Interacting Particle Systems for Expectation Maximization in Latent Variable Models
Offered By: Alan Turing Institute via YouTube
Course Description
Overview
Explore a novel interacting particle system for implementing expectation maximization (EM) in latent variable models. Delve into the continuous-time system's properties, its relation to Langevin diffusion, and how it enables non-asymptotic concentration bounds for optimization error. Compare this approach to existing methods, examine the proof structure, and discuss potential generalizations. Cover topics such as simulated annealing, Langevin dynamics, target measures, and convergence results. Gain insights from Tim Johnston and Francesca Crucinio's research on optimizing parameters in latent variable models using this innovative particle system approach.
Syllabus
Latent Variable Models (LVM)
EM and Variants
Simulated Annealing for LVM
An Interacting Particle System for LVM (Kuntz et al., 2023)
An Optimisation Point of View
Langevin Dynamics
Target Measure
Concentration
Algorithm
Assumptions
Main Convergence Result
Conclusions
Taught by
Alan Turing Institute
Related Courses
Introduction to Artificial IntelligenceStanford University via Udacity Natural Language Processing
Columbia University via Coursera Probabilistic Graphical Models 1: Representation
Stanford University via Coursera Computer Vision: The Fundamentals
University of California, Berkeley via Coursera Learning from Data (Introductory Machine Learning course)
California Institute of Technology via Independent