Pseudo-Non-Linear Data Augmentation: A Constrained Energy Minimization Viewpoint

ICLR 2026 Conference Submission16154 Authors

Keywords: data augmentation, information geometry, energy-based model

TL;DR: We propose a simple, information-geometric approach to data augmentation that is learning-free, efficient, controllable, and broadly applicable to structured data.

Abstract: We propose a simple yet novel data augmentation method for general data modalities based on energy-based modeling and principles from information geometry. Unlike most existing generative models, which rely on learning latent representations with black-box models, our proposed framework enables constructing a geometrically aware latent space that depends on the structure of the data itself, which further supports efficient and explicit encoding and decoding procedures. We then present and discuss how to design latent spaces that will subsequently control the augmentation with the proposed algorithm. Empirical results demonstrate that our data augmentation method achieves competitive downstream task performance compared to other baselines, while offering fine-grained controllability that is lacking in other baselines.

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

Submission Number: 16154