On the Anisotropy of Score-Based Generative Models

Andreas Floros, Seyed-Mohsen Moosavi-Dezfooli, Pier Luigi DragottiSubmission to ICLR 2026 · Rejected

Abstract

We investigate the role of network architecture in shaping the inductive biases of modern score-based generative models. To this end, we introduce the Score Anisotropy Directions (SADs), architecture-dependent directions that reveal how different networks preferentially capture data structure. Our analysis shows that SADs form adaptive bases aligned with the architecture's output geometry, providing a principled way to predict generalization ability in score models prior to training. Through both synthetic data and standard image benchmarks, we demonstrate that SADs reliably capture fine-grained model behavior and correlate with downstream performance, as measured by Wasserstein metrics. Our work offers a new lens for explaining and predicting directional biases of generative models. Code to reproduce our experiments is included in the supplementary material.
BibTeX
@inproceedings{floros2026anisotropy,
  author = {Andreas Floros and Seyed-Mohsen Moosavi-Dezfooli and Pier Luigi Dragotti},
  title = {{On the Anisotropy of Score-Based Generative Models}},
  booktitle = {International Conference on Machine Learning},
  year = {2026},
  url = {https://openreview.net/forum?id=pRgUJfrFw9},
}
Figure