Extracting dependencies between network assets using deep learning

In a computing system, a method comprising:

• receiving, at the computing system, network flow information summarizing packet flow between pairs of network assets over one or more time periods;
• with the computing system, identifying dependencies between multiple network assets based on evaluation of the network flow information, wherein the evaluation includes:
• pre-processing the network flow information to produce input vectors;
• using deep learning to extract patterns in the input vectors, at least in part by applying a neural network to reveal similarities between the input vectors and producing output vectors, and by summarizing the output vectors, wherein the deep learning is machine learning that models high-level abstractions in the input vectors and the summarizing comprises:
• performing hierarchical clustering on weight vectors produced by the neural network;
• associating the clustered weight vector with corresponding details about the network flow information;
• and
• determining the dependencies based on the patterns;
• and
• with the computing system, outputting results of the identifying the dependencies, the results comprising identifications of dependencies between network assets.

A computing system comprising a processor and memory, wherein the memory comprises computer-executable instructions for implementing a network analysis tool, the computer-executable instructions specifying operations comprising:

• receiving network flow information summarizing packet flow between pairs of network assets over one or more time periods;
• pre-processing the network flow information to produce input vectors, the pre-processing comprising:
• combining instances of temporally proximate co-occurrences within the network flow information into higher-order combinations;
• encoding the higher-order combinations as the input vectors;
• and
• annotating the input vectors with details about the network flow information;
• and
• extracting patterns in the input vectors using deep learning, wherein the deep learning is machine learning that models high-level abstractions in the input vectors;
• determining, based on the patterns, dependencies between multiple network assets;
• and
• outputting data setting forth dependencies between network assets.

One or more non-transitory computer-readable media storing computer-executable instructions for causing a computing system, when programmed thereby, to perform:

• receiving network flow information summarizing packet flow between pairs of network assets over one or more time periods;
• identifying dependencies between multiple network assets based on evaluation of the network flow information, wherein the dependencies are identified as recurrent temporal sequence-based patterns, each of the recurrent temporal sequence-based patterns being a randomly-initiated, repeating sequence of time-dependent elements, and the evaluation includes:
• pre-processing the network flow information to produce input vectors;
• using deep learning to extract patterns in the input vectors, wherein the deep learning is machine learning that models high-level abstractions in the input vectors;
• and
• determining the dependencies based on the patterns;
• and
• outputting results of identifying the dependencies, the results comprising identifiers of dependencies between network assets.

One or more non-transitory computer-readable media storing computer-executable instructions for causing a computing system, when programmed thereby, to perform:

• receiving network flow information summarizing packet flow between pairs of network assets over one or more time periods;
• identifying dependencies between multiple network assets based on evaluation of the network flow information, wherein the evaluation includes:
• pre-processing the network flow information to produce input vectors, wherein each of the input vectors includes, for a given time period, information indicating an extent of co-occurrence of activity between network assets in various combinations, selected from among the multiple network asserts, during the given time period;
• using deep learning to extract patterns in the input vectors, wherein the deep learning is machine learning that models high-level abstractions in the input vectors and comprises:
• applying a neural network to reveal similarities between the input vectors, wherein the applying the neural network produces output vectors;
• summarizing the output vectors, the summarizing making the identification of the dependencies more robust over time;
• and
• determining the dependencies based on the patterns;
• and
• outputting results of identifying the dependencies, the results comprising identifiers of dependencies between network assets.