Unsupervised graph similarity learning based on stochastic subgraph sampling
View Patent ↗Methods and systems for detecting abnormal application behavior include determining a vector representation of a first syscall graph that is generated by a first application, the vector representation including a representation of a distribution of subgraphs of the first syscall graph. The vector representation of the first syscall graph is compared to one or more second syscall graphs that are generated by respective second applications to determine respective similarity scores. It is determined that the first application is behaving abnormally based on the similarity scores, and a security action is performed responsive to the determination that the first application is behaving abnormally.
1. A method for detecting abnormal application behavior, comprising:
determining a vector representation of a first syscall graph that is generated by a first application, the vector representation including a representation of a distribution of subgraphs of the first syscall graph;
comparing the vector representation of the first syscall graph to one or more second syscall graphs that are generated by respective second applications to determine respective similarity scores;
determining that the first application is behaving abnormally based on the similarity scores; and
performing a security action responsive to the determination that the first application is behaving abnormally.
2. The method of claim 1 , wherein determining the vector representation of the first syscall graph includes sampling a plurality of sub-graphs from the first syscall graph.
3. The method of claim 2 , wherein sampling the plurality of sub-graphs from the first syscall graph includes performing a walk with earliest visit time.
4. The method of claim 2 , wherein determining the vector representation of the first syscall graph further includes encoding each of the plurality of sub-graphs as a vector.
5. The method of claim 4 , wherein encoding each of the plurality of sub-graphs as a vector uses an encoding function that is implemented as a recurrent neural network.
6. The method of claim 5 , wherein the recurrent neural network is trained using an autoencoder system.
7. The method of claim 4 , wherein determining the vector representation of the first syscall graph further includes fusing multiple subgraph vectors into a single vector representation for the first syscall graph.
8. The method of claim 1 , determining that the first application is behaving abnormally includes determining that the first application has a below-threshold similarity to the second applications.
9. The method of claim 1 , wherein the security action includes an action selected from the group consisting of terminating the first application, quarantining the first application, and performing a security analysis on the first application.
10. A system for detecting abnormal application behavior, comprising:
a hardware processor;
a memory, configured to store a computer program that, when executed by the hardware processor, is configured to implement a graph vector model that determines a vector representation of a first syscall graph that is generated by a first application, the vector representation including a representation of a distribution of subgraphs of the first syscall graph; and
a security console, configured to compare the vector representation of the first syscall graph to one or more second syscall graphs that are generated by respective second applications to determine respective similarity scores, to determine that the first application is behaving abnormally based on the similarity scores, and to perform a security action responsive to the determination that the first application is behaving abnormally.
11. The system of claim 10 , wherein the graph vector model is further configured to sample a plurality of sub-graphs from the first syscall graph.
12. The system of claim 11 , wherein the graph vector model is further configured to perform a walk with earliest visit time.
13. The system of claim 11 , wherein the graph vector model is further configured to encode each of the plurality of sub-graphs as a vector.
14. The system of claim 13 , wherein the graph vector model is further configured to use an encoding function that is implemented as a recurrent neural network.
15. The system of claim 14 , wherein the recurrent neural network is trained using an autoencoder system.
16. The system of claim 13 , wherein the graph vector model is further configured to fuse multiple subgraph vectors into a single vector representation for the first syscall graph.
17. The system of claim 10 , wherein the security console is further configured to determine that the first application has a below-threshold similarity to the second applications.
18. The system of claim 10 , wherein the security console is further configured to perform a security action that includes one or more actions selected from the group consisting of termination of the first application, quarantine of the first application, and a security analysis on the first application.