The Road Towards Autonomous Cybersecurity Agents: Remedies for Simulation Environments

Kód výsledku v IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F00216224%3A14330%2F24%3A00135174" target="_blank" >RIV/00216224:14330/24:00135174 - isvavai.cz</a>

Výsledek na webu

<a href="http://dx.doi.org/10.1007/978-3-031-54129-2_43" target="_blank" >http://dx.doi.org/10.1007/978-3-031-54129-2_43</a>

DOI - Digital Object Identifier

<a href="http://dx.doi.org/10.1007/978-3-031-54129-2_43" target="_blank" >10.1007/978-3-031-54129-2_43</a>

Jazyk výsledku

angličtina

Název v původním jazyce

The Road Towards Autonomous Cybersecurity Agents: Remedies for Simulation Environments

Popis výsledku v původním jazyce

One of the fundamental challenges in developing autonomous cybersecurity agents (AICA) is providing them with appropriate training environments for skills acquisition and evaluation. Current reinforcement learning (RL) algorithms rely on myriads of training runs to instill proper behavior, and this is reasonably achievable only within a simulated environment. In this paper, we explore the topic of simulation models and environments for RL and present an assessment framework to compare simulation models designed for simulating cyberattack scenarios. We examine four existing simulation tools, including a new one by the authors of the paper, and discuss their properties, particularly in terms of deployability, to support RL-based AICA. In the example of complex scenarios, we compare the two most sophisticated simulation tools and discuss their strengths.

Název v anglickém jazyce

The Road Towards Autonomous Cybersecurity Agents: Remedies for Simulation Environments

Popis výsledku anglicky

One of the fundamental challenges in developing autonomous cybersecurity agents (AICA) is providing them with appropriate training environments for skills acquisition and evaluation. Current reinforcement learning (RL) algorithms rely on myriads of training runs to instill proper behavior, and this is reasonably achievable only within a simulated environment. In this paper, we explore the topic of simulation models and environments for RL and present an assessment framework to compare simulation models designed for simulating cyberattack scenarios. We examine four existing simulation tools, including a new one by the authors of the paper, and discuss their properties, particularly in terms of deployability, to support RL-based AICA. In the example of complex scenarios, we compare the two most sophisticated simulation tools and discuss their strengths.

Druh

D - Stať ve sborníku

CEP obor

—

OECD FORD obor

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Projekt

<a href="/cs/project/VJ02010020" target="_blank" >VJ02010020: AI-Dojo: Multiagentní testbed pro výzkum a testování umělé inteligence v kyberbezpečnosti</a><br>

Návaznosti

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Rok uplatnění

2024

Kód důvěrnosti údajů

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Název statě ve sborníku

COMPUTER SECURITY. ESORICS 2023 INTERNATIONAL WORKSHOPS, CPS4CIP, PT II

ISBN

9783031541285

ISSN

0302-9743

e-ISSN

—

Počet stran výsledku

12

Strana od-do

738-749

Název nakladatele

Springer Nature Switzerland AG

Místo vydání

BERLIN

Místo konání akce

The Haague

Datum konání akce

28. 9. 2023

Typ akce podle státní příslušnosti

WRD - Celosvětová akce

Kód UT WoS článku

001212380000033