Notice of Pre-AIA or AIA Status
1. The present application, filed on or after March 16, 2013, is being examined under the first inventor to file provisions of the AIA .
DETAILED ACTION
2. This Office Action is sent in response to Applicant’s Communication received on 06/05/2026 for application number 18/462,717.
Response to Amendments
3. The Amendment filed 06/05/2026 has been entered. Claims 1, 4, 6, 9, 12-16, and 18-20 have been amended. Claims 8 and 17 have been canceled. Claims 21 and 22 have been added. Claims 1-7, 9-16, and 18-22 remain pending in the application.
4. Applicant’s amendment to the claim 4 has been fully considered and is persuasive. The amendment provided to overcome the 101 rejection issued in the last office action is sufficient. The 35 U.S.C § 101 rejection of claims 4-7, 9-16, 18-22 is respectfully withdrawn.
Response to Arguments
Applicant argues that the cited refs do not disclose one or more features recited in the amended independent claims. However, the argument is moot since this is a newly presented limitation, thus changes the scope of the claims. However, newly found references, Xu and Movshovitz, are applied.
Claim Rejections – 35 USC § 103
5. The following is a quotation of 35 U.S.C. 103 which forms the basis for all obviousness rejections set forth in this Office action:
A patent for a claimed invention may not be obtained, notwithstanding that the claimed invention is not identically disclosed as set forth in section 102 of this title, if the differences between the claimed invention and the prior art are such that the claimed invention as a whole would have been obvious before the effective filing date of the claimed invention to a person having ordinary skill in the art to which the claimed invention pertains. Patentability shall not be negated by the manner in which the invention was made.
6. Claim 1 is rejected under 35 U.S.C. 103 as being unpatentable over Sheller et al. (U.S. Patent Application Pub. No. US 20190042937 A1) in view of Walters et al. (U.S. Patent Application Pub. No. US 20200012900 A1), in view of Zoldi et al. (U.S. Patent Application Pub. No. US 20160342963 A1), and further in view of Xu et al. (U.S. Patent Application Pub. No. US 20230368069 A1).
Claim 1: Sheller teaches a system for detecting anomalous data updates at a client device in federated learning, comprising the client device including (i.e. Federated Learning enables a model representing a neural network to be trained using data across many edge systems without having to centralize the data used for such training. Edge devices perform local training, and provide training results to an aggregator device, which aggregates the training results among the multiple edge devices to update a centralized model, which can then be re-distributed to the edge devices for subsequent training and/or use; para. [0015, 0119]):
one or more processors (i.e. a method for federated training of a neural network, the method comprising determining, by executing an instruction with a processor of an edge device; para. [0119]); and
a non-transitory, computer-readable medium storing instructions that, when executed by the one or more processors (i.e. at least one tangible machine readable storage medium comprising instructions which, when executed, cause at least one processor of an edge device to; para. [0113]), cause operations comprising:
in response to receiving a first data update associated with a first interval of time and a second data update associated with a second interval of time (i.e. determines whether to allow edge devices to incorporate new training data in a given training round, and instructs the edge devices 130, 137 concerning the use of new training data for a given training round. In examples disclosed herein, the example training data instructor 260 allows new training data to be used every N rounds. However, any other approach to selecting when new training data will be allowed may additionally or alternatively be used; para. [0033], repeated addition of new local data over different rounds),
determining, based on the anomaly score (i.e. the training data may be compared to previously provided training data to determine a degree of similarity to the prior training data. If the newly provided training data is similar to previously submitted training data, such similarity suggests that the training data is legitimate (as training data is not expected to widely vary from one training iteration to the next). On the contrary, if the training data is not similar to previously submitted training data, such non-similarity suggests that the training data may have been tampered with in an effort to maliciously impact the model; para. [0077]), whether to include the third data update into training data for a local model to be trained at the client device (i.e. determining, by executing an instruction with a processor of an edge device, whether to allow a new local data item to be incorporated into a training process of a neural network implemented at the edge device; para. [0119, 0149]); and
in response to determining to include the third data update: training, based on the third data update, the local model to generate a set of model weights (i.e. training the neural network to create a model update using local data items, the local data items having hashes stored in the hash ledger; para. [0119]), if local data is permitted, the local neural network is trained and a model update is created; and
in response to receiving, from a central node, a request for the set of model weights, transmitting the set of model weights to the central node for federated learning (i.e. means for providing the model update to the aggregator device; para. [0148]), the aggregator provides model parameters/instructions, the edge device trains locally, and then provides the model update to the aggregator.
Sheller does not explicitly teach generating a first data profile based on the first data update and a second data profile based on the second data update, wherein each data profile comprises descriptive statistics regarding a data update comprising one or more of distributions of features in the data update, a frequency of null values, or a covariance matrix between features in the data update; processing, using an extrapolation model that is configured to output predicted values for a set of features, the first data profile and the second data profile to generate a data profile trend, wherein the data profile trend comprises an expectation value that includes the predicted values and a measure of variance; in response to receiving a third data update associated with a third interval of time subsequent to the first interval of time and the second interval of time, generating a third data profile based on the third data update; determining a measure of deviation for the third data profile based on the expectation value of the data profile trend and the third data profile, wherein the measure of deviation is indicative of a difference between the third data profile and the expectation value; processing the measure of deviation and the measure of variance using a prediction model to generate an anomaly score.
However, Walters teaches in response to receiving a first data update associated with a first interval of time and a second data update associated with a second interval of time (i.e. the predictive model may be implemented daily based on received data to generate generated values at five-minute increments over the next week (e.g., stock forecast, weather forecast, etc.), consistent with disclosed embodiments. In some embodiments, step 1808 is performed a plurality of times over a plurality of different model input datasets; para. [0182, 0183]), repeated datasets received on schedules, generating a first data profile based on the first data update (i.e. a baseline data metric of the data profile of the baseline synthetic data is determined, the data profile including the data schema and a statistical profile; para. [0191]) and a second data profile based on the second data update (i.e. a current data metric of a data profile of the current synthetic data is determined; para. [0194]), wherein each data profile comprises descriptive statistics regarding a data update comprising one or more of distributions of features in the data update, a frequency of null values, or a covariance matrix between features in the data update (i.e. the data profile including the data schema and a statistical profile. For example, step 1906 may include determining a baseline covariance matrix of the baseline synthetic data. In some embodiments, the data metric may include a data distribution, a vocabulary (character, word, or the like), a data category, a min, a max, a variance, a quartile, a quantile, a median, a range, or other data metric; para. [0191, 0194]);
processing, using an extrapolation model that is configured to output predicted values (i.e. At step 1808, the predictive model generates predicted data based on the model input data; para. [0182]), the first data profile and the second data profile to generate a data profile trend (i.e. At step 1812, data drift is detected. In some embodiments, detecting data drift is a based on a comparison of predicted data to event data to determine a difference between predicted data and event data. In the embodiments, detecting data drift may be based on known statistical methods. For example, detecting data drift at step 1812 may be based on at least one of a least squares error method, a regression method, a correlation method, or other known statistical method; para. [0184]), wherein the data profile trend comprises an expectation value that includes the predicted values (i.e. Step 1810 includes receiving event data, the event data being of the same category as the predicted data. Event data may be the data that corresponds to the real-world data that the predictive model was used to predict; para. [0183]) and a measure of variance (i.e. the duration and/or periodicity of the schedule may be based on one or more aspects of the disclosed embodiments, such as a characteristic of the training data, predicted data, or event data (e.g., variance, sampling rate, detecting a measured value falls inside or outside a particular range or above or below particular threshold, etc.), detecting a correction in the predictive model, or other such feature; para. [0182, 0191, 0192]);
in response to receiving a third data update associated with a third interval of time subsequent to the first interval of time and the second interval of time, generating a third data profile based on the third data update (i.e. At step 1910, current input data are received, the current input data having the same category as the model input data. Current input data may be entirely composed of actual data, entirely composed of synthetic data, or include a mix of synthetic data and actual data. Step 1910 may be repeated a plurality of times, as shown in FIG. 19; para. [0192]), repeated later/current datasets and corresponding data profiles/metrics;
determining a measure of deviation for the third data profile based on the expectation value of the data profile trend and the third data profile, wherein the measure of deviation is indicative of a difference between the third data profile and the expectation value (i.e. detecting data drift is a based on a comparison of predicted data to event data to determine a difference between predicted data and event data. In the embodiments, detecting data drift may be based on known statistical methods. For example, detecting data drift at step 1812 may be based on at least one of a least squares error method, a regression method, a correlation method, or other known statistical method. In some embodiments, detecting a difference between predicted data and event data includes determining whether a difference between generated data and event data meets or exceeds a threshold difference. In some embodiments, detecting data drift includes determining a difference between the data profile of the predicted data and the data profile of the event data. For example, drift may be detected based on a difference between the covariance matrix of the predicted data and a covariance matrix of the event data; para. [0184]);
processing the measure of deviation and the measure of variance using a prediction model (i.e. detecting data drift is a based on a comparison of predicted data to event data to determine a difference between predicted data and event data. In the embodiments, detecting data drift may be based on known statistical methods. For example, detecting data drift at step 1812 may be based on at least one of a least squares error method, a regression method, a correlation method, or other known statistical method. In some embodiments, the difference is determined using at least one of a Mean Absolute Error, a Root Mean Squared Error, a percent good classification, or the like; para. [0184]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of Sheller to include the feature of Walters. One would have been motivated to make this modification because it improves the reliability and quality of newly received local training data before that data is incorporated into the local training process.
However, Zoldi teaches processing the measure of deviation and the measure of variance using a prediction model to generate an anomaly score (i.e. calculate the mean and variance of the fraud probability, calculate transaction transversal pathway distributions of the model development data and the new transaction data, calculate metrics such as deviation, correlation and K-L distance between the transversal pathway distributions of the developed model and the new transaction dataset; determine a degree of anomaly based on these three metrics and generating an alert according to a predetermined criterion; para. [0018, 0027]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller and Walters to include the feature of Zoldi. One would have been motivated to make this modification because it improves the quality and trustworthiness of the local data used to generate the local model.
However, Xu teaches processing, using an extrapolation model (i.e. The time series predictor 202 applies a regression network 226 to the normalized space representing the set of historical feature values 210; para. [0051]) that is configured to output predicted values (i.e. The anomaly detector 204 can receive the current feature value 220 and a set of predicted values 222 that includes the upper boundary 216, the expected value 214, and the lower boundary 218; para. [0055]) for a set of features (i.e. The time series predictor 160 predicts an expected value and an expected range of values for each feature of the online system … the time series predictor is a multilayer perceptron model that receives multiple time series features such as user clicks, message counts, content posts, document counts, activities by a user interacting with an online system, or other time series features; para. [0037, 0072]), the data profile and the data profile to generate a data profile trend (i.e. the time series predictor 160 applies a seasonality adaption layer 206 that encodes feature cycles (e.g., repeat values or repeating patterns such as a weekly trend from high to low) as features for inferring patterns of each feature within the normalized space; para. [0052]), wherein the data profile trend comprises: an expectation value that includes the predicted values (i.e. The time series predictor 202 outputs the expected value 214, upper boundary 216, and the lower boundary 218; para. [0052]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, and Zoldi to include the feature of Xu. One would have been motivated to make this modification because it improves early identification of anomalous training data.
7. Claim 2 is rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Xu, and further in view of Elsakhawy et al. (U.S. Patent Pub. No. US 11777786 B2).
Claim 2: Sheller, Walters, Zoldi, and Xu teach the system of claim 1. Sheller further teaches wherein the central node to determine whether to include the set of model weights into a central model at the central node (i.e. Example 20 includes a system for federated training of a neural network, the system comprising an aggregator device to aggregate model updates provided by one or more edge devices, and the one or more edge devices to implement respective neural networks, the one or more edge devices to provide the model updates to the aggregator device, at least one of the edge devices to implement the neural network within a trusted execution environment; para. [0125]).
Sheller does not explicitly teach compares the anomaly score against an anomaly threshold predetermined.
However, Elsakhawy teaches compares the anomaly score against an anomaly threshold predetermined (i.e. performing an anomaly detection process. The anomaly detection process includes selecting, from a plurality of historical data points of the feature, a set of historical data points. Using a machine learning module, a set of scores, each corresponding to one historical data point, is selected. An anomaly threshold is determined based on the set of scores, and compared with a score corresponding to a current data point of the feature to determine whether the current data point is anomalous or not; abs).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Xu to include the feature of Elsakhawy. One would have been motivated to make this modification because it improves how the edge device decides whether later-arriving local data should be accepted for local training.
8. Claim 3 is rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Xu, and further in view of Ezrielev et al. (U.S. Patent Application Pub. No. US 20240119148 A1).
Claim 3: Sheller, Walters, Zoldi, and Xu teach the system of claim 1. Sheller further teaches determining, based on the anomaly score, not to include the third data update into training data for a local model to be trained at the client device (i.e. determining, by executing an instruction with a processor of an edge device, whether to allow a new local data item to be incorporated into a training process of a neural network implemented at the edge device; para. [0119]); training the local model based on the first data update and the second data update (i.e. training the neural network to create a model update using local data items; para. [0119]).
Sheller does not explicitly teach removing the third data update and the third data profile from a memory of the client device.
However, Ezrielev teaches removing the third data update and the third data profile from a memory of the client device (i.e. other data obtained from the one or more data collectors (e.g., data including anomalies and/or any other data not used to re-train the inference model) may also be discarded. Discarding data may include deleting the data; para. [0015, 0075]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Xu to include the feature of Ezrielev. One would have been motivated to make this modification because it improves data security and reduces local storage burden by deleting anomalous data from the device.
9. Claims 4 and 13 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, and further in view of Movshovitz et al. (U.S. Patent Application Pub. No. US 20220279045 A1).
Claim 4: Sheller teaches a method for detecting anomalous data updates on a client device in a federated learning system, comprising (i.e. Federated Learning enables a model representing a neural network to be trained using data across many edge systems without having to centralize the data used for such training. Edge devices perform local training, and provide training results to an aggregator device, which aggregates the training results among the multiple edge devices to update a centralized model, which can then be re-distributed to the edge devices for subsequent training and/or use; para. [0015, 0119]):
during a first interval of time, retrieving a plurality of data updates and a corresponding plurality of data (i.e. the local data throttler 325 is instructed that new local data should be committed every N training rounds. In such an example, the local data throttler 325 determines whether N training rounds have elapsed since additional and/or new local model data was allowed to be used as part of the model training process. In some examples, the value for N is provided by the example aggregator device 110 when transmitting the model to the edge device 130; para. [0042, 0043], repeated of new local training data over different rounds);
receiving a subsequent data update and a corresponding subsequent data (i.e. If the example update transmission throttler 325 determines that the at least N training rounds have elapsed since the new local data was last allowed to be incorporated in the model training process, the example local data throttler 325 enables the inclusion of new local data in the model training process; para. [0043]);
based on the anomaly score (i.e. wherein the throttling means is further to determine whether the new local data item is trusted; para. [0149]), determining not to label the subsequent data update as acceptable (i.e. an edge device for federated training of a neural network, the edge device comprising a local data throttler to determine whether to allow a new local data item to be incorporated into a training process of a neural network at the edge device, the neural network implemented within a trusted execution environment of the edge device, a hash ledger to store hashes corresponding to local data items that are permitted to be used in training of the neural network a model receiver to apply model parameters provided to the neural network by an aggregator device, a neural network trainer to train the neural network to create a model update using local data items that have a corresponding hash stored in the hash ledger, and a model update provider to provide the model update to the aggregator device; para. [0106]) for inclusion in training data to generate a local model (i.e. wherein the one or more edge devices are to throttle inclusion of new local training data when training the respective neural networks based on an instruction provided by the aggregator device; para. [0126], deciding whether a new local data item is acceptable for inclusion and whether it may be used in local training process of the edge device); and
training the local model based on the plurality of data updates after determining not to label the subsequent data update as acceptable (i.e. an edge device for federated training of a neural network, the edge device comprising a local data throttler to determine whether to allow a new local data item to be incorporated into a training process of a neural network at the edge device, the neural network implemented within a trusted execution environment of the edge device, a hash ledger to store hashes corresponding to local data items that are permitted to be used in training of the neural network a model receiver to apply model parameters provided to the neural network by an aggregator device, a neural network trainer to train the neural network to create a model update using local data items that have a corresponding hash stored in the hash ledger, and a model update provider to provide the model update to the aggregator device; para. [0106]).
Sheller does not explicitly teach data profiles; generating a first data profile trend based on the plurality of data profiles, wherein the first data profile trend comprises an expectation value and a measure of variance; determining a measure of deviation based on the expectation value of the first data profile trend and the subsequent data profile; processing the measure of deviation and the measure of variance using a clustering machine learning model to generate an anomaly score indicating an extent to which the measure of deviation falls within the measure of variance.
However, Walters teaches during a first interval of time, retrieving a plurality of data updates (i.e. the predictive model may be implemented daily based on received data to generate generated values at five-minute increments over the next week (e.g., stock forecast, weather forecast, etc.), consistent with disclosed embodiments. In some embodiments, step 1808 is performed a plurality of times over a plurality of different model input datasets; para. [0182, 0183]) and a corresponding plurality of data profiles (i.e. data profiles; para. [0191, 0194]), scheduled repeated datasets and a plurality of event/current datasets received over time, and repeated determination of data profiles/current data metrics for those datasets;
generating a first data profile trend based on the plurality of data profiles, wherein the first data profile trend comprises an expectation value and a measure of variance (i.e. a baseline data metric of the data profile of the baseline synthetic data is determined, the data profile including the data schema and a statistical profile. For example, step 1906 may include determining a baseline covariance matrix of the baseline synthetic data. In some embodiments, the data metric may include a data distribution, a vocabulary (character, word, or the like), a data category, a min, a max, a variance, a quartile, a quantile, a median, a range, or other data metric; para. [0182, 0191]);
receiving a subsequent data update and a corresponding subsequent data profile (i.e. step 1910 may be repeated. That is, in some embodiments, a plurality of current synthetic datasets are generated a plurality of times (e.g., based on a schedule). The plurality of synthetic of datasets may be based on respective current input datasets; para. [0183, 0193, 0194]);
determining a measure of deviation based on the expectation value of the first data profile trend and the subsequent data profile (i.e. At step 1916 data drift is detected based on a difference between a current data metric and a baseline data metric. For example, detecting data drift at step 1916 may involve determining whether a difference between a current data metric and a baseline data metric exceeds a threshold. In some embodiments, model optimizer 107 may detect data drift in a manner consistent with the disclosed embodiments. In some aspects, steps 1910-1914 are repeated a plurality of times, and data drift is detected based on a plurality of current data metrics. Data drift may be detected using known statistical methods applied to the plurality of current data metrics (e.g, by using a least squares error method, a regression method, a correlation method, or other known statistical method); para. [0184, 0196]);
generating based on the measure of deviation and the measure of variance (i.e. detecting data drift is a based on a comparison of predicted data to event data to determine a difference between predicted data and event data. In the embodiments, detecting data drift may be based on known statistical methods. For example, detecting data drift at step 1812 may be based on at least one of a least squares error method, a regression method, a correlation method, or other known statistical method. In some embodiments, the difference is determined using at least one of a Mean Absolute Error, a Root Mean Squared Error, a percent good classification, or the like; para. [0184]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of Sheller to include the feature of Walters. One would have been motivated to make this modification because it improves the reliability and quality of newly received local training data before that data is incorporated into the local training process.
However, Zoldi teaches generating an anomaly score based on the measure of deviation and the measure of variance (i.e. calculate the mean and variance of the fraud probability, calculate transaction transversal pathway distributions of the model development data and the new transaction data, calculate metrics such as deviation, correlation and K-L distance between the transversal pathway distributions of the developed model and the new transaction dataset; determine a degree of anomaly based on these three metrics and generating an alert according to a predetermined criterion; para. [0018, 0027]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller and Walters to include the feature of Zoldi. One would have been motivated to make this modification because it improves the quality and trustworthiness of the local data used to generate the local model.
However, Movshovitz teaches processing the measure of deviation (i.e. Each cluster has its own distribution of sessions around its centroid. Thus, when the method evaluates the deviation of a session from its cluster's centroid it should be evaluated according to the distribution of that cluster; para. [0247, 0259]) and the measure of variance (i.e. the system computes for each cluster and for each histogram type the average and the standard deviation of the distance of the cluster's sessions histogram from the cluster's centroid histogram. The result is an average distance and the standard deviation of distance for each cluster's histogram that characterize the cluster's sessions distribution around the cluster's centroid for this type of histogram; para. [0246-0248]) using a clustering machine learning model (i.e. The clustering engine receives a set of sessions' histograms generated by the Session's Histogram Builder component as its input; para. [0223]) to generate an anomaly score indicating an extent to which the measure of deviation falls within the measure of variance (i.e. Each cluster has its own distribution of sessions around its centroid. Thus, when the method evaluates the deviation of a session from its cluster's centroid it should be evaluated according to the distribution of that cluster. This concept can be extended to each cluster's histogram, as follows: when the method compares a session's histogram to its cluster's histogram it should be done based on the cluster's distribution for this specific histogram; para. [0247-0249, 0251]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, and Zoldi to include the feature of Movshovitz. One would have been motivated to make this modification because it reduces false positive anomaly determinations.
Claim 13 is similar in scope to Claim 4 and is rejected under a similar rationale.
10. Claims 5, 6, 14, and 15 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Movshovitz, and further in view of Kloke et al. (U.S. Patent Application Pub. No. US 20160350389 A1).
Claim 5: Sheller, Walters, Zoldi, and Movshovitz teach the method of claim 4. Sheller does not explicitly teach selecting a lens function to project the plurality of data profiles into a real-valued space; selecting a cover function to partition the real-valued space into a plurality of sectors; generating a plurality of sector datasets associated with the plurality of sectors; and processing the plurality of sector datasets to generate the first data profile trend, wherein the first data profile trend is a relational data structure.
However, Kloke teaches selecting a lens function to project the plurality of data profiles into a real-valued space; selecting a cover function to partition the real-valued space into a plurality of sectors; generating a plurality of sector datasets associated with the plurality of sectors; and processing the plurality of sector datasets to generate the first data profile trend, wherein the first data profile trend is a relational data structure (i.e. Generating the topological representations using the receive data set may comprise generating a plurality of reference spaces using each metric-lens combination, mapping the data points of the data set into each reference space using a different metric-lens combination, and for each reference space: clustering data in a cover of the reference space based the data points of the data set, identifying nodes of the plurality of nodes based on the clustered data, and identifying edges between nodes; para. [0011]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Movshovitz to include the feature of Kloke. One would have been motivated to make this modification because it provides a structured way to organize multiple historical data profiles into a relational trend model for subsequent anomaly evaluation.
Claim 6: Sheller, Walters, Zoldi, Movshovitz, and Kloke teach the method of claim 5. Sheller does not explicitly teach using the lens function, projecting the subsequent data profile into the real-valued space; using the cover function, assigning the subsequent data profile to a first sector dataset associated with a first sector; and using the first sector and the first data profile trend, determining the measure of deviation.
However, Kloke further teaches using the lens function, projecting the subsequent data profile into the real-valued space; using the cover function, assigning the subsequent data profile to a first sector dataset associated with a first sector; and using the first sector and the first data profile trend (i.e. Generating the topological representations using the receive data set may comprise generating a plurality of reference spaces using each metric-lens combination, mapping the data points of the data set into each reference space using a different metric-lens combination, and for each reference space: clustering data in a cover of the reference space based the data points of the data set, identifying nodes of the plurality of nodes based on the clustered data, and identifying edges between nodes; para. [0011]), determining the measure of deviation (i.e. the shape containing flares suggests a classification of the data descriptive of ways in which phenomena can deviate from the norm, which would typically be represented by the central core; para. [0077]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Movshovitz to include the feature of Kloke. One would have been motivated to make this modification because it provides a structured way to organize multiple historical data profiles into a relational trend model for subsequent anomaly evaluation.
Claims 14 and 15 are similar in scope to Claims 5, 6 and are rejected under a similar rationale.
11. Claims 7 and 16 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Movshovitz, and further in view of Phillip et al. (U.S. Patent Application Pub. No. US 20230017695 A1).
Claim 7: Sheller, Walters, Zoldi, and Movshovitz teach the method of claim 4. Sheller does not explicitly teach wherein generating the first data profile trend comprises: using a first extrapolation machine learning model, processing the plurality of data profiles to generate the expectation value, wherein the first extrapolation machine learning model comprises a Bayesian regression algorithm; and using a second extrapolation machine learning model, processing the plurality of data profiles to generate the measure of variance, wherein the first extrapolation machine learning model comprises a principal component analysis algorithm.
However, Walters further teaches wherein generating the first data profile trend comprises: using a first extrapolation machine learning model, processing the plurality of data profiles to generate the expectation value (i.e. The operations may include receiving model input data and generating predicted data using the predictive model, based on the model input data; para. [0010]); and using a second extrapolation machine learning model, processing the plurality of data profiles to generate the measure of variance, wherein the first extrapolation machine learning model comprises a principal component analysis algorithm (i.e. a regression check can regress a numerical column in a dataset against other columns in the dataset, determining the predictability of the numerical column given the other columns. In some aspects, a regression error cross check can determine a regression formula for a numerical column of the synthetic data and then evaluate the predictive ability of the regression formula for the numerical column of the actual data. In various aspects, a principal component analysis check can determine a number of principal component analysis columns sufficient to capture a predetermined amount of the variance in the dataset; para. [0131]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of Sheller to include the feature of Walters. One would have been motivated to make this modification because it improves the reliability and quality of newly received local training data before that data is incorporated into the local training process.
However, Phillip teaches wherein the first extrapolation machine learning model comprises a Bayesian regression algorithm (i.e. Bayesian regression on the set of one or more seafloor ripple geometry predictions, wherein the Bayesian regression is based on posterior distribution generated using Markov Chain Monte Carlo sampling. Step 916 may include responsive to performing the Bayesian regression, generating, by the processing device, a probabilistic distribution of predicted seafloor ripple geometry; para. [0103]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Movshovitz to include the feature of Phillip. One would have been motivated to make this modification because using Bayesian regression as a prediction that produces a probabilistic predictive output.
Claim 16 is similar in scope to Claim 7 and is rejected under a similar rationale.
12. Claims 9 and 18 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Movshovitz, and further in view of Elsakhawy et al. (U.S. Patent Pub. No. US 11777786 B2).
Claim 9: Sheller, Walters, Zoldi, and Movshovitz teach the method of claim 4. Sheller further teaches receiving, from a central node of the federated learning system (i.e. the example aggregator device 110 of FIG. 1 to instruct the edge device concerning use of new training data for a given training round; para. [0063]).
Sheller does not explicitly teach an anomaly threshold, wherein the anomaly threshold is a predetermined real value; and comparing the anomaly score to the anomaly threshold.
However, Elsakhawy teaches receiving information identifying an anomaly threshold, wherein the anomaly threshold is a predetermined real value; and comparing the anomaly score to the anomaly threshold (i.e. performing an anomaly detection process. The anomaly detection process includes selecting, from a plurality of historical data points of the feature, a set of historical data points. Using a machine learning module, a set of scores, each corresponding to one historical data point, is selected. An anomaly threshold is determined based on the set of scores, and compared with a score corresponding to a current data point of the feature to determine whether the current data point is anomalous or not; abs).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Movshovitz to include the feature of Elsakhawy. One would have been motivated to make this modification because it improves how the edge device decides whether later-arriving local data should be accepted for local training.
Claim 18 is similar in scope to Claim 9 and is rejected under a similar rationale.
13. Claims 10 and 19 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Movshovitz, and further in view of Achleitner (U.S. Patent Application Pub. No. US 20220368701 A1).
Claim 10: Sheller, Walters, Zoldi, and Movshovitz teach the method of claim 4. Sheller does not explicitly teach a vector, wherein each value in the vector represents an average of a feature in a set of features in a first data update, wherein the first data profile corresponds to the first data update; distributions of the set of features in the data update; a frequency of null values for the set of features; and a covariance matrix between the set of features.
However, Walters further teaches wherein a first data profile in the plurality of data profiles comprises descriptive statistics (i.e. the data profile including the data schema and a statistical profile. For example, step 1906 may include determining a baseline covariance matrix of the baseline synthetic data. In some embodiments, the data metric may include a data distribution, a vocabulary (character, word, or the like), a data category, a min, a max, a variance, a quartile, a quantile, a median, a range, or other data metric; para. [0191, 0194]), comprising: a first data update, wherein the first data profile corresponds to the first data update (i.e. a baseline data metric of the data profile of the baseline synthetic data is determined, the data profile including the data schema and a statistical profile; para. [0191]); distributions of the set of features in the data update (i.e. the similarity metric value can include at least one of a statistical correlation score (e.g., a score dependent on the covariances or univariate distributions of the synthetic data and the normalized reference dataset); para. [0098, 0100]); and a covariance matrix between the set of features (i.e. system 100 can be configured to generate a difference matrix using a covariance matrix of the normalized reference dataset and a covariance matrix of the synthetic dataset. As a further example, the difference matrix can be the difference between the covariance matrix of the normalized reference dataset and the covariance matrix of the synthetic dataset; para. [0099]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of Sheller to include the feature of Walters. One would have been motivated to make this modification because it improves the reliability and quality of newly received local training data before that data is incorporated into the local training process.
However, Achleitner teaches a vector, wherein each value in the vector represents an average of a feature in a set of features (i.e. numerically average feature vectors in the feature vectors 104; para. [0025]); a frequency of null values for the set of features (i.e. Representative vectors can be generated by averaging feature vectors within each cluster or choosing features for each feature placement with highest statistical frequency. Entries of representative vectors where there is no feature with highest statistical frequency for the corresponding feature placement can be left as empty or null features; para. [0047]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Movshovitz to include the feature of Elsakhawy. One would have been motivated to make this modification because it improves the compact statistical representation of profile data for subsequent anomaly analysis.
Claim 19 is similar in scope to Claim 10 and is rejected under a similar rationale.
14. Claims 11 and 20 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Movshovitz, and further in view of Shanableh et al. (U.S. Patent Application Pub. No. US 20100316131 A1).
Claim 11: Sheller, Walters, Zoldi, and Movshovitz teach the method of claim 4. Sheller does not explicitly teach the expectation value is a vector of real values corresponding to a set of features, wherein the set of features is associated with the plurality of data updates and the plurality of data profiles; and the measure of variance is a vector of real values, wherein each value in the measure of variance is derived from a standard deviation of a feature in the set of features.
However, Shanableh teaches the expectation value is a vector of real values corresponding to a set of features, wherein the set of features is associated with the plurality of data updates and the plurality of data profiles; and the measure of variance is a vector of real values, wherein each value in the measure of variance is derived from a standard deviation of a feature in the set of features (i.e. Once the MB features are extracted from both the bitstream and reconstructed video. The feature vectors are normalized to either the frame or the whole sequence. The normalization is applied to each feature separately. The normalization of choice in this work is z-scores defined as: z.sub.i=(x.sub.i-E(x))/.sigma. (5) Where the scalars z.sub.i and x.sub.i are the normalized and non-normalized feature values of feature index i respectively. E(x) is the expected value of the feature variable vector and .sigma. is its standard deviation. Both are computed based on the feature vector population; para. [0053, 0054]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Movshovitz to include the feature of Shanableh. One would have been motivated to make this modification because it makes anomaly comparation more reliable and comparable across features.
Claim 20 is similar in scope to Claim 11 and is rejected under a similar rationale.
15. Claim 12 is rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Movshovitz, and further in view of Ezrielev et al. (U.S. Patent Application Pub. No. US 20240119148 A1).
Claim 12: Sheller, Walters, Zoldi, and Movshovitz teach the method of claim 4. Sheller further teaches wherein: determining, based on the anomaly score, not to include the subsequent data update into training data for a local model to be trained at the client device (i.e. determining, by executing an instruction with a processor of an edge device, whether to allow a new local data item to be incorporated into a training process of a neural network implemented at the edge device; para. [0119]); training the local model based on the plurality of data updates (i.e. training the neural network to create a model update using local data items; para. [0119]).
Sheller does not explicitly teach removing the subsequent data update and the subsequent data profile from a memory of the client device.
However, Ezrielev teaches removing the subsequent data update and the subsequent data profile from a memory of the client device (i.e. other data obtained from the one or more data collectors (e.g., data including anomalies and/or any other data not used to re-train the inference model) may also be discarded. Discarding data may include deleting the data; para. [0015, 0075]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, Zoldi, and Movshovitz to include the feature of Ezrielev. One would have been motivated to make this modification because it improves data security and reduces local storage burden by deleting anomalous data from the device.
16. Claims 21 and 22 are rejected under 35 U.S.C. 103 as being unpatentable over Sheller in view of Walters, Zoldi, Movshovitz, and further in view of Xu.
Claim 21: Sheller, Walters, Zoldi, and Movshovitz teach the method of claim 4. Sheller does not explicitly teach wherein generating the first data profile trend comprises: processing, using an extrapolation model that is configured to output predicted values for a set of features, the plurality of data profiles to generate the first data profile trend.
However, Walters further teaches herein generating the first data profile trend comprises: processing, using an extrapolation model that is configured to output predicted values (i.e. At step 1808, the predictive model generates predicted data based on the model input data; para. [0182]), the plurality of data profiles to generate the first data profile trend (i.e. a baseline data metric of the data profile of the baseline synthetic data is determined, the data profile including the data schema and a statistical profile. For example, step 1906 may include determining a baseline covariance matrix of the baseline synthetic data. In some embodiments, the data metric may include a data distribution, a vocabulary (character, word, or the like), a data category, a min, a max, a variance, a quartile, a quantile, a median, a range, or other data metric; para. [0182, 0184, 0191]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of Sheller to include the feature of Walters. One would have been motivated to make this modification because it improves the reliability and quality of newly received local training data before that data is incorporated into the local training process.
However, Xu teaches processing, using an extrapolation model (i.e. The time series predictor 202 applies a regression network 226 to the normalized space representing the set of historical feature values 210; para. [0051]) that is configured to output predicted values (i.e. The anomaly detector 204 can receive the current feature value 220 and a set of predicted values 222 that includes the upper boundary 216, the expected value 214, and the lower boundary 218; para. [0055]) for a set of features (i.e. The time series predictor 160 predicts an expected value and an expected range of values for each feature of the online system … the time series predictor is a multilayer perceptron model that receives multiple time series features such as user clicks, message counts, content posts, document counts, activities by a user interacting with an online system, or other time series features; para. [0037, 0072]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, and Zoldi, and Movshovitz to include the feature of Xu. One would have been motivated to make this modification because it improves early identification of anomalous training data.
Claim 22: Sheller, Walters, Zoldi, Movshovitz, and Xu teach the method of claim 21. Sheller does not explicitly teach wherein the data profile trend comprises: an expectation value that includes the predicted values, and a measure of variance.
However, Walters further teaches wherein the data profile trend comprises: an expectation value that includes the predicted values (i.e. Step 1810 includes receiving event data, the event data being of the same category as the predicted data. Event data may be the data that corresponds to the real-world data that the predictive model was used to predict; para. [0183]), and a measure of variance (i.e. the duration and/or periodicity of the schedule may be based on one or more aspects of the disclosed embodiments, such as a characteristic of the training data, predicted data, or event data (e.g., variance, sampling rate, detecting a measured value falls inside or outside a particular range or above or below particular threshold, etc.), detecting a correction in the predictive model, or other such feature; para. [0182, 0191, 0192]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the invention of Sheller to include the feature of Walters. One would have been motivated to make this modification because it improves the reliability and quality of newly received local training data before that data is incorporated into the local training process.
However, Xu further teaches wherein the data profile trend comprises: an expectation value (i.e. The time series predictor 160 predicts an expected value and an expected range of values for each feature of the online system; para. [0037, 0050]) that includes the predicted values (i.e. The time series predictor 160 predicts, using the set of historical feature values 210 and he set of auxiliary feature values 212, an expected value 214, an upper boundary 216, and a lower boundary 218; para. [0050]).
Therefore, it would have been obvious to one of ordinary skill in the art before the effective filling date of the claimed invention to modify the combination of Sheller, Walters, and Zoldi, and Movshovitz to include the feature of Xu. One would have been motivated to make this modification because it improves early identification of anomalous training data.
Conclusion
The prior art made of record and not relied upon is considered pertinent to applicant’s disclosure.
Wojnowicz et al. (Pub. No. US 20200057854 A1), the proximity module 120 may determine a distance (e.g., Mahalanobis distance) between lower k-dimensional representations of the unclassified file and of the file set. As noted above, the distance, between the unclassified file and the file set may correspond to an amount of deviation between the features of the unclassified file and the features of the files in the file set.
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 extension fee 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 date of this final action.
It is noted that any citation to specific pages, columns, lines, or figures in the prior art references and any interpretation of the references should not be considered to be limiting in any way. A reference is relevant for all it contains and may be relied upon for all that it would have reasonably suggested to one having ordinary skill in the art. In re Heck, 699 F.2d 1331, 1332-33, 216 U.S.P.Q. 1038, 1039 (Fed. Cir. 1983) (quoting In re Lemelson, 397 F.2d 1006, 1009, 158 U.S.P.Q. 275, 277 (C.C.P.A. 1968)).
Any inquiry concerning this communication or earlier communications from the examiner should be directed to TAN TRAN whose telephone number is (303)297-4266. The examiner can normally be reached on Monday - Thursday - 8:00 am - 5:00 pm MT.
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, Matt Ell can be reached on 571-270-3264. The fax phone number for the organization where this application or proceeding is assigned is 571-273-8300.
Information regarding the status of an application may be obtained from the Patent Application Information Retrieval (PAIR) system. Status information for published applications may be obtained from either Private PAIR or Public PAIR. Status information for unpublished applications is available through Private PAIR only. For more information about the PAIR system, see http://pair-direct.uspto.gov. Should you have questions on access to the Private PAIR system, contact the Electronic Business Center (EBC) at 866-217-9197 (toll-free). If you would like assistance from a USPTO Customer Service Representative or access to the automated information system, call 800-786-9199 (IN USA OR CANADA) or 571-272-1000.
/TAN H TRAN/Primary Examiner, Art Unit 2141
Prosecution Insights