Soft Model Assertions Using Severity Scores for ML Error Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing model assertion systems for Machine Learning (ML) models in autonomous systems, such as autonomous vehicle control, rely on manual specification of potential errors, which can be complex and prone to missing unknown errors.
Innovation Solution
A soft model assertion (SMA) system that monitors ML models to detect errors by generating data associations, priors, and application objective functions, and applies a severity score to automatically and accurately identify errors in ML model predictions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual specification of potential errors is used in model assertion systems, then the system can detect known errors, but the complexity increases and unknown errors may be missed
Solution Approach 1:
The system enables ML models to self-monitor and self-diagnose errors through automated assertion checking. The models generate their own predictions and have these predictions automatically evaluated against assertion conditions, eliminating the need for manual error specification while maintaining comprehensive error detection capability.
Solution Approach 2:
The patent replaces manual mechanical specification of error conditions with automated computational assertion checking. Instead of manually defining error boundaries and conditions, the system uses automated ML models to evaluate predictions against learned assertion patterns, substituting manual processes with intelligent automation.
2Reliability
If manual specification of potential errors is used, then the system can be implemented with existing methods, but it is prone to missing unknown errors
Solution Approach 1:
The system dynamically adapts to unknown error types by continuously learning from new data and adjusting assertion conditions. Rather than relying on static manual specifications, the ML models evolve their error detection capabilities by processing diverse prediction scenarios, enabling automatic adaptation to previously unseen error patterns.
Solution Approach 2:
The system changes the parameters of error detection by transforming fixed manual assertion thresholds into dynamic, learned parameters. The ML models automatically adjust detection sensitivity and criteria based on incoming data distributions, allowing the system to detect unknown errors by adapting its detection parameters rather than relying on predetermined settings.
3Adaptability or versatility
If automated error detection is implemented, then the system can detect unknown errors, but the computational complexity and processing requirements increase
Solution Approach 1:
The system applies partial assertion checking by prioritizing critical assertion conditions and evaluating only the most relevant predictions in detail. Rather than exhaustively checking every possible error condition for every prediction, the system focuses computational resources on high-risk scenarios and uses approximate checking for lower-priority cases, reducing overall processing requirements while maintaining effective error detection.
Data Source
AI summary
Systems and methods are provided for implementing soft model assertions (SMA) system and techniques designed to monitor and improve Machine Learning (ML) model quality by to detecting errors within the one or more ML models. SMA techniques and systems are distinctly designed to leverage: 1) a user's ability to specify features over data; and 2) large, existing datasets of organizations, in a manner that can improve the accuracy and quality of predicting potential errors in Machine Learning (ML) models. A SMA system can include a controller device receiving predictions generated based on the ML models and output from the SMA system. The controller performs autonomous operations of the system in response to determining that the one or more detected errors within the one or more ML models yield a high certainty of errors in the predictions. The SMA system also includes a domain specific language and a severity score module.