Povećanje sigurnosti računalnih mreža potpornim učenjem

master's thesis

• Title (english): Reinforcement learning for computer network security
• Author: Ante Žužul
• Mentor: Domagoj Jakobović (mentor)
• Committee member: Domagoj Jakobović (predsjednik povjerenstva)
• Committee member: Marko Đurasević (član povjerenstva)
• Committee member: Stjepan Groš (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2021-07-16, Croatia
• Scientific / art field, discipline and subdiscipline: TECHNICAL SCIENCES; Computing

Abstract

Cyber napadi sve su inovativniji i učestaliji. Stoga se javlja potreba za novim strategijama osiguranja računalnih mreža. Simulirana okruženja omogućavaju razvoj obrambenih taktika u kontroliranim uvjetima s jasnom definiranim parametrima. Implementiran je simulator računalne mreže, napadača na mrežu i branitelja mreže. Cilj napadača je osvojiti što više čvorova u mreži, dok ga branitelj u tome pokušava spriječiti. Potporno učenje zbog svojih karakteristika omogućava stjecanje novih znanja kroz simulacije. Dubokim Q-učenjem napadač uspijeva osvojiti većinu čvorova računalne mreže. Suparničkim dubokim Q-učenjem branitelj uspješno pobjeđuje iskusnog i neiskusnog napadača. Braniteljeve mogućnosti teško je definirati te su se pokazale poprilično ograničenima.

Abstract (english)

Cyber attacks are becoming more innovative and frequent. Therefore, new computer network security strategies are needed. Simulated environments allow the development of defensive tactics in controlled conditions with clearly defined parameters. A computer network simulator with its attacker and defender has been implemented. Attacker and defender compete against each other. Attacker's goal is to win as many nodes in the network as possible, while defender tries to prevent him from doing so. Due to its characteristics, reinforcement learning allows the acquisition of new knowledge through simulations. Using deep Q-learning, attacker manages to exploit most nodes of a computer network. Using adversarial deep Q-learning, defender successfully defeats experienced and inexperienced attacker. Defender's capabilities are difficult to define and have proven to be quite limited.

• Keywords: sigurnost; računalna mreža; potporno učenje; suparničko učenje
• Keywords (english): security; computer network; reinforcement learning; adversarial learning
• Language: croatian
• URN:NBN: urn:nbn:hr:168:574378
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-01-13 14:21:12