Reproduce “Improved Analysis of the Accelerated Noisy Power Method with Applications to Decentralized PCA” 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
Pierre Aguié, Mathieu Even, Laurent Massoulié
The authors analyze accelerated noisy power iteration with inexact matrix-vector products and apply it to decentralized PCA. Momentum gives, up to logarithmic factors, a square-root dependence on the leading-eigenvalue-to-eigengap ratio, versus the standard method’s ratio. Under suitable initialization and bounds on both components of adversarial error, ADePM reduces subspace error faster in two experiments.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.