Tamara Kolda, Distinguished Member of Technical Staff, Sandia National Laboratories @ MLconf SF Part
summary
A focused survey of canonical polyadic (CP/PARAFAC) decompositions shows how 3D data such as time-evolving networks is written as sums of rank-one components, exposes how the factors relate to matrix SVD, compares optimization objectives under different statistical assumptions, and demonstrates algorithms that scale to large sparse data and practical implementations where a small number of components (often 3–10) recover meaningful structure and anomalies.
AI-written summary · about
related talks
10mDale Smith, Data Scientist, Nexidia
58mTensor Decomposition, Sparse Representations and Applications
34mDimensionally reducing data squeezing out the good stuff
01h 03mRecent Progress in High-Dimensional Learning
01h 00mSmoothed Analysis of Tensor Decompositions
01h 11mTensor Decompositions: Uniqueness and Smoothed Analysis
52mAverage-Case Overcomplete Tensor Decomposition
37mA Bluffer's Guide to Dimension Reduction