This paper introduces a novel method for asynchronous eigenspace computation in distributed systems, focusing on the Grassmannian manifold. The proposed Grassmannian incremental aggregation technique minimizes per-update costs by refreshing only arriving components and reusing cached gradients, thus avoiding global synchronization. The method utilizes an extrinsic polar update to maintain intrinsic subspace geometry and offers a two-phase linear convergence characterized by broad-basin and local regimes, demonstrating improved sample efficiency and wall-clock convergence in experiments on principal component analysis. AI
IMPACT This research could improve the efficiency of distributed machine learning algorithms that rely on eigenspace computations, such as principal component analysis.
RANK_REASON The cluster contains a research paper detailing a new mathematical method for eigenspace computation. [lever_c_demoted from research: ic=1 ai=0.7]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
