Unrestrained Simplex Denoising treats discrete data generation as a non-Markovian process on the probability simplex.
March 31, 2026
Original Paper
Unrestrained Simplex Denoising for Discrete Data. A Non-Markovian Approach Applied to Graph Generation
arXiv · 2603.28572
The Takeaway
It avoids the abrupt state changes of discrete diffusion by operating in a continuous probability space, consistently outperforming state-of-the-art graph generation baselines.
From the abstract
Denoising models such as Diffusion or Flow Matching have recently advanced generative modeling for discrete structures, yet most approaches either operate directly in the discrete state space, causing abrupt state changes. We introduce simplex denoising, a simple yet effective generative framework that operates on the probability simplex. The key idea is a non-Markovian noising scheme in which, for a given clean data point, noisy representations at different times are conditionally independent.