A NeurIPS 2026 Workshop · Sydney, Australia
Beyond next-token prediction — exploring the theory, algorithms, and applications of discrete diffusion and flow models for parallel, non-causal generation.
01 Overview
Autoregressive models produce their output one element at a time, in a fixed order. The recipe is remarkably general, and it has carried the field a long way. However, it also ties the order of computation to the order of the result, leaving parallelism, revision, and control hard to reach. Discrete diffusion and flow models take a different route. They generate by iterative denoising: refining many positions at once, and revisiting earlier choices as a sample takes shape.
That shift brings a different set of computational properties — non-causal, parallel, and controllable — along with open questions spanning mathematics, algorithms, and engineering, and a growing body of work that combines the two paradigms rather than choosing between them. This workshop joins two conversations: the theory and algorithms that make discrete generative models work, and the applications and systems that put them to use.
Discrete diffusion and flow models at the intersection of generative modeling, optimal transport, and stochastic optimal control.
Novel applications and scalable systems that exploit non-causal, parallel generation.
02 Scope
This workshop considers, but is not limited to, the following topics.
03 Call for Papers
We invite submissions on any topic within the workshop's scope, across both tracks, from theory and algorithms to applications and systems. All submissions are handled through OpenReview.
Accepted papers are presented as posters, with a selection of contributed orals. A Best Paper Award recognizes outstanding work.
04 Important Dates
All deadlines are 23:59 AoE. Dates are indicative and subject to change.
05 Schedule
A full day of invited talks, contributed orals, two poster sessions, and a panel. Speaker slot assignments are being finalized.
Tentative program · times shown in the local conference time zone. Invited-talk slots will be mapped to speakers closer to the event.
06 Invited Speakers
Talk titles are tentative and may be refined before the workshop.
Talk TitleProduction-Grade Diffusion LLMs
Associate Professor of Computer Science at Stanford, where his group advances generative modeling, probabilistic inference, and diffusion methods. His recent work pushes diffusion language models toward production-scale text generation.
Talk TitleEfficient Diffusion Language Models from Autoregressive LLMs
Research Director at NVIDIA Research, focused on efficient deep learning — model compression, acceleration, and foundation models. His talk explores distilling fast diffusion language models from autoregressive LLMs.
Talk TitleDiffusion Models for the Natural Sciences
Senior Principal Research Manager at Microsoft Research, known for foundational work on normalizing flows and discrete diffusion. She applies generative models to problems across the natural sciences.
Talk TitleEliciting Non-Causal Reasoning in Diffusion Language Models
Assistant Professor at Harvard working on the theoretical foundations of machine learning, including the analysis of diffusion models and sampling algorithms, and how non-causal generation enables new forms of reasoning.
Talk TitleLLaDA: A New Paradigm for Large Language Modeling
Associate Professor at the Gaoling School of AI, Renmin University of China, specializing in deep generative models and diffusion. His group developed LLaDA, a large language diffusion model.
Talk TitleSynthesizing Autoregression and Diffusion for Flexible Generation and Planning
Researcher at Meta Superintelligence Labs working on probabilistic machine learning and generative models, unifying autoregression and diffusion for flexible, controllable generation and planning.
07 Panel
A moderated discussion on the open problems and near-term future of diffusion and flow models.
08 Organizers
General enquiries: bento-neurips@googlegroups.com
Associate Professor at KAIST working on 3D vision, computer graphics, and generative models.
mhsung@kaist.ac.kr
PhD student at KAIST, working on diffusion models and generative AI.
jh27kim@kaist.ac.kr
Professor at Georgia Tech working on Generative AI such as diffusion models.
mtao@gatech.edu
Assistant Professor at the University of Pennsylvania developing generative models for biological and therapeutic design.
pranam@engineering.upenn.edu
Undergraduate Researcher at the University of Pennsylvania researching generative models for biomolecular design.
sophtang@engineering.upenn.edu
Research Scientist at Cerebras Systems working on large-scale training, scaling laws, and efficient deep learning.
nolan@cerebras.net
Researcher at MBZUAI focused on discrete diffusion language models and efficient generative modeling.
subham.sahoo@mbzuai.ac.ae09 Sponsors & Support
Sponsorship helps us fund travel grants and broaden participation for students and early-career researchers. We gratefully acknowledge the organizations supporting this workshop.
Interested in supporting the workshop? bento-neurips@googlegroups.com
Join us at NeurIPS 2026 to push generation beyond next-token prediction. Submissions open July 20 and close August 29, 2026.
Submit on OpenReviewSubmissions open Jul 20, 2026 at NeurIPS.cc/2026/Workshop/BeNTo · bento-neurips@googlegroups.com