Proofs to Algorithms in High-Dimensional Statistics (an introduction, with applications to Gaussian Mixture Models)
12th March 2021, 3:00 pm – 4:00 pm
Abstract: I will give an introduction to a powerful technique for algorithm design in high-dimensional statistics, the “Sum of Squares method”. In recent years, this technique has led to novel polynomial-time algorithms with strong provable guarantees for a wide range of computationally-challenging problems: high-dimensional robust estimation, clustering, learning of latent-variable models, matrix and tensor completion, and more. I will discuss the scope of the method, the basic tools, and then describe one of the early (and simpler) applications, to learning high-dimensional Gaussian mixture models.