Workshop Papers
4-to-8-page papers following the ICML 2026 main conference template, with unlimited references and appendix; maximum file size of 50 MB. Papers may include supplementary material, though reviewers are not obligated to review it.
Machine Learning/CORE Unranked
SPIGM @ ICML 2026
ICML 2026 Workshop on Structured Probabilistic Inference & Generative Modeling
The SPIGM @ ICML 2026 workshop focuses on structured probabilistic inference and generative modeling, bringing together researchers from academia and industry to address challenges in encoding domain knowledge into probabilistic methods for structured data. It emphasizes applications in science, foundation models, and structured modalities like graphs, time series, and 3D data, while fostering collaboration on theory, methodology, and practical implementations.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Research areas in scope
01
Generative Methods and Inference
Generative methods for graphs, 3D, time series, text, video, and other structured modalitiesProbabilistic inference in these models for reward fine-tuning, alignment, acceleration, watermarking, etc.Unsupervised representation learning of high dimensional structured dataIntersection between probabilistic inference and LLMs, VLMs, VLAs, and foundation modelsScaling and accelerating inference and generative models on structured dataApplications and practical implementations of existing methods to areas in scienceEmpirical analysis comparing different architectures for a given data modality and applicationSimilar to last year, we also encourage submissions that explore the relevance of probabilistic inference in the era of foundation modelsWe welcome submissions that explore the intersection of probabilistic inference and foundation models
02
Inference Techniques
Sampling and variational inferenceUncertainty quantification in AI systemsApplications in sampling, optimization, decision making, etc.
Policies worth checking twice
- The review process is double-blind; all submissions must be anonymized and identification information must not be leaked.
- Submissions of papers already published in journals or presented at non-machine-learning conferences or workshops are welcome.
- Papers previously presented at machine-learning venues (e.g., ICLR, NeurIPS, ICML, CVPR, AISTATS) will be considered only if they include substantial extensions or new results.
- An exception is made for papers that were published but not presented due to unforeseen circumstances (e.g., visa issues).
- Reviewers are not obliged to review supplementary material.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026