Machine learning to monitor operations of a device
A multi-tier machine learning engine receives signal data characterizing a monitored signal of the computing platform. The machine learning engine can include a plurality of tiers that employ frequency domain analysis on the signal data to identify an application executing on the computing platform and a module and/or loop of the identified application and employ time domain analysis on the signal data to identify timing of events within the identified module and/or loop of the identified application.
1. A system for monitoring operations of a computing platform comprising:
a non-transitory memory for storing machine readable instructions; and
a processing unit that accesses the memory and executes the machine readable instructions, the machine readable instructions comprising:
a multi-tier machine learning engine that receives signal data characterizing a monitored signal of the computing platform, the machine learning engine comprising:
a plurality of tiers that employ frequency domain analysis on the signal data to identify an application executing on the computing platform and a module and/or loop of the identified application and employ time domain analysis on the signal data to identify timing of events within the identified module and/or loop of the identified application, wherein a number of tiers executed by the multi-tier machine learning algorithm is selected based on resources allocated to the system, and each tier in the plurality of tiers the multi-tier machine learning algorithm executes a corresponding machine learning algorithm that is trained independently.
2. The system of claim 1 , wherein the plurality of tiers of the multi-tier machine learning engine further employ the identified timing of the events to determine control flow paths being executed by the computing platform.
3. The system of claim 1 , wherein the monitored signal is one of radio frequency (RF) emissions from the computing platform and consumption of a power signal consumed by the computing platform.
4. The system of claim 1 , wherein the computing platform is an Internet of Things (IoT) device, the IoT device comprising:
a memory that stores machine readable instructions; and
a processing unit that accesses the memory and execute the machine readable instructions.
5. The system of claim 1 , wherein at least one tier of the plurality of tiers comprises a machine learning model trained with data characterizing code structure of at least one of authorized code and unauthorized code executable by the computing platform.
6. The system of claim 1 , wherein the signal data characterizing the monitored signal is provided from a sensor, wherein the system is air gapped from the computing platform.
7. The system of claim 6 , wherein the sensor receives an interference signal generated by a device separate from the computing platform and each tier in the plurality of tiers of the multi-tier machine learning engine comprises operations to curtail the interference signal.
8. The system of claim 7 , wherein the operations of a first tier of the plurality of tiers in the multi-tier machine learning engine comprises a short-time Fourier transform (STFT) that determines a sinusoidal frequency and phase content of local sections of the monitored signal as the sinusoidal frequency and phase content changes over time.
9. The system of claim 8 , wherein a second tier of the plurality of tiers in the multi-tier machine learning engine comprises an average magnitude difference function (AMDF) that compares segments of the monitored signal with other segments offset by a trial period to find a match.
10. The system of claim 1 , wherein the multi-tier machine learning engine comprises a hierarchical Dynamic Bayesian Network (DBN).
11. The system of claim 10 , wherein the hierarchical DBN receives data characterizing a flow chart of a given application stored on the computing platform.
12. The system of claim 11 , wherein each tier in the plurality of tiers of the hierarchical DBN employs the data characterizing flow of the given application to correlate the monitored signal with the flow of the given application.
13. The system of claim 11 , wherein a given tier in the plurality of tiers of the hierarchical DBN employs hyperdimensional Bayesian time mapping to correlate the monitored signal with the flow of the given application.
14. The system of claim 1 , wherein the computing platform comprises a plurality of computing platforms, and the multi-tier machine learning engine identifies hardware for at least a subset of the plurality of computing platforms.
15. The system of claim 1 , wherein the multi-tier machine learning engine generates an alert indicating that the computing platform is potentially executing malicious code in response to determining that the determined control flow paths being executed by the computing platform deviates from an expected pattern by more than a threshold level.
16. A non-transitory machine readable medium having machine executable instructions, the machine executable instructions comprising:
a hierarchical dynamic Bayesian network (DBN) that receives signal data characterizing a monitored signal of a computing platform, and an interference signal generated by a device separate from the computing platform, the hierarchical DBN comprising:
a first tier that employs frequency domain analysis on the signal data to identify an application executing on the computing platform;
a second tier that employs frequency domain analysis on the signal data to identify a module and/or loop of the identified application;
a third tier that employs time domain analysis on the signal data to identify timing of events within the identified module and/or loop of the identified application; and
a fourth tier that employs the identified timing of the events to determine control flow paths being executed by the computing platform;
wherein the first tier, the second tier and the third tier each comprises operations to curtail the interference signal from the signal data.
17. The non-transitory medium of claim 16 , wherein the monitored signal is a radio frequency (RF) signal emanating from the computing platform and the signal data is provided from a sensor is air gapped from the computing platform.
18. A method comprising:
receiving, at a multi-tier machine learning engine, signal data characterizing a monitored signal of a computing platform and an interference signal, wherein the monitored signal and the interference signal are detected by a sensor air gapped from the computing device, such that the sensor and the computing device are physically and logically separated, and the monitored signal is one of radio frequency (RF) emissions and consumption of a power signal consumed by the computing platform;
preprocessing, by a plurality of tiers of the multi-tier machine learning engine, the signal data;
executing frequency domain analysis on the signal data to identify a module and/or loop of an identified application executing on the computing platform; and
identifying timing of events within an identified module and/or loop of the identified application.
19. The method of claim 18 , wherein at least one tier of the multi-tier machine learning engine employs time domain analysis on the signal data.