Семинар HDI & TFAIM Lab: Streaming PCA: Independent and Markovian Data

This talk will focus on the theoretical analysis of Streaming Principal Component Analysis (Streaming PCA), an online method for estimating the top eigenvectors of a covariance matrix without storing the full dataset. We will first review classical convergence guarantees for batch PCA and then discuss finite-sample convergence bounds for Oja’s algorithm in the streaming setting. The analysis will cover both the standard case of independent observations and the more general setting where data points are sampled from an irreducible, aperiodic, and reversible Markov chain

Доклад основан на работах: 

Streaming PCA for Markovian Data (2023)
Streaming PCA: Matching Matrix Bernstein and Near-Optimal Finite Sample Guarantees for Oja’s Algorithm (2016)