Smoothed Analysis for Tensor Decompositions and Unsupervised Learning

Explore a 33-minute conference talk on smoothed analysis for tensor decompositions and unsupervised learning. Delve into a powerful paradigm for proving robust, polynomial time guarantees of algorithms on worst-case computationally intractable problems. Discover a general framework for showing polynomial time smoothed analysis guarantees for tensor methods and related unsupervised learning problems. Learn about new high confidence lower bounds on the least singular value for random matrix ensembles with highly dependent entries. Understand how these concepts apply to robust decompositions of symmetric order-(2t) overcomplete tensors and learning overcomplete latent variable models like HMMs. Gain insights from Aravindan Vijayaraghavan's presentation at the Institute for Pure & Applied Mathematics' workshop on Tensor Methods and Emerging Applications to the Physical and Data Sciences.

Aravindan Vijayaraghavan: "Smoothed Analysis for Tensor Decompositions and Unsupervised Learning"