Reliable ML from Unreliable Data

Workshop @ NeurIPS 2025

San Diego Convention Center | Dec 6, 2025

Distributions shift, chatbots get jail‑broken, users game algorithms — how do we build reliable machine learning when data are missing, corrupted, or strategically manipulated?

This workshop bridges theory and practice to tackle these challenges, bringing together researchers working on distribution shift, adversarial robustness, and strategic behaviour to chart principled yet deployable solutions for Reliable ML from Unreliable Data.

Call for Papers

We invite work that advances theory, empirical understanding, or systems design for robust and reliable machine learning under imperfect data – including, but not limited to, the following topics:

Distribution shift and transfer learning
Adversarial robustness and defenses
Strategic behavior in socio-technical systems
Learning with missing or biased data; and truncated statistics
Causal inference beyond overlap, with confounders, or with errors
LLM safety and alignment
Robustness in interactive environments

Submissions may report new results, negative findings, benchmarks, or visionary perspectives.

Long track: up to 9 pages (excluding references).
Short track: up to 4 pages (excluding references).
Formatting: use the official NeurIPS 2025 style files and submit a single PDF (which should be anonymized, like NeurIPS submissions; see these instructions on anonymization).
Appendix: include any supplementary material in the same PDF — no page limit for the appendix.

Important Dates

Submission deadline: Fri Aug 29 2025 (AoE) ~~Fri Aug 22 2025 (AoE) — previous deadline~~
Notification of decisions: Mon Sep 22 2025 (AoE)
Workshop: Sat Dec 6, 2025 (at NeurIPS 2025)
Submission link: OpenReview

The workshop is non‑archival; authors are free to publish revised versions elsewhere. Every submission will receive at least two reviews from our program committee, and accepted papers will be presented as talks or posters.