Granted Patent B2

US 11,392,821 · App. 16/877,631 · Granted Jul 19, 2022

Inventor: Mohammad Rafey (Bangalore, IN)

Assignee: Dell Products L.P.

G06N3/0445G06K9/6223G06K9/6256G06K9/6269G16Y40/10

Abstract

Claims (52)

1. An apparatus comprising:

at least one processing device comprising a processor coupled to a memory;

the at least one processing device being configured to perform steps of:

obtaining time series diagnostic data associated with a plurality of assets in an information technology infrastructure, the time series diagnostic data comprising feature values for a plurality of features measured for a plurality of time periods;

generating first modality information for the time series diagnostic data, the first modality information comprising behavior labels assigned to each of the plurality of time periods, a given behavior label for a given one of the plurality of time periods being based at least in part on measured feature values for the plurality of features collectively in the given time period;

generating second modality information for the time series diagnostic data, the second modality information comprising feature deltas characterizing differences between measured feature values for interdependent pairs of the plurality of features in each of the plurality of time periods;

performing multi-modal analysis of the time series diagnostic data to detect behavior patterns in the time series diagnostic data utilizing a machine learning model trained using the first modality information and the second modality information; and

initiating at least one remedial action in the information technology infrastructure responsive to detecting an anomalous behavior pattern in the time series diagnostic data.

2. The apparatus of claim 1 wherein the time series diagnostic data is obtained from a plurality of Internet of Things (IoT) devices associated with the information technology infrastructure, and wherein the time series diagnostic data characterizes device health of one or more of the plurality of assets of the information technology infrastructure.

3. The apparatus of claim 1 wherein obtaining the time series diagnostic data associated with the plurality of assets comprises:

separating the time series diagnostic data into two or more chunks, wherein the two or more chunks of the time series diagnostic data comprise feature data for the plurality of features collected from the plurality of assets in the information technology infrastructure, the feature data collected from different ones of the plurality of assets in the information technology infrastructure having different sample lengths per a designated time unit; and

transforming the feature data in the two or more chunks of the time series diagnostic data to have an equal sample length per the designated time unit.

4. The apparatus of claim 3 wherein the designated time unit comprises a duration of each of the plurality of time periods.

5. The apparatus of claim 3 wherein transforming the feature data in the two or more chunks of the time series diagnostic data to have the equal sample length per the designated time unit comprises:

computing a mean time sample length of the feature data across the two or more chunks of the time series diagnostic data;

truncating sample lengths for feature data in the two or more chunks of the time series diagnostic data that are longer than the mean time sample length; and

padding dummy data for feature data in the two or more chunks of the time series diagnostic data that are shorter than the mean time sample length.

6. The apparatus of claim 1 wherein generating the first modality information for the time series diagnostic data comprises assigning the asset behavior labels for each of the plurality of time periods utilizing a binary classifier model.

7. The apparatus of claim 6 wherein the binary classifier model is configured to label asset behavior during each of the plurality of time periods as one of normal behavior and abnormal behavior.

8. The apparatus of claim 1 wherein generating the second modality information for the time series diagnostic data comprises:

analyzing the time series diagnostic data to generate a feature dependency matrix characterizing interdependencies between pairs of the plurality of features; and

identifying the interdependent pairs of the plurality of features based at least in part on the feature dependency matrix.

9. The apparatus of claim 1 wherein a given interdependent pair of the plurality of features comprises a first one of the plurality of features and a second one of the plurality of features, and wherein feature deltas for the given interdependent pair of the plurality of features characterize relative differences between the first feature and the second feature across the plurality of time periods.

10. The apparatus of claim 9 wherein the feature deltas for the given interdependent pair of the plurality of features are non-absolute and independent of positioning in the time series diagnostic data.

11. The apparatus of claim 1 wherein the machine learning model comprises a long short term memory (LSTM) model configured to detect behavior patterns in the time series diagnostic data based at least in part on (i) first patterns in the generated first modality information corresponding to feature values of the plurality of features across the plurality of time periods and (ii) second patterns in the generated second modality information corresponding to differences between the interdependent pairs of features across the plurality of time periods.

12. The apparatus of claim 11 wherein utilizing the trained machine learning model to detect behavior patterns in the time series diagnostic data comprises identifying at least a subset of the plurality of assets in the information technology infrastructure affected by the detected anomalous behavior pattern in the time series diagnostic data.

13. The apparatus of claim 12 wherein initiating the at least one remedial action in the information technology infrastructure responsive to detecting the anomalous behavior pattern in the time series diagnostic data comprises applying the at least one remedial action to the identified subset of the plurality of assets in the information technology infrastructure affected by the detected anomalous behavior pattern in the time series diagnostic data.

14. The apparatus of claim 1 wherein initiating the at least one remedial action comprises at least one of:

applying one or more security hardening procedures to one or more of the plurality of assets associated with the detected anomalous behavior;

modifying a configuration of one or more of the plurality of assets associated with the detected anomalous behavior; and

modifying access, by one or more of a plurality of users, to one or more of the plurality of assets associated with the detected anomalous behavior.

15. A computer program product comprising a non-transitory processor-readable storage medium having stored therein program code of one or more software programs, wherein the program code when executed by at least one processing device causes the at least one processing device to perform steps of:

obtaining time series diagnostic data associated with a plurality of assets in an information technology infrastructure, the time series diagnostic data comprising feature values for a plurality of features measured for a plurality of time periods;

generating first modality information for the time series diagnostic data, the first modality information comprising behavior labels assigned to each of the plurality of time periods, a given behavior label for a given one of the plurality of time periods being based at least in part on measured feature values for the plurality of features collectively in the given time period;

generating second modality information for the time series diagnostic data, the second modality information comprising feature deltas characterizing differences between measured feature values for interdependent pairs of the plurality of features in each of the plurality of time periods;

performing multi-modal analysis of the time series diagnostic data to detect behavior patterns in the time series diagnostic data utilizing a machine learning model trained using the first modality information and the second modality information; and

initiating at least one remedial action in the information technology infrastructure responsive to detecting an anomalous behavior pattern in the time series diagnostic data.

16. The computer program product of claim 15 wherein generating the second modality information for the time series diagnostic data comprises:

analyzing the time series diagnostic data to generate a feature dependency matrix characterizing interdependencies between pairs of the plurality of features; and

identifying the interdependent pairs of the plurality of features based at least in part on the feature dependency matrix.

17. The computer program product of claim 15 wherein a given interdependent pair of the plurality of features comprises a first one of the plurality of features and a second one of the plurality of features, and wherein feature deltas for the given interdependent pair of the plurality of features characterize relative differences between the first feature and the second feature across the plurality of time periods.

18. A method comprising:

obtaining time series diagnostic data associated with a plurality of assets in an information technology infrastructure, the time series diagnostic data comprising feature values for a plurality of features measured for a plurality of time periods;

generating first modality information for the time series diagnostic data, the first modality information comprising behavior labels assigned to each of the plurality of time periods, a given behavior label for a given one of the plurality of time periods being based at least in part on measured feature values for the plurality of features collectively in the given time period;

generating second modality information for the time series diagnostic data, the second modality information comprising feature deltas characterizing differences between measured feature values for interdependent pairs of the plurality of features in each of the plurality of time periods;

performing multi-modal analysis of the time series diagnostic data to detect behavior patterns in the time series diagnostic data utilizing a machine learning model trained using the first modality information and the second modality information; and

initiating at least one remedial action in the information technology infrastructure responsive to detecting an anomalous behavior pattern in the time series diagnostic data;

wherein the method is performed by at least one processing device comprising a processor coupled to a memory.

19. The method of claim 18 wherein generating the second modality information for the time series diagnostic data comprises:

analyzing the time series diagnostic data to generate a feature dependency matrix characterizing interdependencies between pairs of the plurality of features; and

identifying the interdependent pairs of the plurality of features based at least in part on the feature dependency matrix.

20. The method of claim 18 wherein a given interdependent pair of the plurality of features comprises a first one of the plurality of features and a second one of the plurality of features, and wherein feature deltas for the given interdependent pair of the plurality of features characterize relative differences between the first feature and the second feature across the plurality of time periods.

Assignments (9)

Continuity (1)