Reproduce “A Lie Group Approach to Riemannian Batch Normalization” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Ziheng Chen, Yue Song, Yunmei Liu, Nicu Sebe
University of Trento and University of Louisville researchers developed LieBN, a unified framework for Riemannian Batch Normalization that precisely controls Riemannian mean and variance on Lie groups. This method improves geometric deep learning model performance and efficiency across radar, action, and EEG classification tasks, demonstrating its generality and flexibility with deformed Lie groups.
Key insights
- Normalization operations for mean and variance control can be rigorously translated from Euclidean space to general Lie groups.
- The proposed LieBN framework naturally reduces to standard Euclidean Batch Normalization in Euclidean space, affirming its foundational correctness.
- The optimal Lie group structure and metric for LieBN vary across datasets, highlighting the benefit of flexible, parameterized "deformed Lie groups" for adapting to data geometry.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.