Cross-Domain Data Filtering with Machine Learning Classifiers

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing cross-domain solutions are less effective in detecting sophisticated malware and covert channels due to their reliance on signature-based approaches, which require frequent updates and fail to identify zero-day attacks.

Innovation Solution

Implementing a machine learning-based malicious content detection approach that uses multiple classifiers trained with labeled data, including both publicly identified and domain-specific malware, to filter data across security domains, combined with traditional filtering methods for redundant and independent verification.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Reliability

If signature-based detection is used, then known malware can be detected, but zero-day attacks and sophisticated malware cannot be detected

Engineering Contradiction:
Improvedetection effectivenessVSAvoidability to detect novel threats
Core Design Contradiction:
ReliabilityVSAdaptability or versatility

Solution Approach 1:

The system performs preliminary machine learning training with domain-specific malware samples before deployment. This pre-training enables the classifier to recognize domain-specific attack patterns and zero-day threats before they encounter the cross-domain solution, overcoming the limitation of signature-based detection that only identifies known threats.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The patent replaces the mechanical signature-matching system with a machine learning-based classification system. Instead of comparing file signatures against a database of known malware signatures, the system uses trained machine learning classifiers to analyze and classify files, enabling detection of novel and domain-specific threats that lack existing signatures.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

2Reliability

If virus definition updates are performed frequently, then detection accuracy improves, but update cycles in classified environments are longer than desired

Engineering Contradiction:
Improvedetection accuracyVSAvoidupdate cycle duration
Core Design Contradiction:
ReliabilityVSLoss of time

Solution Approach 1:

The system performs preliminary machine learning training with domain-specific malware samples before deployment. This pre-training enables the classifier to recognize domain-specific attack patterns and zero-day threats before they encounter the cross-domain solution, overcoming the limitation of signature-based detection that only identifies known threats.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

The machine learning model is trained offline with domain-specific malware samples and then deployed for autonomous classification. Once trained, the system can independently detect novel and domain-specific threats without requiring frequent updates, as the trained model generalizes to unseen threats within the domain.

Inventive Principle:
Principle #25Self-service

3Reliability

If multiple machine learning classifiers are used, then detection capability against diverse threats improves, but system complexity increases

Engineering Contradiction:
Improvedetection capabilityVSAvoidclassifier system complexity
Core Design Contradiction:
ReliabilityVSDevice complexity

Solution Approach 1:

The system segments the detection task by deploying different machine learning classifiers at different domains (source domain and destination domain). Each classifier is trained with domain-specific malware samples, allowing specialized detection for each domain while maintaining overall system manageability through modular deployment.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The machine learning classifiers serve multiple functions: they detect known malware, identify zero-day threats, and recognize domain-specific attack patterns. This multi-functionality reduces the need for separate specialized systems while maintaining comprehensive detection capability across diverse threat types.

Inventive Principle:
Principle #6Universality (Multi-functionality)

Data Source

PatentEP3599566B1Machine learning data filtering in a cross-domain environment
Publication Date: 2022.04.06 THE BOEING CO
  • EP3599566B1 patent drawingFigure 1
  • EP3599566B1 patent drawingFigure 2
  • EP3599566B1 patent drawingFigure 3

AI summary

Techniques for transferring data from a first domain to a second domain in a cross-domain environment are presented. The techniques can include accepting computer readable data in the first domain for transfer to the second domain, passing the computer readable data to a first machine learning classifier at the first domain trained with at least malware files publicly identified as malicious, passing the computer readable data to a second machine learning classifier at the first domain trained with at least malware files specific to the first domain, and transferring the computer readable data to a destination in the second domain.