NETWORK ANOMALY DETECTION

摘要

Methods, systems, and apparatus, including computer programs encoded on computer storage media, for determining network related anomaly scores. One of the methods includes generating a network map including at least a plurality of network nodes and a plurality of edges that indicate communications paths between the plurality of network nodes, obtaining first data indicating network activity over the edges and between the plurality of network nodes for a first time period, generating a model of expected network activity over the edges and between the plurality of network nodes for a future time period using the network map and the first data, obtaining second data indicating network activity over the edges and between the plurality of network nodes for a second time period, and determining an anomaly score using a comparison between the second data and the model of expected network activity.

主权项

1. A computer implemented method, comprising:
generating, by one or more computers, a network map comprising at least a plurality of network nodes and a plurality of edges that indicate communications paths between the plurality of network nodes; obtaining, by at least one of the one or more computers, network node information comprising an indication of a node type for each of the plurality of network nodes and network activity data indicating typical network activity for each of the node types; obtaining, by at least one of the one or more computers, first data indicating network activity over the edges and between the plurality of network nodes for a first time period; generating, by at least one of the one or more computers, a model of expected network activity over the edges and between the plurality of network nodes for a future time period using the network map, the network node information, and the first data; obtaining, by at least one of the one or more computers, second data indicating network activity over the edges and between the plurality of network nodes for a second time period; and determining, by at least one of the one or more computers, a node anomaly score for each of at least some of the plurality of network nodes using a comparison between the second data and the model of expected network activity.