The training condition may relate to the performance of the model during model training. For example, the training condition may involve comparing an accuracy score of the model to a predetermined threshold or tracking a rate of improvement
of an accuracy score. the command may include a at least one of a command to adjust the model, to train the model. adjusting the model may include at least one
of changing the number of layers in a neural network),
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Walters. The motivation for doing so is to allow the system to optimize the model ( Abstract – Walters) .
Regarding claim 16,
Fenichel does not explicitly teach
wherein the instructions, when executed, cause the processor circuitry to update a quantity of neural network layers in the local model based on the modeling plan retrieved from the server.
However, Walters teaches
wherein one or more of the at least one processor circuit is to update a quantity of neural network layers in the local model based on the modeling plan retrieved from the server ( ¶ 0044, ¶ 0193 – ¶ 0194 -model optimizer 107 receives a command from interface 113. In some embodiments, the command is based on user inputs (i.e., debugging may be a supervised process). The command may comprise a training condition to be used for machine learning. The training condition may relate to the performance of the model during model training. For example, the training condition may involve comparing an accuracy score of the model to a predetermined threshold or tracking a rate of improvement
of an accuracy score. the command may include a at least one of a command to adjust the model, to train the model. adjusting the model may include at least one
of changing the number of layers in a neural network),
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Walters. The motivation for doing so is to allow the system to optimize the model ( Abstract – Walters) .
Claim 9 is rejected under 35 U.S.C. 103 as being unpatentable over Fenichel in view of Estrada further in view of Song further in view of Chu et al. Publication No. US 2022/0156589 A1 ( Chu hereinafter).
Regarding claim 9,
Fenichel does not explicitly teach
wherein one or more of the at least one processor circuit causes the local model to train with the client telemetry data for a threshold quantity of iterations
However, Chu teaches
wherein one or more of the at least one processor circuit causes the local model to train with the client telemetry data for a threshold quantity of iterations (¶ 0104 - The model trainer 130 can first train the main model 119 to perform the machine learning task, and prune sources of sensor data that make up the sensor input 116. For example, the model trainer 130 can train the main model 119 on training data that includes sensor data from two voice accelerometers of the sensors 113. The model trainer 130 can continue this process of pruning and evaluation for predetermined number of iterations, or until model performance falls below the minimum accuracy threshold).
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Chu . The motivation for doing so is to allow the system to improve the overall accuracy of the system ( ¶ 0143 – Chu).
Claim 10 is rejected under 35 U.S.C. 103 as being unpatentable over Fenichel in view of Estrada further in view of Song further in view of Bendre et al. Publication No.US 2018/0322415 A1 ( Bendre hereinafter).
Regarding claim 10,
Fenichel does not explicitly teach
wherein the first modeling plan includes a schedule for at least one of updating the local model, training the local model, or transmitting model updates to the server.
However, Bendre teaches
wherein the first modeling plan includes a schedule for at least one of updating the local model, training the local model, or transmitting model updates to the server (¶ 0140, ¶ 0142 - the solution definition 630 could specify a periodic training schedule. For instance, the solution definition 630 could specify that the ML model should be updated once per day. As a result, the scheduler device 604 could periodically receive ML training requests based on the solution definition 630 and could assign ML trainer process(es) to respectively serve those ML training requests according to the periodic training schedule, so as to periodically update the ML model. Other examples are also possible.),
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Bendre . The motivation for doing so is to allow the system to specify that the ML model should be updated once per day ( ¶ 0142 – Bendre).
Claims 21-22 are rejected under 35 U.S.C. 103 as being unpatentable over Fenichel in view of Estrada further in view of Song further in view of Cruz Mota et al. Publication No. US 2015/0193695 A1 ( Cruz Mota hereinafter).
Regarding claim 21,
Fenichel further teaches wherein one or more of the at least one processor circuit is to initialize the local model with data [..] in the first modeling plan retrieved from the server (Col.11, lines 55-70,Col.12,lines 1-10, Col.11, lines 40-55). However, Fenichel does not explicitly teach that the data is activation function.
Cruz Mota teaches
initialize the local model with activation function received from a server (¶ 0125 - After communicating all the sequence information to each TE, the controller/NMS sends a Collect Step message, which is a newly defined IPv6 unicast message, to the first TE. This message may contain any or all of the following TLVs: The activation functions of the neurons; the NMS may send the above model parameters to the first TE selected to perform the collective training. a network device may receive ANN weights, activation functions, etc. of the model to be trained. In step 815, the received model parameters are applied to a local set of data to generate new model parameters.)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Cruz Mota . The motivation for doing so is to allow the system to determine whether a neuron fires or passes information forward in a neural network
Regarding claim 22,
Fenichel further teaches wherein the instructions, when executed, cause the processor circuitry to initialize the local model with data [..] in the first modeling plan retrieved from the server (Col.11, lines 55-70,Col.12,lines 1-10, Col.11, lines 40-55). However, Fenichel does not explicitly teach that the data is activation function.
Cruz Mota teaches
initialize the local model with activation function received from a server (¶ 0125 - After communicating all the sequence information to each TE, the controller/NMS sends a Collect Step message, which is a newly defined IPv6 unicast message, to the first TE. This message may contain any or all of the following TLVs: The activation functions of the neurons; the NMS may send the above model parameters to the first TE selected to perform the collective training. a network device may receive ANN weights, activation functions, etc. of the model to be trained. In step 815, the received model parameters are applied to a local set of data to generate new model parameters.)
It would have been obvious to a person of ordinary skill in the art before the effective filing date of the claimed invention to modify the teachings of Fenichel to include the teachings of Cruz Mota . The motivation for doing so is to allow the system to determine whether a neuron fires or passes information forward in a neural network
Conclusion
Applicant's amendment necessitated the new ground(s) of rejection presented in this Office action. Accordingly, THIS ACTION IS MADE FINAL. See MPEP § 706.07(a). Applicant is reminded of the extension of time policy as set forth in 37 CFR 1.136(a).
A shortened statutory period for reply to this final action is set to expire THREE MONTHS from the mailing date of this action. In the event a first reply is filed within TWO MONTHS of the mailing date of this final action and the advisory action is not mailed until after the end of the THREE-MONTH shortened statutory period, then the shortened statutory period will expire on the date the advisory action is mailed, and any nonprovisional extension fee (37 CFR 1.17(a)) pursuant to 37 CFR 1.136(a) will be calculated from the mailing date of the advisory action. In no event, however, will the statutory period for reply expire later than SIX MONTHS from the mailing date of this final action.
Any inquiry concerning this communication or earlier communications from the examiner should be directed to YOUNES NAJI whose telephone number is (571)272-2659. The examiner can normally be reached Monday - Friday 8:30 AM -5:30 PM.
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, Oscar A Louie can be reached at (571) 270-1684. 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.
/YOUNES NAJI/Primary Examiner, Art Unit 2445
Prosecution Insights