Research map / Optimization / Nonconvex and smooth optimization
Averaging on the Bures-Wasserstein manifold: dimension-free convergence of gradient descent
Jason Altschuler, Sinho Chewi, Patrik Robert Gerber, Austin J Stromme
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use Riemannian GD instead of Euclidean GD for the Bures-Wasserstein barycenter | Riemannian GD · Barycenter of n=50 covariance matrices, d=50, kappa=1000 | convergence accuracy | Improved |
| use Riemannian GD instead of an SDP solver (SCS) | Riemannian GD · Barycenter of Gaussians | runtime: ~15 seconds | Improved |
| use Riemannian GD instead of Riemannian Frank-Wolfe | Riemannian GD · Barycenter, Appendix F | convergence | Improved |
| use Riemannian SGD instead of Euclidean SGD | Riemannian SGD · 300x300 covariance matrices with identity barycenter | convergence | Improved |
| increase the dimension d | Riemannian GD and SGD · Barycenter, passes until error 10^-r with r in {3, 5} | passes to convergence | No clear effect |
| increase the regularization strength gamma | regularized barycenter with KL to N(0, I) · 20x20 covariances, n=100, perturbation rho in [0, 10] | closeness of barycenter to isotropy | Improved |
| decrease the entropic regularization epsilon | geometric median Algorithm 4 · n=d=30 covariances | approximation quality vs convergence speed | Mixed |
| perturb a fraction of input matrices (20%, 45%, 55%) | Wasserstein geometric median vs barycenter · n=d=20 covariances | shift of the estimate | Improved |
Closest papers by meaning
- Optimal Transport Barycenter via Nonconvex-Concave Minimax Optimization (ICML 2025 · similarity 0.91)
- On the Convergence of Projected Bures-Wasserstein Gradient Descent under Euclidean Strong Convexity (ICML 2024 · similarity 0.88)
- Continuous-time Riemannian SGD and SVRG Flows on Wasserstein Probabilistic Space (NeurIPS 2025 · similarity 0.88)
- Acceleration via silver step-size on Riemannian manifolds with applications to Wasserstein space (NeurIPS 2025 · similarity 0.88)
- A Unified Approach for Computing Wasserstein Barycenters of Discrete and Continuous Measures (NeurIPS 2026 · similarity 0.87)
- Computing Optimal Transport Maps and Wasserstein Barycenters Using Conditional Normalizing Flows (ICML 2025 · similarity 0.87)
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.
