Research map / Learning theory and algorithms / Kernels and regression theory
Riemannian Ordinary Least Squares
Xiaoyu Chen, Yujing Huang, Yingyan Zeng
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's full text. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes Riemannian Ordinary Least Squares, a regression framework for manifold-valued responses that accepts both scalar and manifold-valued predictors and returns per-coefficient hypothesis tests and effect sizes.
Problem
Regression with Riemannian-manifold-valued responses and mixed scalar/manifold-valued predictors lacks a framework that supports point prediction, formal hypothesis testing, and effect-size estimation for individual coefficients.
Approach
A manifold-valued OLS-style regression that handles both predictor types in one framework and returns per-coefficient hypothesis tests and effect sizes, which existing manifold regression methods do not jointly provide.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use ROLS instead of NL-Frechet regression | manifold regression · DTI-CC | Frechet RMSE: 1.134 -> .400 | Improved |
| use ROLS instead of geodesic regression | manifold regression · Protein backbone (T2) | Frechet RMSE: 1.215 -> 1.177 | Improved |
| use ROLS instead of NL-Frechet regression | manifold regression · Wind (S2) | Frechet RMSE: 0.414 -> .095 | Improved |
| add scalar predictors to manifold-only ROLS (ablation) | ROLS (m) · Protein backbone (T2) | Frechet RMSE: 1.395 -> 1.177 | Improved |
| add scalar predictors to manifold-only ROLS (ablation) | ROLS (m) · DTI-CC | Frechet RMSE: .400 -> .400 | No clear effect |
| add scalar predictors to manifold-only ROLS (ablation) | ROLS (m) · Wind (S2) | Frechet RMSE: .098 -> .095 | No clear effect |
| let data spread approach the cut locus (ablation) | ROLS on S2 · S2 simulation | test RMSE: 0.11 -> 1.10 | Hurt |
| use ROLS under tangent misspecification | ROLS vs NL-Frechet · S2 quadratic distortion gamma=0.6 | Frechet RMSE: 0.438 -> 0.300 | Improved |
| use sandwich covariance instead of model-based (ablation) | coefficient test · misspecified S2 gamma=0.6 | type I rate: 0.36 -> 0.084 | Improved |
| use ROLS in simulation | manifold regression · 45 simulated configs | RMSE: 44 of 45 lowest | Improved |
| use tangent linear-ridge full d x d map instead of scalar alpha (ablation) | ROLS · simulation | RMSE: median 1.2% higher | No clear effect |
Closest papers by meaning
- Robust Geodesic Regression (IJCV 2022 · similarity 0.90)
- RMLR: Extending Multinomial Logistic Regression into General Geometries (NeurIPS 2024 · similarity 0.88)
- Intrinsic Riemannian Cross-covariance for Manifold-valued Random Objects (NeurIPS 2026 · similarity 0.86)
- Non-parametric Online Change Point Detection on Riemannian Manifolds (ICML 2024 · similarity 0.86)
- Fréchet Regression on the Bures-Wasserstein Manifold (NeurIPS 2026 · similarity 0.85)
- Follow the Energy, Find the Path: Riemannian Metrics from Energy-Based Models (NeurIPS 2025 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
