Call for Papers

“Your Model is Wrong: Robustness and misspecification in probabilistic modeling”

A virtual workshop at the Thirty-fifth Annual Conference on Neural Information Processing Systems (NeurIPS 2021).

Date: December 14, 2021 (TBC)

Location: Online!

Contact: robustbayesworkshop@gmail.com

Submission link: https://cmt3.research.microsoft.com/WrongModels2021/

• September 1, 2021: Submission link posted

• September 13, 2021: Abstract Submission Deadline (11:59pm AoE)

• September 17, 2021: Full Paper Submission Deadline (11:59pm AoE)

• October 23, 2021: Notification

• November 1, 2021: SlidesLive Upload Deadline (02:00 AM UTC) (for livestreamed talks only)

• November 22, 2021: Camera Ready deadline (via CMT) (11:59pm AoE)

• December 1, 2021: Poster Upload deadline (Neurips.cc)

• December 3, 2021: Optional 3-minute video deadline (11:59pm AoE)

• December 14, 2021: Workshop at NeurIPS

Probabilistic modeling is a foundation of modern data analysis -- due in part to the flexibility and interpretability of these methods -- and has been applied to numerous application domains, such as the biological sciences, social and political sciences, engineering, and health care. However, any probabilistic model relies on assumptions that are necessarily a simplification of complex real-life processes; thus, any such model is inevitably misspecified in practice. In addition, as data set sizes grow and probabilistic models become more complex, applying a probabilistic modeling analysis often relies on algorithmic approximations, such as approximate Bayesian inference, numerical approximations, or data summarization methods. Thus in many cases, approximations used for efficient computation lead to fitting a misspecified model by design (e.g., variational inference). Importantly, in some cases, this misspecification leads to useful model inferences, but in others it may lead to misleading and potentially harmful inferences that may then be used for important downstream tasks for, e.g., making scientific inferences or policy decisions.

The goal of the workshop is to bring together researchers focused on methods, applications, and theory to outline some of the core problems in specifying and applying probabilistic models in modern data contexts along with current state-of-the-art solutions. Participants will leave the workshop with (i) exposure to recent advances in the field, (ii) an idea of the current major challenges in the field, and (iii) an introduction to methods meeting these challenges.

We encourage submissions identifying problems and potential solutions regarding misspecification, approximation, robustness, and reliability of probabilistic inference on topics including but not limited to:

• Nonparametric Bayesian models

• Probabilistic surrogate modeling

• Approximate Bayesian inference algorithms (e.g., MCMC and variational inference)

• Probabilistic numerical methods

• Data summarization for probabilistic models (e.g., sketching, subsampling, coresets)

• Missing data and data imputation

• Application areas highlighting issues and harms arising from misspecification (e.g., health care, social science, economics, and scientific engineering applications).

Some concrete examples of problems, solutions, and applications include:

• Generalized Bayesian methods: e.g., power posteriors and Gibbs posteriors

• (In)consistency for the number of components in nonparametric and finite mixture models

• Misspecified Bayesian nonparametric models for real-world scaling behavior, e.g., microclustering, sparse graphs, and local exchangeability

• Bayesian robustness via bootstrap sampling

• Sensitivity/robustness to model choice and nonparametric Bayes

• Bayesian robustness to particular influential data points

• Variational inference under misspecification

• Uncertainty quantification of numerical approximations

• Model misspecification issues and model checking for gene expression data applications

• Model checking and fast, reliable evaluation more generally for Bayesian methods

• Missing data imputation in specific application domains, e.g., social & health data

• Robustness of probabilistic modeling software implementations

Submission Style and Format Guidelines

Submissions should be short research papers of up to 4 pages in PDF format using the workshop style files (use these modified NeurIPS LaTeX templates). Author names should be anonymized (use the default submission line in the style file). Submissions that violate the provided style or page limits may be rejected without further review.

References may extend as far as needed beyond the 4 page limit. The main 4-page paper should include all text and figures, and should adequately describe the work, its contributions, and its limitations. Submissions may include supplementary material as an appendix within the submitted PDF file, but reviewers are not required to read any supplementary material beyond the 4 page limit.

Peer Review and Acceptance Criteria

If authors' research has previously appeared in a journal, workshop, or conference (including the NeurIPS 2021 conference), their workshop submission should meaningfully extend that previous work.

All submissions will undergo peer review via the workshop’s program committee. Accepted papers will be chosen based on technical merit and suitability to the workshop's goals.

Accepted Papers and Presentations

There will be no published proceedings for this workshop (although accepted papers will be posted on the workshop website if authors wish); we hope that authors will find discussion and feedback at the workshop beneficial for developing the research they present, and we encourage authors to submit their resulting work for archival publication in other venues after the workshop.

All accepted papers will be included in a poster presentation session on the day of the workshop and (optionally) may provide a link to a 3-minute spotlight video. Some accepted papers of top quality will be invited to give short contributed talk presentations at the workshop.

Poster uploads and instructions are available at: https://neurips.cc/PosterUpload.

For the camera ready submission, we allow an additional page beyond the 4 page pdf limit to account for the addition of author names and acknowledgements; this page limit does not include references or appendices.