DeMTS: Denoising Trajectories as Multivariate Time Series for Hallucination Detection in Diffusion Language Models
A novel framework for detecting hallucinations in diffusion models using multivariate temporal analysis.
Diffusion large language models (D-LLMs) have emerged as a promising paradigm for text generation. However, similar to autoregressive LLMs, D-LLMs remain vulnerable to hallucinations, where fluent outputs may contain factually incorrect or unsupported content. Although existing h...
ArXiv CS.CL
Source
Generated insight by OnAirToday AI
By treating denoising trajectories as multivariate time series, DeMTS captures complex patterns of hallucination that traditional dimension-compression methods ignore. This approach significantly enhances the reliability of diffusion language models by leveraging the structural dependencies within the generation process.
Comments
Keep comments relevant and respectful. No links, advertising, self-promotion, or spam.