Introduction
We are pleased to announce our latest contribution to the IDUNN project with the publication of our new article in IEEE Access. Titled “Introducing PAC-GPT: A Novel Framework for Synthetic Data Generation in Cybersecurity”, this paper addresses the critical challenge of data scarcity in the development of machine learning models for cybersecurity applications.
Abstract
The application of machine learning models in cybersecurity has surged in recent years. However, the effectiveness of these models is heavily dependent on the quality and breadth of the training data they use. The scarcity of realistic datasets in the cybersecurity field presents a significant hurdle, as many available datasets are outdated or insufficient. Organizations are often reluctant to share data due to privacy concerns or potential threats to trade secrets.
Key Contribution: PAC-GPT
To address this challenge, we introduce PAC-GPT, a novel framework leveraging OpenAI’s Generative Pre-trained Transformer 3 (GPT-3) to generate reliable synthetic data for machine learning methods. PAC-GPT comprises two core components:
- Flow Generator: Captures and regenerates patterns in a series of network packets.
- Packet Generator: Generates individual network packets based on the network flow.
Methodology
We propose a packet generator based on Large Language Model (LLM) chaining and evaluate its performance using metrics such as loss, accuracy, and success rate. Our findings indicate that transformers are a suitable approach for synthetic packet generation, requiring minimal fine-tuning.
Tools and Applications
To facilitate access to this innovative data generation strategy, we have developed a streamlined command line interface (CLI) tool. This tool enables professionals from various disciplines to seamlessly integrate PAC-GPT into their workflows, enhancing the development of industry-grade cybersecurity tools.
Publication Details
- Published in: IEEE Access (Volume 11)
- Pages: 114936 – 114951
- Date of Publication: October 18, 2023
- Electronic ISSN: 2169-3536
- DOI: 10.1109/ACCESS.2023.3325727
- Publisher: IEEE
Authors
Danial Khosh Kholgh, Panos Kostakos
Conclusion
PAC-GPT represents a significant advancement in synthetic data generation for cybersecurity. By addressing the limitations of existing datasets, PAC-GPT enables the development of more effective machine learning models, ultimately contributing to more robust and secure cybersecurity solutions.
For a detailed exploration of our framework and findings, please refer to the full article published.