Attribution Explanations for Deep Neural Networks: A Theoretical Perspective

2025-08-11

AI Summary

This study addresses the **faithfulness evaluation challenge** in deep neural network attribution methods—i.e., whether existing techniques accurately reflect the true contribution of input features to model predictions. Confronting key bottlenecks—including methodological heterogeneity, weak theoretical foundations, and the absence of unified, rigorous evaluation criteria—we propose an integrated research framework comprising *theoretical unification*, *principled clarification*, and *rigorous evaluation*. Through formal mathematical modeling and analysis, we systematically unify major attribution methods, exposing their fundamental commonalities and essential distinctions. We establish, for the first time, the necessary and sufficient conditions for faithfulness and provide formal proofs of key theoretical properties. The work yields the first attribution theory framework that bridges deep theoretical rigor with practical applicability, offering a principled foundation for method selection, trustworthy AI design, and the development of novel attribution algorithms.

Technology Category

• Computer Vision: Adversarial Attacks & Robustness
• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Other Foundations of Reasoning under Uncertainty

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Graph Algorithms and Modeling for the Web: Graph neural networks and deep learning approaches for Web-related graphs

Abstract

Attribution explanation is a typical approach for explaining deep neural networks (DNNs), inferring an importance or contribution score for each input variable to the final output. In recent years, numerous attribution methods have been developed to explain DNNs. However, a persistent concern remains unresolved, i.e., whether and which attribution methods faithfully reflect the actual contribution of input variables to the decision-making process. The faithfulness issue undermines the reliability and practical utility of attribution explanations. We argue that these concerns stem from three core challenges. First, difficulties arise in comparing attribution methods due to their unstructured heterogeneity, differences in heuristics, formulations, and implementations that lack a unified organization. Second, most methods lack solid theoretical underpinnings, with their rationales remaining absent, ambiguous, or unverified. Third, empirically evaluating faithfulness is challenging without ground truth. Recent theoretical advances provide a promising way to tackle these challenges, attracting increasing attention. We summarize these developments, with emphasis on three key directions: (i) Theoretical unification, which uncovers commonalities and differences among methods, enabling systematic comparisons; (ii) Theoretical rationale, clarifying the foundations of existing methods; (iii) Theoretical evaluation, rigorously proving whether methods satisfy faithfulness principles. Beyond a comprehensive review, we provide insights into how these studies help deepen theoretical understanding, inform method selection, and inspire new attribution methods. We conclude with a discussion of promising open problems for further work.

Problem

• Evaluating faithfulness of attribution methods for DNNs
• Unifying heterogeneous attribution methods theoretically
• Providing theoretical rationale for attribution methods

Innovation

• Theoretical unification for comparing attribution methods
• Theoretical rationale clarifying method foundations
• Theoretical evaluation proving faithfulness principles