Multivariate network structure anomaly detector
A multivariate anomaly detector can detect a cyber-attack using incremental malicious actions distributed across multiple devices in a network. A multivariate anomaly detector can collect input data describing communication connections between devices in the network. The multivariate anomaly detector can group the input data into a graph data batch based on a fixed batch increment of time to identify incremental actions. The multivariate anomaly detector can calculate a multivariate centrality score for two or more devices based on the graph data batch describing device centrality to the network. The multivariate anomaly detector can identify whether the two or more devices are in an anomalous state from normal device network interactions based on the multivariate centrality score to identify malicious activity distributed across multiple devices in the network. The multivariate anomaly detector can identify a cyber-attack upon identifying the incremental malicious actions distributed across multiple devices in the network.
1. A method for a cyber threat defense system, comprising:
detecting a cyber-attack using incremental malicious actions distributed across multiple devices in a network;
collecting input data describing communication connections between devices in the network;
grouping the input data into a graph data batch based on a fixed batch increment of time to identify incremental actions;
calculating a multivariate centrality score for two or more devices based on the graph data batch describing device centrality to the network;
identifying whether the two or more devices are in an anomalous state from normal device network interactions based on the multivariate centrality score in order to identify the malicious activity distributed across multiple devices in the network; and
generating a notification about the cyber-attack using the incremental malicious actions distributed across multiple devices in the network.
2. The method for the cyber threat defense system of claim 1 , further comprising:
comparing the multivariate centrality score to a multivariate centrality score history to identify the malicious activity distributed across multiple devices in the network.
3. The method for the cyber threat defense system of claim 1 , further comprising:
generating a graph with nodes to represent devices in the network and edges of the graph to represent connections between the devices in the network.
4. The method for the cyber threat defense system of claim 3 , further comprising:
calculating the multivariate centrality score for two or more nodes representing two or more devices based on the graph data batch describing device centrality to the network using a different heuristic for each variate of the multivariate centrality score in order to identify at least a first node and a second node both acting in the anomalous state from the normal device network interactions.
5. The method for the cyber threat defense system of claim 1 , further comprising:
using the graph data batch to calculate a network binding score describing a connection density of a first node with other nodes in the network as a variate of the multivariate centrality score.
6. The method for the cyber threat defense system of claim 1 , further comprising:
using the graph data batch to calculate an eigenvector centrality score describing an influence of a first node on the network by weighting connections to other influential nodes as a variate of the multivariate centrality score.
7. The method for the cyber threat defense system of claim 1 , further comprising:
collecting a score history over a maturation period for each variate of the multivariate centrality score.
8. The method for the cyber threat defense system of claim 1 , further comprising:
computing a matrix of two-point correlations between each variate of the multivariate centrality score.
9. The method for the cyber threat defense system of claim 1 , further comprising:
using the graph data batch to calculate an access entropy score describing a diversity of visited nodes by a first node as a variate of the multivariate centrality score.
10. A non-transitory computer readable medium comprising computer readable code operable, when executed by one or more processing apparatuses in the cyber threat defense system to instruct a computing device to perform the method of claim 1 .
11. A multivariate anomaly detector, comprising:
a cyber-threat module configured to detect a cyber-attack using incremental malicious actions distributed across multiple devices in a network;
an ingestion module configured to collect input data describing communication connections between devices in the network;
a batch module configured to group the input data from the ingestion module into a graph data batch based on a fixed batch increment of time to identify incremental actions;
a centrality processing module configured to calculate a multivariate centrality score for two or more devices based on the graph data batch describing device centrality to the network from the batch module;
an anomaly detector module configured to identify whether the two or more devices are in an anomalous state from normal device network interactions based on the multivariate centrality score in order to identify the malicious activity distributed across multiple devices in the network; and
a user interface module to generate a notification about the cyber-attack using the incremental malicious actions distributed across multiple devices in the network.
12. The multivariate anomaly detector of claim 11 , wherein the anomaly detector module is further configured to compare the multivariate centrality score to a multivariate centrality score history to identify the malicious activity distributed across multiple devices in the network.
13. The multivariate anomaly detector of claim 11 , further comprising:
a graph detection module configured to generate a graph with nodes to represent devices in the network and edges of the graph to represent connections between the devices in the network.
14. The multivariate anomaly detector of claim 13 , wherein the centrality processing module is further configured to calculate the multivariate centrality score for two or more nodes representing two or more devices based on the graph data batch describing device centrality to the network using a different heuristic for each variate of the multivariate centrality score in order to identify at least a first node and a second node both acting in the anomalous state from the normal device network interactions.
15. The multivariate anomaly detector of claim 11 , wherein the centrality processing module is further configured use the graph data batch to calculate a network binding score describing a connection density of a first node with other nodes in the network as a variate of the multivariate centrality score.
16. The multivariate anomaly detector of claim 11 , wherein the centrality processing module is further configured to use the graph data batch to calculate an eigenvector centrality score describing an influence of a first node on the network by weighting connections to other influential nodes as a variate of the multivariate centrality score.
17. The multivariate anomaly detector of claim 11 , wherein the anomaly detector module is further configured to collect a score history over a maturation period for each variate of the multivariate centrality score.
18. The multivariate anomaly detector of claim 11 , wherein the anomaly detector module is further configured to compute a matrix of two-point correlations between each variate of the multivariate centrality score.
19. The multivariate anomaly detector of claim 11 , wherein the anomaly detector module is further configured to feed an anomaly radius for the multivariate centrality score into a probabilistic distribution such as a one-tail Cauchy distribution to compute a survival probability.
20. The multivariate anomaly detector of claim 11 , wherein the centrality processing module is configured use the graph data batch to calculate an access entropy score describing a diversity of visited nodes by a first node as a variate of the multivariate centrality score.