Schedule optimization for tau-leaping in masked discrete diffusion

arXiv:2609.21960 · math.ST, cs.LG, stat.ML, stat.TH · Submitted 2026-09-18 · Read on arXiv

math.ST, cs.LG, stat.ML, stat.TH

Submitted: 2026-09-18

Updated: 2026-10-05

License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/

Terminology

• Diffusion Model — 14× in this paper · explained in 8 episodes
• Diffusion Models — 14× in this paper · explained in 6 episodes
• Accuracy — 3× in this paper · explained in 1 episode
• Exchangeability — 2× in this paper · explained in 2 episodes
• Generative Modeling — 2× in this paper · explained in 3 episodes
• Adaptation — 1× in this paper · explained in 1 episode
• Convergence Guarantees — 1× in this paper · explained in 3 episodes
• Theoretical Guarantees — 1× in this paper · explained in 2 episodes

Sources

• A Sharp Fannes-type Inequality for the von Neumann Entropy
• Accelerated Sampling from Masked Diffusion Models via Entropy Bounded Unmasking
• Confidence-Based Decoding is Provably Efficient for Diffusion Language Models
• Optimal Inference Schedules for Masked Diffusion Models
• Non-Asymptotic Convergence of Discrete Diffusion Models: Masked and Random Walk dynamics
• Efficient Sampling with Discrete Diffusion Models: Sharp and Adaptive Guarantees
• Improved Sampling Schedules for Discrete Diffusion Models
• Train for the Worst, Plan for the Best: Understanding Token Ordering in Masked Diffusions
• Error Bounds and Optimal Schedules for Masked Diffusions with Factorized Approximations
• Breaking AR's Sampling Bottleneck: Provable Acceleration via Diffusion Language Models
• Discrete Diffusion Modeling by Estimating the Ratios of the Data Distribution
• Plan for Speed: Dilated Scheduling for Masked Diffusion Language Models
• Your Absorbing Discrete Diffusion Secretly Models the Conditional Distributions of Clean Data
• Score-Based Generative Modeling through Stochastic Differential Equations
• The data geometry of masking diffusion: Certified-optimal schedules via unmasking growth complexity
• The Cosine Schedule is Fisher-Rao-Optimal for Masked Discrete Diffusion Models
• Adaptation to Intrinsic Dependence in Diffusion Language Models
• Masked Diffusion Models are Secretly Time-Agnostic Masked Models and Exploit Inaccurate Categorical Sampling

Related papers

• Conformal Prediction for Dyadic Regression Under Complex Missingness
• Bentkus-type asymptotic e-values
• High-Dimensional Asymptotics of Differentially Private PCA
• KL Convergence Guarantees for Score diffusion models under minimal data assumptions
• Geometric bias in eigenspace perturbation under random heterogeneous noise
• On the Asymptotic Inadmissibility of Double Machine Learning Estimators Under Structure-Agnostic Models