In this paper, we propose to adopt the diffusion approximation tools to study the dynamics of Ojas iteration which is an online stochastic gradient descent method for the principal component analysis. Ojas iteration maintains a running estimate of the true principal component from streaming data and enjoys less temporal and spatial complexities. We show that the Ojas iteration for the top eigenvector generates a continuous-state discrete-time Markov chain over the unit sphere. We characterize the Ojas iteration in three phases using diffusion approximation and weak convergence tools. Our three-phase analysis further provides a finite-sample error bound for the running estimate, which matches the minimax information lower bound for principal component analysis under the additional assumption of bounded samples.
Download