Home | Publications | AKS+25

Symmetry as Intervention; Causal Estimation With Data Augmentation

MCML Authors

Niki Kilbertus

Prof. Dr.

Principal Investigator

Ethics in Systems Design and Machine Learning

Abstract

To our knowledge, we provide the first analysis of causal estimation under hidden confounding using only observational data and knowledge of symmetries in data generation via data augmentation (DA) transformations. We show that such DA is equivalent to interventions on the treatment , mitigating bias from hidden confounding, and that framing DA as a relaxation of instrumental variables (IVs)-sources of randomization that are conditionally independent of the outcome -can further improve causal estimation beyond simple DA.

inproceedings AKS+25

NeurReps @NeurIPS 2025

Workshop on Symmetry and Geometry in Neural Representations at the 39th Conference on Neural Information Processing Systems. San Diego, CA, USA, Nov 30-Dec 07, 2025. To be published. Preprint available.