Automated Audit Triggering via Provenance Data Monitoring
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current AI model audit processes are manual, slow, non-transparent, and error-prone, especially in complex and safety-critical environments, requiring an efficient and reliable automated solution for real-time validation and re-validation.
Innovation Solution
A computer-implemented method that monitors provenance data structures of AI models, automatically triggers audits based on detected changes, and adapts the audit process using metadata and predefined rules to ensure efficient and reliable auditing, reducing manual intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If manual audit processes are used for AI models, then flexibility and adaptability to complex contexts are maintained, but the process becomes slow, error-prone, and inefficient
Solution Approach 1:
The system enables self-service auditing by automatically monitoring provenance data structures and triggering audit processes without manual intervention. The audit system autonomously detects changes in AI models, training data, or evaluation data and initiates appropriate audit workflows, eliminating the need for manual audit triggering while maintaining comprehensive coverage.
Solution Approach 2:
The system implements continuous feedback loops by monitoring provenance data structures and automatically responding to detected changes. When changes are detected in AI models or their underlying data, the system triggers re-audit processes and notifies relevant stakeholders, ensuring that audit results are及时反馈 to maintain model reliability throughout the AI lifecycle.
2Productivity
If automated audit processes are implemented, then efficiency and speed are improved, but complexity of the system increases
Solution Approach 1:
The system performs preliminary actions by continuously monitoring and analyzing provenance data structures before audit triggers are needed. The monitoring infrastructure is pre-configured to detect relevant changes, and audit workflows are pre-defined based on change types, enabling rapid automated response without complex real-time decision-making logic.
Solution Approach 2:
The audit system is segmented into distinct modular components: provenance data monitoring, change detection, audit triggering, audit execution, and notification. Each component operates independently with well-defined interfaces, reducing overall system complexity while enabling comprehensive automated auditing functionality.
3Reliability
If comprehensive audit coverage is maintained for all changes, then reliability is improved, but loss of time in audit processing increases
Solution Approach 1:
The system applies partial auditing by triggering re-audits selectively based on the type and significance of detected changes. Not all provenance data changes require full audit processes - the system intelligently determines the appropriate audit scope based on change characteristics, performing comprehensive audits only when necessary while maintaining reliability for critical changes.
Data Source
Figure 1
Figure 2~3
Figure 4
AI summary
There is provided a computer implemented method, a platform and a system configured to perform the method, a computer-readable medium and a computer program comprising instructions to carry out the method, the method comprising: monitoring a provenance data structure of one or more algorithms and/or of an artefact generated by one or more algorithms, the provenance data structure comprising provenance data on which the one or more algorithms and/or the artefact are based and, preferably, the one or more algorithms and/or the artefact; determining a change of the monitored provenance data structure; and automatically triggering an audit process for the one or more algorithms and/or the artefact based on the determined change of the monitored provenance data structure.