Projected Power Method: An Efficient Algorithm for Joint Discrete Assignment - Lecture 3

Explore the third lecture in the TRIAD Distinguished Lecture Series featuring Yuxin Chen from Princeton University. Delve into the topic of "Projected Power Method: An Efficient Algorithm for Joint Discrete Assignment" in this 46-minute presentation. Learn about a proposed low-complexity, model-free procedure that operates in a lifted space by representing distinct label values in orthogonal directions and optimizes quadratic functions over hypercubes. Discover how the algorithm refines iterates through projected power iterations, starting with an initial guess computed via a spectral method. Understand the proof that demonstrates the algorithm's convergence to the maximum likelihood estimate without errors for a broad class of statistical models. Gain insights into the practical applications of this algorithmic framework for various discrete assignment problems through numerical experiments conducted on both synthetic and real data.