Denoising attribution maps through gradient analysis of critical parameters

• We revisit the concept of widely used attribution maps and share the observation that many gradient signals correspond to irrelevant features, obscuring the true drivers behind model decisions. • We introduce CriGrad , a new framework that improves the clarity of feature attribution maps by filtering out noise through critical parameter analysis, while maintaining the faithfulness of attribution maps. • CriGrad can be integrated as an add-on module in various attribution methods, and extensive experiments show that it significantly improves the reliability and explainability in 93% of the cases. Many post-hoc explainable feature attribution techniques analyze gradient propagation and understand decisions of deep learning models. However, conventional gradient analysis used to generate such maps can be noisy, potentially compromising the reliability of explanations. In this work, we introduce a robust method for computing feature attributions by identifying critical parameters and refining gradient propagation through these parameters. This method reduces the impact of non-critical parameters, mitigating the effect of feature leakage and randomized initialization, which introduce noise in attribution maps. We implemented this concept as an add-on module, called CriGrad , and evaluated its efficacy using three benchmarks and seven explainable models. Our results show that focusing on critical parameters improves explanability in 93% of the cases, demonstrating its effectiveness and improved reliability.