Explore the power of active sampling in unsupervised learning through this 52-minute lecture by Aarti Singh from Carnegie Mellon University, presented at the Simons Institute. Delve into topics such as big data systems, partially observed data, statistical learning tradeoffs, and Gaussian graphical models. Examine passive and active graphical model selection, incoherence and leverage scores, and random sampling for matrix completion. Investigate the computational complexity and effects of row coherence through simulations and theory. Learn about column subset selection, active matrix approximation, and active sampling for clustering. Gain insights into the latest research and applications in interactive learning and unsupervised machine learning techniques.
Power of Active Sampling for Unsupervised Learning
Overview
Syllabus
Intro
Big Data, Bigger Systems
Partially observed data
Statistical Learning Tradeoffs
Power of Active Sampling
Gaussian Graphical Models
Passive Graphical Model Selection
Active Graphical Model Selection
Outline
Incoherence & leverage scores
Random sampling for matrix completion
Active sampling for matrix completion
Low-rank Matrix completion
Computational Complexity - Simulations
Effect of Row Coherence - Simulations
Effect of Row Coherence - Theory
Column subset selection (CSS)
Active Matrix Approximation
Summary of results and assumptions
Active sampling for clustering
Acknowledgements
References
Taught by
Simons Institute
