Conference Website (requires ICML 2022 login)

List of Accepted Papers

Folder with PDFs of Accepted Papers

Overview

The importance of robust, trustworthy predictions continues to grow as machine learning models are increasingly relied upon for decision-making in high-stakes settings. Ensuring reliability in real-world applications remains an enormous challenge, particularly because data in the wild frequently differs substantially from the data on which models were trained.

With the growing interest in addressing this problem has come a growing awareness of the multitude of possible meanings of “distribution shift” and the importance of understanding the distinctions between them: which types of shift occur in the real world, and under which of these is generalization feasible?

This workshop aims to provide a forum for discussing promising directions for making rigorous progress on existing problems and identifying meaningful, well-defined new problems to pursue. The following is an (incomplete) list of the kinds of questions we hope to address:

1. What are the most useful models/formalisms of distribution shift? What aspects of real-world data do they capture, and what other desiderata do they or should they meet? When can we hope to design statistically and computationally efficient methods for robust generalization?

2. What are the implicit assumptions behind existing methods which perform well empirically? Under what conditions do they work, and how can they be generalized to more diverse settings? When do they fail, and how can these failures be prevented/mitigated?

3. What properties make for a good benchmark? To what extent is there a gap between the challenges of real-life systems deployed in specific application domains and current benchmarks, and what can we do to close it?

We encourage both theory and application-driven submissions, with an emphasis on rigor in both analysis and evaluation. We also encourage “evaluation” and “opinion” submissions, which focus not on novelty or state-of-the-art accuracy but instead on collating, summarizing, assessing, and/or critiquing existing approaches and identifying promising directions for future research.

Please contact distribution-shift-principles-workshop@googlegroups.com for inquiries.

Invited Speakers

Panelists

Organizers