Please use this identifier to cite or link to this item: http://hdl.handle.net/10609/149599
Title: Anàlisi arrel IA: ciberseguretat pràctica
Author: Mata Melia, Arnau
Tutor: Farràs Ballabriga, Gerard  
Others: Isern-Deya, Andreu Pere  
Abstract: Using a methodological framework that integrates Large Language Models (LLMs) with Reinforcement Learning (RL) supervision (Question-Answer), a solution has been developed that includes APIs for interacting with the models and for training using Python language data. Despite technical challenges and time constraints in the design and execution of the project, which limited the acquisition of real datasets for intensive training of the models and their subsequent processing, promising results have been achieved. It has been observed that supervising the LLM model with an RL model enhances the accuracy of LLM responses, largely due to the RL model training based on rewards for correct answers. The containerization of the solution in a Docker image, which includes a small server that launches the API endpoints, facilitates the interaction and continuous training of the models in a local environment, allowing for uture improvements such as automated validation of training data and programmatic generation of training files. The findings suggest that establishing a learning cycle for both models in a real network environment, training with real incidents, could significantly improve the system's response capability. This research paves the way for future studies where the dynamic integration of LLMs and RL in real cybersecurity environments could be key for a more accurate and effective analysis of the root causes of security incidents.
Keywords: ciberseguretat
resposta a incidents
intel·ligència artificial
Document type: info:eu-repo/semantics/bachelorThesis
Issue Date: 9-Jan-2024
Publication license: http://creativecommons.org/licenses/by-nc-nd/3.0/es/  
Appears in Collections:Trabajos finales de carrera, trabajos de investigación, etc.

Files in This Item:
File Description SizeFormat 
arnmataTFG0124memoria.pdfMemòria del TFG3,59 MBAdobe PDFThumbnail
View/Open
Share:
Export:
View statistics

This item is licensed under aCreative Commons License Creative Commons