HTTP Header Perturbation Generator for Robust Malicious PDU Detection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
ML models trained to detect malicious protocol data units (PDUs) are vulnerable to reordering and structure-based attacks, leading to evasion by cyber attackers, as they rely on biased training data that does not account for perturbations in PDU fields.
Innovation Solution
An adversarial training data generator applies perturbations like reordering, replacement, and grid search to mimic malicious attacks, focusing on high and medium importance fields, and builds a knowledge base to optimize training data, enhancing the robustness of ML models against such attacks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If ML models are trained on conventional training data without perturbations, then training data processing is simple and fast, but the models become vulnerable to reordering and structure-based attacks by cyber attackers
Solution Approach 1:
The system performs preliminary actions by generating adversarial training data with perturbations (reordering, replacement, grid search) before actual attack occurs. The adversarial training data generator creates perturbed versions of benign and malicious PDUs in advance, so when real attacks happen, the model has already been exposed to similar variations during training, preventing vulnerability to reordering and structure-based attacks.
Solution Approach 2:
The system converts the harmful effect of attack perturbations into a beneficial training mechanism. Instead of trying to prevent attackers from reordering or modifying PDU structures, the system uses these same perturbation techniques to generate training data that teaches the model to recognize malicious patterns despite structural variations. The harm of potential attack vectors becomes the basis for robust model training.
2Measurement precision
If ML models are trained with adversarial perturbations to improve robustness, then detection accuracy against malicious PDUs improves, but training time and computational resources increase
Solution Approach 1:
The system applies local quality by focusing perturbation generation on specific high-importance and medium-importance fields rather than treating all fields equally. The adversarial training data generator identifies which PDU fields are most critical for malicious detection and concentrates perturbation efforts on those fields, reducing the overall computational burden while maintaining detection accuracy on the most vulnerable aspects.
Solution Approach 2:
The system uses parameter changes by varying perturbation intensity and scope based on field importance. High-importance fields receive more aggressive perturbation (including grid search exhaustively), while low-importance fields receive minimal or no perturbation. This selective parameter adjustment optimizes the balance between training robustness and computational efficiency.
3Adaptability or versatility
If the training data generator pertubs all PDU fields, then coverage of attack vectors is complete, but processing efficiency decreases due to unnecessary perturbations of low importance fields
Solution Approach 1:
The system implements local quality by differentiating treatment based on field importance. The adversarial training data generator assigns different perturbation strategies to different fields: high-importance fields receive comprehensive perturbation (reordering, replacement, grid search), medium-importance fields receive selective perturbation, and low-importance fields are either excluded or receive minimal perturbation. This localized approach maintains attack vector coverage for critical fields while improving overall processing efficiency.
Data Source
AI summary
A modular adversarial training data generator (“generator”) generates adversarial training data for a machine learning (ML) model to detect malicious protocol data units (PDUs). The adversarial training data mimics high volume cyberattacks by perturbing PDUs to bypass malicious detection systems. For Hypertext Transfer Protocol (HTTP) PDUs, the generator rearranged, replaces, and grid searches values of HTTP header fields to generate the adversarial training data. The generator further biases grid search based on metrics for values of HTTP header fields that quantify impact of replacing the values on malicious verdicts by ML models trained on the adversarial training data.