Exploring the influence of the choice of prior of the Variational Auto-Encoder on cybersecurity anomaly detection

Tengfei Yang, Yuansong Qiao, Brian Lee

    Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

    Abstract

    The Variational Auto-Encoder (VAE) is a popular generative model as the variance inference in the latent layer, the prior is an important element to improve inference efficient. This research explored the prior in the VAE by comparing the Normal family distributions and other location-scale family distributions in three aspects (performance, robustness, and complexity) in order to find a suitable prior for cybersecurity anomaly detection. Suitable distributions can improve the detection performance, which was verified at UNSW-NB15 and CIC-IDS-2017.

    Original languageEnglish
    Title of host publicationARES 2024 - 19th International Conference on Availability, Reliability and Security, Proceedings
    PublisherAssociation for Computing Machinery
    ISBN (Electronic)9798400717185
    DOIs
    Publication statusPublished - 30 Jul 2024
    Event19th International Conference on Availability, Reliability and Security, ARES 2024 - Vienna, Austria
    Duration: 30 Jul 20242 Aug 2024

    Publication series

    NameACM International Conference Proceeding Series

    Conference

    Conference19th International Conference on Availability, Reliability and Security, ARES 2024
    Country/TerritoryAustria
    CityVienna
    Period30/07/242/08/24

    Keywords

    • Cybersecurity anomaly detection
    • Latent representation
    • Location-scale Distribution
    • Normal family Distribution
    • Prior distribution
    • Variational Auto-Encoder

    Fingerprint

    Dive into the research topics of 'Exploring the influence of the choice of prior of the Variational Auto-Encoder on cybersecurity anomaly detection'. Together they form a unique fingerprint.

    Cite this