Big Data Is Low Rank - Madeleine Udell, Cornell University

Explore the intersection of statistics and computer science in machine learning through this 46-minute lecture by Madeleine Udell from Cornell University. Delve into the concept of low-rank structures in big data, covering topics such as principal components analysis, generalized low-rank models, and their applications in various fields. Learn about fitting GLRMs using alternating minimization, imputing missing and heterogeneous data, and validating models. Discover how low-rank models can be applied to dimensionality reduction, autoML, and experiment design for timely model selection. Examine real-world examples, including hospitalizations and political data tables, to understand the practical implications of low-rank structures. Gain insights into latent variable models and their approximate low-rank nature, bridging the gap between statistical theory and computational efficiency in the era of Big Data.

Intro
Data table: politics
Principal components analysis
Generalized low rank model
Fitting GLRMs with alternating minimization
Impute missing data
Impute heterogeneous data
Example: Julia implementation
Validate model
Hospitalizations are low rank
Low rank models for dimensionality reduction
Low rank autoML
Low rank fit correctly identifies best algorithm type
Experiment design for timely model selection
Latent variable models examples
Approximate low rank
Rank of nice latent variable models?