Few-Shot Learning based Anomaly Detection in Security Applications

Abstract

One of the most important areas of Computer Science is network security. Due to the high volume of data being transferred, the network is now open to different security-related attacks. As a result, it is now of utmost importance to detect network based attacks. In this paper, we propose a simple, adaptable and general framework aiming to detect Anomaly in Security Applications using Relation Network based Few-Shot Learning (RNFSL) model, which is cheaper to compute and needs less data compared to the traditional Machine Learning (ML) and Deep Learning (DL) models that are data hungry. We perform extensive experiments on a publicly available dataset where RNFSL on 1% data resulted in a loss value of 0.032 and also outperformed traditional ML and DL models.