Rethinking Saliency Maps: A Cognitive Human Aligned Taxonomy and Evaluation Framework for Explanations

2025-11-17

AI Summary

Saliency maps are widely used for visual explanation in deep learning, yet a lack of consensus between their intended explanatory purpose and users’ query types leads to inaccurate evaluation and limited applicability. Method: We propose the Reference Frame × Granularity (RF×G) taxonomy—the first systematic framework disentangling explanation intent (pointwise vs. contrastive) from semantic granularity (class-level vs. group-level)—and uncover cognitive biases inherent in existing faithfulness metrics. Based on this theory, we design four novel faithfulness evaluation metrics and establish a cross-dimensional benchmark across ten saliency methods, four model architectures, and three datasets. Results: Experiments reveal significant deficiencies in mainstream methods for contrastive explanation and group-level semantic modeling. This work provides an interpretable AI framework aligned with human cognition and delivers reproducible, theoretically grounded evaluation tools.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Humans and AI: Explainable AI (XAI) for Human Understanding
• Natural Language Processing: Interpretability, Analysis, and Evaluation of NLP Models

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Web Mining and Content Analysis: Robustness and generalizability of Web mining methods

Abstract

Saliency maps are widely used for visual explanations in deep learning, but a fundamental lack of consensus persists regarding their intended purpose and alignment with diverse user queries. This ambiguity hinders the effective evaluation and practical utility of explanation methods.We address this gap by introducing the Reference-Frame $ imes$ Granularity (RFxG) taxonomy, a principled conceptual framework that organizes saliency explanations along two essential axes:Reference-Frame: Distinguishing between pointwise ("Why this prediction?") and contrastive ("Why this and not an alternative?") explanations.Granularity: Ranging from fine-grained class-level (e.g., "Why Husky?") to coarse-grained group-level (e.g., "Why Dog?") interpretations.Using the RFxG lens, we demonstrate critical limitations in existing evaluation metrics, which overwhelmingly prioritize pointwise faithfulness while neglecting contrastive reasoning and semantic granularity. To systematically assess explanation quality across both RFxG dimensions, we propose four novel faithfulness metrics. Our comprehensive evaluation framework applies these metrics to ten state-of-the-art saliency methods, four model architectures, and three datasets.By advocating a shift toward user-intent-driven evaluation, our work provides both the conceptual foundation and the practical tools necessary to develop visual explanations that are not only faithful to the underlying model behavior but are also meaningfully aligned with the complexity of human understanding and inquiry.

Problem

• Addressing ambiguity in saliency maps' purpose and user alignment
• Developing taxonomy to organize explanations by reference-frame and granularity
• Creating evaluation framework for faithful and human-aligned visual explanations

Innovation

• RFxG taxonomy organizes saliency explanations along two axes
• Four novel faithfulness metrics assess explanation quality systematically
• Framework evaluates saliency methods across models and datasets