Diffusion-CAM: Faithful Visual Explanations for dMLLMs

TL;DR

Diffusion-CAM is a novel interpretability method designed specifically for diffusion-based multimodal large language models, providing accurate and faithful visual explanations of their non-autoregressive token generation process.

Contribution

It introduces the first tailored interpretability approach for dMLLMs, capturing both latent features and class-specific gradients to improve understanding of their parallel generation.

Findings

Diffusion-CAM outperforms state-of-the-art methods in localization accuracy.

It enhances visual fidelity of explanations for diffusion multimodal models.

The method effectively resolves spatial ambiguity and reduces confounders in explanations.

Abstract

While diffusion Multimodal Large Language Models (dMLLMs) have recently achieved remarkable strides in multimodal generation, the development of interpretability mechanisms has lagged behind their architectural evolution. Unlike traditional autoregressive models that produce sequential activations, diffusion-based architectures generate tokens via parallel denoising, resulting in smooth, distributed activation patterns across the entire sequence. Consequently, existing Class Activation Mapping (CAM) methods, which are tailored for local, sequential dependencies, are ill-suited for interpreting these non-autoregressive behaviors. To bridge this gap, we propose Diffusion-CAM, the first interpretability method specifically tailored for dMLLMs. We derive raw activation maps by differentiably probing intermediate representations in the transformer backbone, accordingly capturing both latent features and their class-specific gradients. To address the inherent stochasticity of these raw signals, we incorporate four key modules to resolve spatial ambiguity and mitigate intra-image confounders and redundant token correlations. Extensive experiments demonstrate that Diffusion-CAM significantly outperforms SoTA methods in both localization accuracy and visual fidelity, establishing a new standard for understanding the parallel generation process of diffusion multimodal systems.