TDGNet: Hallucination Detection in Diffusion Language Models via Temporal Dynamic Graphs

TL;DR

TDGNet introduces a novel temporal dynamic graph approach for detecting hallucinations in diffusion language models by analyzing evolving token attention graphs over the denoising process, improving robustness and accuracy.

Contribution

The paper presents TDGNet, a new framework that leverages temporal dynamic graphs to detect hallucinations in diffusion language models, addressing limitations of previous single-pass detectors.

Findings

Achieves higher AUROC scores on QA benchmarks.

Effective in single-pass inference with modest computational overhead.

Highlights importance of temporal reasoning in hallucination detection.

Abstract

Diffusion language models (D-LLMs) offer parallel denoising and bidirectional context, but hallucination detection for D-LLMs remains underexplored. Prior detectors developed for auto-regressive LLMs typically rely on single-pass cues and do not directly transfer to diffusion generation, where factuality evidence is distributed across the denoising trajectory and may appear, drift, or be self-corrected over time. We introduce TDGNet, a temporal dynamic graph framework that formulates hallucination detection as learning over evolving token-level attention graphs. At each denoising step, we sparsify the attention graph and update per-token memories via message passing, then apply temporal attention to aggregate trajectory-wide evidence for final prediction. Experiments on LLaDA-8B and Dream-7B across QA benchmarks show consistent AUROC improvements over output-based, latent-based, and static-graph baselines, with single-pass inference and modest overhead. These results highlight the importance of temporal reasoning on attention graphs for robust hallucination detection in diffusion language models.