Vladislav Smirnov

Publications  /  Uncertainty Quantification for Large Language Diffusion Models

Uncertainty Quantification for Large Language Diffusion Models

A. Vazhentsev, V. Smirnov, D. Li, M. Panov, T. Baldwin, A. Shelmanov

Accepted NeurIPS 2026

arXiv ↗Method figure

D-CoCoA-G workflow: an input produces a diffusion trajectory; semantic dynamics, trajectory statistics, and masked likelihood provide signals for uncertainty estimation.
D-CoCoA-G method overview. Open full-resolution figure (PDF) ↗

A detailed write-up of this work is coming soon. For now, the full paper is available on arXiv.

← Back to home