Ataques a un Sistema de Detecci ́on de Intrusiones mediante Redes Generativas Adversarias
Loading...
Official URL
Full text at PDC
Publication date
2023
Advisors (or tutors)
Editors
Journal Title
Journal ISSN
Volume Title
Publisher
Citation
Abstract
La inteligencia artificial está experimentando un fuerte impacto en la sociedad, la potencia de estos algoritmos no para crecer y no paran de aparecer nuevas aplicaciones: clasificadores de imágenes, reconocedores de voz, chatbots, etc. La inteligencia artificial, como el resto de ramas de la inform ́atica, debe preocuparse por la seguridad de sus algoritmos y del tratamiento de datos. Este trabajo es un an ́alisis de los ataques adversarios, los cuales utilizan las herramientas de la inteligencia artificial contra sí misma. Si un modelo puede entrenarse con datos para detectar patrones y hacer predicciones también se puede entrenar para generar datos que puedan falsear estas predicciones. A lo largo de este proyecto se analizan distintas formas de realizar ataques adversarios centrándose principalmente en ataques a clasificadores de software malicioso. Además se da un ejemplo de implementación de ataque adversario a un sistema de detección de intrusos del que se analizarán los resultados.
The artificial intelligence is experiencing a boom in the society, the power of these algorithms is growing and new applications are constantly appearing: image classifiers, voice recognizers, chatbots, etc. Artificial intelligence, like all other branches of computer science, must be concerned about the security of its algorithms and data processing. This work is an analysis of adversarial attacks, which use the tools of artificial intelligence against itself. If a model can be trained with data to detect patterns and make predictions it can also be trained to generate data that can falsify these predictions. Throughout this project different ways of performing adversarial attacks are analyzed focusing mainly on attacking malware classifiers. In addition, an example implementation of an adversarial attack on an intrusion detection system is given and the results will be analyzed.
The artificial intelligence is experiencing a boom in the society, the power of these algorithms is growing and new applications are constantly appearing: image classifiers, voice recognizers, chatbots, etc. Artificial intelligence, like all other branches of computer science, must be concerned about the security of its algorithms and data processing. This work is an analysis of adversarial attacks, which use the tools of artificial intelligence against itself. If a model can be trained with data to detect patterns and make predictions it can also be trained to generate data that can falsify these predictions. Throughout this project different ways of performing adversarial attacks are analyzed focusing mainly on attacking malware classifiers. In addition, an example implementation of an adversarial attack on an intrusion detection system is given and the results will be analyzed.
Description
Trabajo de Fin de Grado en Ingeniería Informática, Facultad de Informática UCM, Departamento de Ingeniería de Software e Inteligencia Artificial (ISIA), Curso 2022/2023.