Tangentially Aligned Integrated Gradients for User-Friendly Explanations

2025-03-11

AI Summary

To address the subjectivity and instability of baseline selection in Integrated Gradients (IG), which often leads to explanations misaligned with the underlying data manifold, this paper proposes an automatic baseline optimization method guided by **maximizing tangential alignment**. We formally define the alignment degree of explanation vectors within the Riemannian tangent space of the data manifold, derive the theoretical conditions under which IG vectors lie in this tangent space, and design a differentiable approximation enabling end-to-end optimization. Unlike conventional approaches relying on heuristic baselines (e.g., zero vector or dataset mean), our method requires no manual baseline specification and seamlessly integrates into the standard IG framework. Experiments on ImageNet and CIFAR-10 demonstrate substantial improvements in explanation consistency, stability, and human interpretability over zero-baseline IG, mean-baseline IG, and Grad-CAM, empirically validating the effectiveness of manifold-aware attribution.

Technology Category

• Computer Vision: Interpretability, Explainability, and Transparency
• Machine Learning: Learning with Manifolds
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Integrated gradients is prevalent within machine learning to address the black-box problem of neural networks. The explanations given by integrated gradients depend on a choice of base-point. The choice of base-point is not a priori obvious and can lead to drastically different explanations. There is a longstanding hypothesis that data lies on a low dimensional Riemannian manifold. The quality of explanations on a manifold can be measured by the extent to which an explanation for a point lies in its tangent space. In this work, we propose that the base-point should be chosen such that it maximises the tangential alignment of the explanation. We formalise the notion of tangential alignment and provide theoretical conditions under which a base-point choice will provide explanations lying in the tangent space. We demonstrate how to approximate the optimal base-point on several well-known image classification datasets. Furthermore, we compare the optimal base-point choice with common base-points and three gradient explainability models.

Problem

• Addresses black-box problem in neural networks using integrated gradients.
• Proposes optimal base-point choice for better explanation alignment.
• Measures explanation quality on low-dimensional Riemannian manifolds.

Innovation

• Maximizes tangential alignment for explanations
• Formalizes tangential alignment theoretical conditions
• Approximates optimal base-point on image datasets