We invite submissions that take principled approaches to advancing the understanding of generative modeling. This understanding may be pursued from diverse perspectives, including mathematical theory, physical modeling, and rigorous empirical analysis. In particular, we encourage contributions that address the following foundational questions:
Model classes and expressivity: What classes of functions, distributions, or algorithms can modern generative models represent, and how do architectural choices shape this expressive power? What makes transformers, diffusion models, discrete diffusions, and state-space models particularly effective at capturing linguistic or visual structures? How do repeated token-level computations expand the computational capabilities of these models, potentially enabling general reasoning?
Learning, generalization and inductive bias: How do generative models acquire high-level structure and capabilities such as in-context learning, compositional generalization, and reasoning from data? What aspects of structure and skill acquisition are governed by unifying principles across architectures and modalities, and which are shaped by the specific inductive biases of data, objective, architecture, and optimization? Are there fundamental limits to current self-supervised learning paradigms, such as next-token prediction?
Inference-time computation and adaptation: Modern generative systems increasingly rely on computation performed at inference time, from simple sampling to test-time adaptation. When and why does additional inference-time computation improve generation quality and reasoning? What are the statistical and computational limits of adapting generative models to distribution shifts, and how do these limits depend on model class, data structure, and the nature of the shift?
Post-training: Which properties of pretrained models enable successful post-training? Does post-training create new capabilities, reveal latent ones, or reweight behaviors already present in the pretrained model? What are the fundamental distinctions between supervised fine-tuning, reinforcement learning, and distillation, and when does each offer genuine advantages?
Submission instructions
Submissions are limited to four single-column pages, plus unlimited pages for references and appendices.
The reviewing process will be double-blind and all submissions must be anonymized. Please do not include author names, affiliations, acknowledgements, or any other identifying information in your submission. Submissions and reviews of rejected papers will not be made public.
All submissions must be made through OpenReview at this link. We ask you to use the standard LaTeX NeurIPS style files. It is not required to fill the NeurIPS checklist. Appendices can be submitted in a the same PDF file as the main text.
Note: If you are creating a new OpenReview profile, we strongly recommend using your institutional email address. Profiles created without an institutional email may require a moderation process, which can take up to two weeks.
Timeline
Paper submission deadline: September 5th, 2026, AoE
Review period: September 6th - 22nd
Notification date: September 29th, 2026, AoE
Dual submissions: This workshop is non-archival and will not have official proceedings. Workshop submissions can be submitted to other venues. We welcome ongoing and unpublished work, including papers that are under review at the time of submission. We do not accept submissions that have been accepted for publication in other venues with archival proceedings.
Attending the workshop: Our workshop is an in-person event, and authors are asked to present a poster at the workshop. A subset of papers will be selected for presentation in 15-minute contributed talks.
Contact: prigm-eurips-2026@googlegroups.com