DAVE: Distribution-aware Attribution via ViT Gradient Decomposition

TL;DR

DAVE is a novel attribution method for Vision Transformers that decomposes gradients to produce stable, high-resolution explanations by isolating architecture-induced artifacts from meaningful signals.

Contribution

It introduces a mathematically grounded gradient decomposition technique tailored for ViTs, improving attribution stability and resolution.

Findings

Produces more stable attribution maps for ViTs

Separates architecture artifacts from true input signals

Enhances interpretability of Vision Transformer explanations

Abstract

Vision Transformers (ViTs) have become a dominant architecture in computer vision, yet producing stable and high-resolution attribution maps for these models remains challenging. Architectural components such as patch embeddings and attention routing often introduce structured artifacts in pixel-level explanations, causing many existing methods to rely on coarse patch-level attributions. We introduce DAVE \textit{(\underline{D}istribution-aware \underline{A}ttribution via \underline{V}iT Gradient D\underline{E}composition)}, a mathematically grounded attribution method for ViTs based on a structured decomposition of the input gradient. By exploiting architectural properties of ViTs, DAVE isolates locally equivariant and stable components of the effective input--output mapping. It separates these from architecture-induced artifacts and other sources of instability.