Power Method and Applications

This lecture covers the power method and its applications in topic models and document analysis. Starting with tensors, the instructor explains the iterative process to find the dominant eigenvector and eigenvalue. The lecture then delves into the power method for matrices, emphasizing the goal of recovering the highest eigenvalue and eigenvector. The instructor demonstrates the formalization of the model using tensors and dictionaries, highlighting the reconstruction of probability distributions. The lecture concludes with the application of the power method to empirical data, focusing on whitening procedures and the explicit verification of results.