Riemannian Diffusion Adaptation for Distributed Optimization on Manifolds

Authors: Xiuheng Wang, Ricardo Augusto Borsoi, Cédric Richard, Ali H. Sayed

ICML 2025 | Venue PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply the algorithm to the online decentralized principal component analysis problem and Gaussian mixture model inference. Experimental results with both synthetic and real data illustrate its performance.
Researcher Affiliation Academia 1Universit e de Lorraine, CNRS, CRAN, France 2Universit e Cˆote d Azur, CNRS, OCA, France 3 Ecole Polytechnique F ed erale de Lausanne, Switzerland.
Pseudocode Yes Algorithm 1 Riemannian Diffusion Adaptation
Open Source Code Yes Opensource code to reproduce the results is publicly available on https://github.com/xiuheng-wang/diffusion_manifold_release.
Open Datasets Yes Real data: We also obtain numerical results on the MNIST dataset (Le Cun, 1998).
Dataset Splits Yes We randomly shuffle the images, partition them into K = 20 subsets, and then run the algorithms to compute the first p = 5 principal components with the fixed step sizes µ = 0.002 and α = 0.005.
Hardware Specification Yes These experiments were performed on a computer with an Apple M4 Pro processor and 24GB of RAM.
Software Dependencies No Our method is implemented in Python with the Pymanopt toolbox (Townsend et al., 2016).
Experiment Setup Yes The simulations used fixed step sizes µ = 0.05 and α = 0.8.