Manifold-Aligned Guided Integrated Gradients for Reliable Feature Attribution

TL;DR

The paper introduces MA-GIG, a novel method that constructs feature attribution paths in a learned latent space to produce more reliable and faithful explanations for deep neural networks.

Contribution

It proposes a manifold-aligned approach to Guided Integrated Gradients using a variational autoencoder to bias attribution paths toward the data manifold.

Findings

MA-GIG reduces off-manifold noise in attributions.

It outperforms prior path-based attribution methods.

Produces more faithful explanations across datasets.

Abstract

Feature attribution is central to diagnosing and trusting deep neural networks, and Integrated Gradients (IG) is widely used due to its axiomatic properties. However, IG can yield unreliable explanations when the integration path between a baseline and the input passes through regions with noisy gradients. While Guided Integrated Gradients reduces this sensitivity by adaptively updating low-gradient-magnitude features, input-space guidance still produces intermediate inputs that deviate from the data manifold. To address this limitation, we propose \emph{Manifold-Aligned Guided Integrated Gradients} (MA-GIG), which constructs attribution paths in the latent space of a pre-trained variational autoencoder. By decoding intermediate latent states, MA-GIG biases the path toward the learned generative manifold and reduces exposure to implausible input-space regions. Through qualitative and quantitative evaluations, we demonstrate that MA-GIG produces faithful explanations by aggregating gradients on path features proximal to the input. Consequently, our method reduces off-manifold noise and outperforms prior path-based attribution methods across multiple datasets and classifiers. Our code is available at https://github.com/leekwoon/ma-gig/.