Using Neurobiological Frameworks for Anomaly Detection in System Log Streams
Gustaf Rydholm · KTH Publication Database DiVA (KTH Royal Institute of Technology) · 2018
Artificial Intelligence (AI) has shown enormous potential,and is predicted to be a prosperous fieldthat will likely revolutionise entire industries andbring forth a new industrial era. However, mostof today’s AI is either, as in deep learning, anoversimplified abstraction of how an actual mammalianbrains neural network function, or methodssprung from mathematics. But, with the foundationof the bold ideas of Vernon Mountcastle statedin 1978 about the neocortical functionality, newframeworks for creating true machine intelligencehave been developed, and continues to be.In this thesis, we study one such theory, calledHierarchical Temporal Memory (HTM).We use thisframework to build a machine learning model inorder to solve the task of detecting and classifyinganomalies in system logs belonging to Ericsson’scomponent based architecture applications.The results are then compared to an existing classifier,called Linnaeus, which uses classical machinelearning methods. The HTM model is able to showpromising capabilities of classifying system log sequenceswith similar results compared with the Linnaeusmodel. The HTM model is an appealing alternative,due to the limited need of computationalresources and the algorithms ability to effectivelylearn with “one-shot learning”.