Pooling Image Datasets with Multiple Covariate Shift and Imbalance

Sotirios Panagiotis Chytas, Vishnu Suresh Lokhande, Vikas Singh

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Keywords: image harmonization, medical imaging

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2024/AuthorGuide.

TL;DR: Use Category Theory to provide a general harmonization tool that can handle multi-equivariance and multi-invariance with respect to the images' covariates.

Abstract:

Small sample sizes are common in many disciplines,
which necessitates pooling roughly similar datasets across
multiple sites/institutions to study weak but relevant
associations between images and disease incidence. Such
data often manifest shifts and imbalances in covariates
(secondary non-imaging data).
These issues are well-studied for classical models, but
the ideas simply do not apply to overparameterized DNN models.
Consequently, recent work has shown how strategies from
fairness and invariant representation learning provides
a meaningful starting point, but the current repertoire
of methods remains limited to accounting for shifts/imbalances in just a couple of covariates at a time. In this paper, we show how
viewing this problem from the perspective of Category theory
provides a simple and effective solution that completely avoids
elaborate multi-stage training pipelines that would otherwise be
needed. We show the effectiveness of this approach via
extensive experiments on real datasets. Further, we
discuss how our style of formulation offers a unified
perspective on at least 5+ distinct
problem settings in vision, from self-supervised learning
to matching problems in 3D reconstruction.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors' identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Primary Area: unsupervised, self-supervised, semi-supervised, and supervised representation learning

Submission Number: 1512