Explore a groundbreaking algorithm for learning mixtures of linear regressions in subexponential time using Fourier moments in this 25-minute conference talk. Delve into solving multiple linear and Gaussian systems, examine mixture-of-experts and mixture models, and analyze previous results in the field. Investigate the potential barrier at exp(k) and discover the innovative approach that overcomes it. Learn about techniques for identifying individual components, understanding the minimum variance of Gaussian mixtures, and grasping the significance of exp(VR) in the algorithm's complexity. Gain insights into methods for learning all components simultaneously and consider open questions for future research in this cutting-edge area of machine learning and statistical analysis.
Learning Mixtures of Linear Regressions in Subexponential Time via Fourier Moments
Association for Computing Machinery (ACM) via YouTube
Overview
Syllabus
SOLVING MANY LINEAR SYSTEMS
SOLVING MANY GAUSSIAN LINEAR SYSTEMS
MIXTURE-OF-EXPERTS
MIXTURE MODELS
PREVIOUS RESULTS
A BARRIER AT exp(k)?
OUR RESULTS
LEARNING ONE COMPONENT
MIN VARIANCE OF A GAUSSIAN MIXTURE
WHERE DOES exp(VR) COME FROM?
LEARNING ALL COMPONENTS
OPEN QUESTIONS
Taught by
Association for Computing Machinery (ACM)
