Data Validation for Error Detection in Multi-Dimensional Systems
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Complex computer systems face challenges in detecting errors in software code, as errors may not immediately manifest and can result in significant negative consequences, such as revenue loss, due to corrupted user-related data used for content delivery campaigns, where distinguishing normal changes from bugs is difficult and false alarms can occur.
Innovation Solution
A data validation process that tracks and analyzes changes in count distribution for attribute values across multiple users, setting thresholds based on historical analysis or statistical models to generate notifications for significant changes, and using machine learning classifiers to differentiate between natural and problematic changes.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If data validation tracks changes in count distribution for attribute values, then error detection capability is improved, but false alarms increase due to difficulty in distinguishing normal changes from bugs
Solution Approach 1:
The patent segments the data validation process into multiple independent components: (1) collecting historical data for each attribute value, (2) calculating separate statistical parameters (mean, standard deviation) for each attribute value, (3) computing individual significance scores for each attribute value change, and (4) aggregating scores only when multiple attribute values show concurrent changes. This segmentation allows the system to distinguish between isolated normal variations and systematic error patterns, reducing false alarms while maintaining error detection capability.
2Measurement precision
If thresholds are set based on historical analysis or statistical models, then measurement precision is improved, but device complexity increases due to machine learning classifiers
Solution Approach 1:
The patent implements feedback mechanisms where the system continuously collects validation results and uses them to refine statistical parameters. Historical data from previous validation runs is fed back into the statistical models to update mean values and standard deviations, allowing thresholds to adapt to changing data patterns over time. This feedback loop improves measurement precision without requiring complex machine learning classifiers, as the system learns from its own operational history.
Solution Approach 2:
The system performs self-service by automatically calculating statistical parameters and determining thresholds based on its own historical data, without requiring external configuration or complex external systems. The validation system uses its accumulated historical data to self-adjust its sensitivity and thresholds, reducing the need for manual tuning and simplifying the overall system architecture while maintaining high measurement precision.
3Productivity
If the system analyzes changes in user-related data, then productivity is improved through faster error detection, but loss of time increases due to days or weeks before errors manifest
Solution Approach 1:
The patent applies preliminary action by continuously monitoring and validating data in the background before errors manifest into visible problems. The system performs proactive validation checks on data pipelines, attribute value distributions, and count changes before they can cause downstream issues. By detecting anomalies in data patterns early through statistical analysis of attribute value distributions, the system identifies potential errors days or weeks before they would normally manifest, allowing preventive action to be taken and eliminating the time loss associated with delayed error detection.
Data Source
AI summary
Techniques for detecting errors in a computing system are provided. In one technique, first member data that comprises first attribute values of a particular attribute is stored. Each attribute value in the first attribute values is associated with a different member of a first plurality of members. Second member data that comprises second attribute values of the particular attribute is stored. Each attribute value in the second attribute values is associated with a different member of a second plurality of members. For each attribute value of the first attribute values, a first number of instances of the attribute value in the first member data is determined, a second number of instances of the attribute value in the second member data is determined, and based on the first number and the second number, a determination of whether a notification indicating the particular attribute will be generated is made.