Multi-layered Framework for ML Safety Critical System Design
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine learning-based safety critical systems face challenges in ensuring strict safety due to uncertainty in ML models and incomplete requirements specifications, particularly lacking sufficient guidance for systematically engineering data requirements involving various stakeholders.
Innovation Solution
A computing apparatus with a multi-layered framework is designed to provide guidance for a safety assurance process by ensuring data quality through quantified uncertainty, belief, and plausibility. The framework includes a problem layer for defining operational domains, safety critical goals, and risk factors, a data layer for specifying data requirements, and an evidence layer for evaluating data uncertainty.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If machine learning models are applied to safety critical systems, then system functionality and intelligence are improved, but uncertainty and safety assurance difficulty increase
Solution Approach 1:
The patent segments the safety assurance process into three distinct layers: problem layer (defining operational domain and safety goals), data layer (specifying data requirements), and evidence layer (evaluating data quality). This segmentation allows systematic handling of ML uncertainty by addressing each aspect separately with appropriate methods and stakeholders.
Solution Approach 2:
The patent performs preliminary actions by defining problem spaces, data requirements, and evaluation criteria before actual ML model deployment. The framework establishes operational domains, identifies safety critical goals, and specifies data requirements in advance, enabling proactive safety assurance rather than reactive validation.
2Reliability
If data requirements are systematically engineered involving various stakeholders, then safety assurance quality is improved, but process complexity increases
Solution Approach 1:
The patent divides the complex stakeholder collaboration process into three manageable layers with specific tasks and deliverables. Each layer has defined inputs and outputs, making the overall complex process tractable through structured decomposition into problem space exploration, data requirement specification, and evidence evaluation.
Solution Approach 2:
The patent introduces a vertical dimension to stakeholder collaboration by organizing activities across three hierarchical layers. This dimensional organization transforms complex multi-stakeholder interactions into structured layer-specific processes, where each layer addresses specific aspects with appropriate expertise.
3Reliability
If data quality is ensured through quantified uncertainty and belief, then ML model reliability is improved, but data engineering difficulty increases
Solution Approach 1:
The patent introduces an evidence layer as an intermediary between data collection and ML model training. This layer provides standardized evaluation mechanisms (uncertainty quantification, belief assessment, plausibility checking) that mediate between raw data and model requirements, simplifying the data engineering process through systematic evaluation protocols.
Solution Approach 2:
The patent transforms qualitative data quality assessment into quantitative parameters by introducing measurable metrics for uncertainty, belief, and plausibility. This parameter transformation enables objective data quality evaluation and comparison, making data engineering more systematic and less subjective.
4Measurement precision
If blind spots in training data are detected through intellectual diversity, then data quality is improved, but evaluation time increases
Solution Approach 1:
The patent performs preliminary identification of potential blind spots during the problem space exploration and data requirement specification phases. By proactively identifying areas of concern before full data collection and evaluation, the framework reduces the time needed for comprehensive blind spot detection while maintaining thoroughness through targeted assessment.
Data Source
AI summary
An embodiment relates to a design of a safety critical system, and more particularly, to a computing apparatus having a multi-layered framework including a problem layer, a data layer, and an evidence layer to provide guidance for a safety assurance process in designing a machine learning-based safety critical system, and a method of designing a safety critical system using the multi-layered framework.