Network intrusion detection with a novel hierarchy of distances between embeddings of hash IP addresses

作者：

Highlights：

• Network addresses (IP and Port) are high cardinality categorical variables.

• Novel method to encode a network address into low-dimensional embeddings.

• Estimation of the probability that two network addresses share a network flow.

• Hash and embedding transforms applied to a hierarchy of address components.

• Neural network trained under self-supervised framework.

摘要

•Network addresses (IP and Port) are high cardinality categorical variables.•Novel method to encode a network address into low-dimensional embeddings.•Estimation of the probability that two network addresses share a network flow.•Hash and embedding transforms applied to a hierarchy of address components.•Neural network trained under self-supervised framework.

论文关键词：Hash function,Self-supervised learning,Neural network,Network address embedding,Network intrusion detection

论文评审过程：Received 27 September 2020, Revised 11 January 2021, Accepted 20 February 2021, Available online 25 February 2021, Version of Record 3 March 2021.

论文官网地址：https://doi.org/10.1016/j.knosys.2021.106887