Unknown malicious program behavior detection using a graph neural network

1. A computer-implemented method executed on a processor for detecting unknown malicious program behavior, the method comprising:

performing, via the processor, program verification based on system activity data;

analyzing, via the processor, unverified program data identified from the program verification to detect abnormal events, including analyzing host-level events to detect abnormal host-level events by learning a program representation as a graph embedding through an attentional architecture based on an invariant graph between different system entities;

generating detection results based on the analysis;

performing, via the processor, at least one corrective action based on the detection results;

performing, via the processor, a contextual search based on the invariant graph to determine path-relevant neighbor sets;

computing, via the processor, a node-wise attentional weight for each node in the path-relevant neighbor sets to generate node embeddings from different layers;

aggregating, via the processor, the node embeddings toward a dense-connected node embedding; and

automatically learning path-wise attention weights for respective meta-paths to compute a joint embedding.

2. The method of claim 1 , wherein:

analyzing the unverified program data further includes analyzing network communications to detect abnormal network communication events, the network communications being associated with Transmission Control Protocol (TCP) or User Datagram Protocol (UDP); and

the host-level events are selected from the group consisting of: process-to-process, process-to-file, process-to-Internet socket, and combinations thereof.

3. The method of claim 1 , wherein analyzing the host-level events further includes:

modeling the system event data as the invariant graph to capture a program behavior profile; and

learning the program representation as a graph embedding through an attentional architecture including an attentional heterogeneous graph neural network (AHGNN).

4. The method of claim 3 , further comprising training the attentional architecture to distinguish between an unknown program and a known benign program, including learning a similarity metric and the program graph representation jointly for improved graph matching between the unknown program and the known benign program.

5. The method of claim 1 , wherein generating the detection results further includes integrating the abnormal events to obtain integrated data, and refining the integrated data for trustworthy events.

6. The method of claim 1 , wherein performing the corrective action further includes performing at least one corrective action selected from the group consisting of: transmitting the detection results to at least one computing device associated with at least one end-user, changing a security setting for an application or hardware component, changing an operational parameter of an application or hardware component, halting or restarting an application or hardware component, changing an environmental condition, and changing status of a network interface.

7. A computer program product comprising a non-transitory computer readable storage medium having program instructions embodied therewith, the program instructions executable by a computer to cause the computer to perform a method for detecting unknown malicious program behavior, the method performed by the computer comprising:

performing program verification based on system activity data;

analyzing unverified program data identified from the program verification to detect abnormal events, including analyzing host-level events to detect abnormal host-level events by learning a program representation as a graph embedding through an attentional architecture based on an invariant graph between different system entities;

generating detection results based on the analysis;

performing a corrective action based on the detection results;

performing a contextual search based on the invariant graph to determine path-relevant neighbor sets;

computing a node-wise attentional weight for each node in the path-relevant neighbor sets to generate node embeddings from different layers;

aggregating the node embeddings toward a dense-connected node embedding; and

automatically learning path-wise attention weights for respective meta-paths to compute a joint embedding.

8. The computer program product of claim 7 , wherein:

analyzing the unverified program data further includes analyzing network communications to detect abnormal network communication events, the network communications being associated with Transmission Control Protocol (TCP) or User Datagram Protocol (UDP); and

the host-level events are selected from the group consisting of: process-to-process, process-to-file, process-to-Internet socket, and combinations thereof.

9. The computer program product of claim 7 , wherein analyzing the host-level events further includes:

modeling the system event data as the invariant graph to capture a program behavior profile; and

learning the program representation as a graph embedding through an attentional architecture including an attentional heterogeneous graph neural network (AHGNN).

10. The computer program product of claim 9 , wherein the method further includes training the attentional architecture to distinguish between an unknown program and a known benign program, including learning a similarity metric and the program graph representation jointly for improved graph matching between the unknown program and the known benign program.

11. The computer program product of claim 7 , wherein generating the detection results further includes integrating the abnormal events to obtain integrated data, and refining the integrated data for trustworthy events.

12. The computer program product of claim 7 , wherein performing the corrective action further includes performing at least one corrective action selected from the group consisting of: transmitting the detection results to at least one computing device associated with at least one end-user, changing a security setting for an application or hardware component, changing an operational parameter of an application or hardware component, halting or restarting an application or hardware component, changing an environmental condition, and changing status of a network interface.

13. A system for detecting unknown malicious program behavior, comprising:

a memory device for storing program code; and

at least one processor device operatively coupled to a memory device and configured to execute program code stored on the memory device to:

perform program verification based on system activity data;

analyze unverified program data identified from the program verification to detect abnormal events by analyzing host-level events to detect abnormal host-level events by learning a program representation as a graph embedding through an attentional architecture based on an invariant graph between different system entities;

generate detection results based on the analysis;

perform at least one corrective action based on the detection results;

perform a contextual search based on the invariant graph to determine path-relevant neighbor sets;

compute a node-wise attentional weight for each node in the path-relevant neighbor sets to generate node embeddings from different layers;

aggregate the node embeddings toward a dense-connected node embedding; and

automatically learn path-wise attention weights for respective meta-paths to compute a joint embedding.

14. The system of claim 13 , wherein the at least one processor device is further configured to analyze the unverified program data further by analyzing network communications to detect abnormal network communication events, the network communications being associated with Transmission Control Protocol (TCP) or User Datagram Protocol (UDP), wherein the host-level events are selected from the group consisting of: process-to-process, process-to-file, process-to-Internet socket, and combinations thereof.

15. The system of claim 13 , wherein the at least one processor device is further configured to analyze the host-level events by:

modeling the system event data as the invariant graph to capture a program behavior profile; and

learning the program representation as a graph embedding through an attentional architecture including an attentional heterogeneous graph neural network (AHGNN).

16. The system of claim 15 , wherein the at least one processor device is further configured to execute program code stored on the memory device to train the attentional architecture to distinguish between an unknown program and a known benign program by learning a similarity metric and program graph representation jointly for improved graph matching between the unknown program and the known benign program.

17. The system of claim 13 , wherein the at least one processor device is further configured to perform the corrective action by performing at least one corrective action selected from the group consisting of: transmitting the detection results to at least one computing device associated with at least one end-user, changing a security setting for an application or hardware component, changing an operational parameter of an application or hardware component, halting or restarting an application or hardware component, changing an environmental condition, and changing status of a network interface.