mlsanityData to 14 Oct 2026

Does Uniform Discrete Diffusion Need Time?

arXiv 2609.30977 ↗First posted 25 Sep 2026cs.LG cs.AI cs.CLcited by 0

On the map: Diffusion models → Discrete and text diffusion

Abstract (arXiv)

Uniform discrete diffusion models (UDMs) commonly use explicit time conditioning, but we find that it can often be unnecessary in practice. In this paper, we first show that the population-optimal UDM predictor generally depends on time: time controls how much the model should trust the observed context. We then show that this dependence can become negligible in finite-data settings relevant to language. When a corrupted training sequence remains much closer to its original clean sequence than to competing training sequences, the empirical-optimal predictor is nearly insensitive to time over most of the diffusion trajectory, where the guarantee weakens toward the high-noise endpoint. Empirically, trained language UDMs exhibit limited time sensitivity over most of the trajectory, while time-agnostic predictors remain competitive with, and often outperform, time-conditioned models across datasets and training objectives. These results challenge the use of explicit time conditioning in UDMs: although the population optimum depends on time, explicitly conditioning on it may often be unnecessary in practice.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Discrete and text diffusion, one of 14 directions in Diffusion models. Builds on 33 of its 36 references that are in the map. First posted on arXiv 25 Sep 2026.

Builds on

33 of 36 references.

Closest ancestors · 15

Same direction, newest first

  1. 2026Unlocking Lossless Speedups in LLMs via Discrete Diffusion3 Sep
  2. 2026Scaling Beyond Masked Diffusion Language Models16 Feb
  3. 2025Scaling Behavior of Discrete Diffusion Language Models11 Dec
  4. 2025Improving Discrete Diffusion Unmasking Policies Beyond Explicit Reference Policies7 Oct
  5. 2025Dream 7B: Diffusion Large Language Models21 Aug
  6. 2025The Diffusion Duality12 Jun
  7. 2025Block Diffusion: Interpolating Between Autoregressive and Diffusion Language Models12 Mar
  8. 2025Large Language Diffusion Models14 Feb
  9. 2024Think While You Generate: Discrete Diffusion with Planned Denoising8 Oct
  10. 2024Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Sampling4 Sep
  11. 2024Simple and Effective Masked Diffusion Language Models11 Jun
  12. 2024Simplified and Generalized Masked Diffusion for Discrete Data6 Jun
3 more
  1. 2024Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data6 Jun
  2. 2023Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution25 Oct
  3. 2022DiffusionBERT: Improving Generative Masked Language Models with Diffusion Models28 Nov

Other references · 18

  1. 2020Denoising Diffusion Probabilistic Modelscited by 21,076
  2. 2019Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformercited by 14,599
  3. 2020Score-Based Generative Modeling through Stochastic Differential Equationscited by 9,122
  4. 2015Deep Unsupervised Learning using Nonequilibrium Thermodynamicscited by 7,058
  5. 2022Scalable Diffusion Models with Transformerscited by 6,054
  6. 2021Structured Denoising Diffusion Models in Discrete State-Spacescited by 1,719
  7. 2021Argmax Flows and Multinomial Diffusion: Learning Categorical Distributionscited by 574
  8. 2022A Continuous Time Framework for Discrete Denoising Modelscited by 451
  9. 2013One Billion Word Benchmark for Measuring Progress in Statistical Language Modelingcited by 308
  10. 2024Simple Guidance Mechanisms for Discrete Diffusion Modelscited by 155
  11. 2025The Principles of Diffusion Modelscited by 76
  12. 2025Is Noise Conditioning Necessary for Denoising Generative Models?cited by 44
6 more
  1. 2025Equilibrium Matching: Generative Modeling with Implicit Energy-Based Modelscited by 21
  2. 2025Why Masking Diffusion Works: Condition on the Jump Schedule for Improved Discrete Diffusioncited by 15
  3. 2026Uniform Diffusion Models Revisited: Leave-One-Out Denoiser and Absorbing State Reformulationcited by 13
  4. 2025Effective Test-Time Scaling of Discrete Diffusion through Iterative Refinementcited by 8
  5. 2026The Geometry of Noise: Why Diffusion Models Don't Need Noise Conditioningcited by 8
  6. 2026What Time Is It? How Data Geometry Makes Time Conditioning Optional for Flow Matchingcited by 2

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Optimizers for Diffusion Models: A Controlled Benchmark
  2. 2026Consistent Diffusion Language Models
  3. 2024Towards Faster Training of Diffusion Models: An Inspiration of A Consistency Phenomenon
  4. 2026Understanding and Accelerating the Training of Masked Diffusion Language Models
  5. 2026Less Uniform Discrete Diffusion is More Powerful and Scalable
  6. 2025Diffusion Language Models are Super Data Learners
  7. 2026Sumi: Open Uniform Diffusion Language Model from Scratch
  8. 2025Inference-Time Scaling of Diffusion Language Models via Trajectory Refinement