Researchers propose post-hoc group averaging to enforce symmetries in pre-trained ML models
A new arXiv paper describes post-hoc group averaging, also known as test-time augmentation, as a simple method to impose symmetry on a trained predictor without changing its architecture or retraining.
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-11-09 05:00 · arXiv stat.ML
Enforcing Symmetries on Pre-trained ML Models for Physics