Distributed learning anomaly detector
In a network discovery and management system, a machine learning (ML) DLAD processor trains, validates, updates, and stores machine learning models. A ML training data preparation program performs operations to process and format input data to generate ML training data that can be used to train ML models. ML training program uses the ML training data to train ML models, thereby generating trained ML models. The ML training program can re-train or update the training of ML models as the system collects additional data and produces additional estimates, predictions, and forecasts. ML model validation program performs validation testing on trained ML models to generate one or more metrics that can indicate accuracy of predictions generated by the trained models. The resulting ML model(s) can be used to manage the network including but not limited to retrieve, instantiate and execute dynamic applications based on predictions made based on the models.
1. A network comprising:
a non-transitory storage device that stores a portable encoding of an initial machine learning-trained hyperparameter data set parameterizing operating metrics and information characterizing operation of executing software components of at least one proto-typical network device as a Distributed Learning Anomaly Detector (DLAD) dynamic application configured to use the initial machine learning-trained hyperparameter data set parameterizing the operating metrics and information comprising at least one collected data set from at least one data source specified by the DLAD dynamic application; and
a machine-learning environment configured to incorporate information from the DLAD dynamic application including data from the at least one data source specified by the DLAD dynamic application,
the machine-learning environment configured to use, as initial parameters, the initial machine learning-trained hyperparameter data set for local machine learning using local data and a non-locally initialized model configured to model executing software component characteristics of devices, the initial machine learning-trained hyperparameter data set comprising a configuration for the non-locally initialized model;
the machine-learning environment being further configured to use (a) the initial machine learning-trained hyperparameter data set or a hyperparameter data set derived, at least in part, from the machine learning-trained hyperparameter data set and (b) the at least one collected data set from the at least one data source, to discover operational condition events,
wherein the DLAD dynamic application selectively instantiates at least one additional dynamic application based at least on part on the discovered operational condition events.
2. The network of claim 1 wherein the network includes a target network management domain and further includes a data collector that collects the at least one collected data set from the target network management domain.
3. The network of claim 1 further comprising at least one processor configured to train the initial machine learning-trained hyperparameter data set with the at least one collected data set to derive an at least in part locally trained hyperparameter data set.
4. The network of claim 3 further comprising a deep neural network trained, at least in part, by the locally trained hyperparameter data set.
5. The network of claim 2 further comprising:
a monitoring device that uses the initial machine learning trained hyperparameter data set to configure monitoring of discovered operational condition events and predict whether the discovered operational condition events are anomalous; and
a network management device configured to use results of the predicting to manage at least one aspect of the target network management domain.
6. The network of claim 5 further comprising:
an alerting component configured for applying anomalous value thresholds of monitored data set in accordance with the anomalous events predicting, and
a comparator that compares a collected data set value against the anomalous value thresholds, and
wherein the alerting component generates an alert or notification upon the comparator indicating the collected data set value is outside at least one of the anomalous value thresholds.
7. The network of claim 6 , further comprising the at least one processor configured to calculate and store one or more predicted future data values for the monitored data set.
8. The network of claim 1 further comprising at least one processor that is configured to determine whether a discovered operational condition event associated with one or more data set elements comprises an anomaly event.
9. The network of claim 1 further comprising a predictor that uses event tags to predict whether observed events represented by the collected data set are anomalous.
10. The network of claim 1 further comprising a model registry and a training handler configured to interface with the initial machine learning trained hyperparameter data set.
11. The network of claim 1 wherein the DLAD dynamic application is configured using a DLAD specification.
12. The network of claim 1 further comprising a DLAD engine of a network management device configured to run the DLAD dynamic application to discover anomalies.
13. The network of claim 12 wherein the initial machine learning-trained hyperparameter data set comprises a configuration for a trained model that is external to the trained model and is used to parameterize use of the trained model.
14. The network of claim 1 wherein the DLAD dynamic application includes one or more of program specifications, DLAD parameters, applicable data types, and data sources.
15. A network comprising:
a non-transitory storage that stores a portable encoding of an initial machine learning-trained hyperparameter data set comprising a configuration for a trained model that is external to the trained model and is used to parameterize use of the trained model, the hyperparameter data set parameterizing information relating to executing software components of at least one proto-typical network device as a Distributed Learning Anomaly Detector (DLAD) dynamic application configured for execution on the network,
the initial machine learning-trained hyper-parameter data set providing initial parameters for local machine learning using local data and a non-locally initialized model that models executing software component characteristics of devices that are part of a target network management domain; and
at least one processor operatively coupled to the non-transitory storage and configured to execute a machine-learning environment incorporating information from the DLAD dynamic application including data from at least one data source the DLAD dynamic application specifies, the DLAD dynamic application using the initial machine learning-trained hyperparameter data set to process at least one set of data collected from the network;
the machine-learning environment being configured to, based on the processing, use (a) the initial machine learning-trained hyperparameter data set or a hyperparameter data set derived, at least in part, from the machine learning-trained hyperparameter data set, and (b) the at least one data source, to discover operational condition events in the at least one set of data collected from the network,
wherein the DLAD dynamic application selectively instantiates at least one additional DLAD dynamic application based at least on part on the discovered operational condition events.
16. The network of claim 15 wherein the at least one processor is further configured to further train the initial machine learning-trained hyperparameter data set with the at least one set of data collected from the network to derive a locally trained hyperparameter data set.
17. The network of claim 16 wherein the at least one processor executes a deep neural network configured to use the locally trained hyperparameter data set.
18. The network of claim 15 wherein the at least one processor is further configured to determine whether a discovered operational condition event associated with one or more data set elements comprises an anomaly event.
19. The network of claim 15 wherein the at least one processor is further configured to:
use the initial machine learning-trained hyperparameter data set to predict whether the discovered operational condition events are anomalous; and
use results of the predicting to manage at least one aspect of the network.
20. The network of claim 19 wherein the at least one processor is further configured to:
execute an alerting routine for anomalous value thresholds of a monitored data set in accordance with predicting whether the discovered operational condition events are anomalous,
compare a collected data set value against the anomalous value thresholds, and
generate an alert or notification upon the compare indicating the collected data set value is outside at least one of the anomalous value thresholds.
21. The network of claim 19 , wherein the at least one processor is further configured to calculate one or more predicted future data values for the monitored data set and store the calculated predicted future data values.
22. The network of claim 15 wherein the at least one processor is further configured to use event tags to predict whether observed events represented by the collected data set are anomalous.
23. The network of claim 15 wherein the at least one processor is further configured to use a model registry and a training handler to interface with the initial machine learning-trained hyperparameter data set.
24. The network of claim 15 wherein the DLAD dynamic application is configured using a DLAD specification.
25. The network of claim 15 wherein the at least one processor is further configured to provide a DLAD engine of a network management device running the DLAD dynamic application to discover anomalies.
26. The network of claim 15 wherein the DLAD dynamic application includes one or more of program specifications, DLAD parameters, applicable data types, and data sources.