Causality-Preserving Time Series Mixing for Attack Training Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Generating appropriate time series content for training adaptive systems, such as neural networks, is challenging due to the difficulty in obtaining representative data for various scenarios, especially for malicious attacks and anomalous events, as injecting malicious payloads into software applications is arduous and resource-intensive.
Innovation Solution
Collecting runtime system trace content from isolated instances of software and malicious code execution, and performing causality-preserving time series mixing to generate training content, maintaining causal relationships between independently gathered time series to create diverse and representative training data.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If malicious payloads are injected into software applications to generate training data, then representative training content for malicious attacks can be obtained, but the process becomes arduous and resource-intensive
Solution Approach 1:
The patent segments the training data generation process into two independent parts: (1) collecting runtime system trace content from isolated instances of software and malicious code execution, and (2) performing causality-preserving time series mixing to combine these traces. This segmentation eliminates the need for injecting malicious payloads into full software applications, thereby reducing resource requirements while maintaining training data quality
Solution Approach 2:
The patent introduces causality-preserving time series mixing as an intermediary process that combines runtime system traces from isolated executions. This mixing technique preserves causal relationships between system events while generating diverse training scenarios, serving as a mediator that replaces the need for direct malicious payload injection into applications
2Adaptability or versatility
If diverse training scenarios are generated through extensive data injection, then adaptive systems can be trained for various attack types, but the cost and complexity increase significantly
Solution Approach 1:
The patent creates copies of runtime system traces from isolated executions and combines them through causality-preserving mixing. Instead of generating entirely new training data through complex injection processes for each attack scenario, the system copies and recombines existing trace data, maintaining diversity while simplifying the generation process
Solution Approach 2:
The causality-preserving time series mixing mechanism serves multiple functions: it diversifies training scenarios, preserves causal relationships, and generates representative data for various attack types simultaneously. This universal approach replaces multiple specialized data injection processes with a single multi-functional mixing operation
Data Source
AI summary
Subject matter disclosed herein may relate to time-series mixing for adaptive system training and may relate more particularly to causality-preserving time series mixing for adaptive system training.