arXiv:2605.14255v3 Announce Type: replace Abstract: Industrial visual inspection systems increasingly rely on deep classifiers whose heatmap explanations may appear visually plausible while failing to identify the image regions that actually drive model decisions. This paper operationalizes an architecture-aware explanation audit protocol grounded in the native-readout hypothesis: the perturbation-based faithfulness of an explanation method is bounded by its structural distance from the model's native decision mechanism. On WM-811K wafer maps (9 classes, 172k images) under a three-seed zero-fi
The increasing reliance on AI in critical industrial applications, particularly visual inspection, necessitates robust methods for validating model decisions and ensuring reliability.
This research addresses the core problem of AI explainability, crucial for deploying trustworthy AI in high-stakes industrial settings, impacting regulatory acceptance and operational safety.
The proposed audit protocol offers a structured approach to evaluate AI explanation methods, moving beyond merely 'visually plausible' explanations to those truly reflective of model decision-making.
- · Industrial AI solution providers
- · Manufacturing sector (quality control)
- · AI safety researchers
- · Regulatory bodies for AI
- · Companies with opaque AI systems
- · Explanation methods that lack faithfulness guarantees
Improved reliability and trustworthiness of AI systems in industrial visual inspection.
Faster adoption of AI in critical manufacturing processes due to enhanced explainability and auditability.
Potential for new industry standards and regulatory frameworks for AI explainability across various high-assurance domains.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.LG