Tag
This paper systematically evaluates how post-training quantization affects the interpretability of CNN models using Grad-CAM and LIME, revealing that classification accuracy is not a reliable indicator of interpretability stability and that architecture selection is critical for trustworthy deployment.
This paper introduces an architecture-aware explanation audit protocol for industrial visual inspection, demonstrating that the faithfulness of explanation methods is bounded by their structural compatibility with a model's native decision mechanism, using experiments on wafer map and anomaly detection datasets.