MIT Department of Electrical Engineering & Computer Science
Graphical Models, Neural Networks and Variational Methods
Michael I. Jordan
Department of Brain and Cognitive Sciences, MIT
Friday, May 16, 1997
3:00 PM (2:45 refreshments)
NE43-8th floor, AI Playroom
EECS Special Seminar
Abstract
Graphical models provide a useful formalism for understanding
network-based statistical and machine learning systems.
The key computation in all of these systems is that of computing
a conditional probability distribution over hidden (latent)
variables, given visible (observed) variables. For certain
classes of architecture (including chain and tree-like structures),
this calculation can be carried out exactly and efficiently.
For other architectures, the exact calculation is inefficient
and approximations must be developed. In this talk I describe
the use of variational methods for the efficient calculation of
bounds of probabilities on graphs, exemplifying these methods
with applications to Markovian probabilistic decision trees,
factorial Hidden Markov models, Boltzmann machines and Bayesian
belief networks with logistic or noisy-OR nodes.
Hosts: Professor T. Lozano-Pérez and Professor E. Grimson
URL of this page:
http://www-eecs.mit.edu/AY96-97/events/55.html
Created: May 14, 1997
|
Modified: Jun 24, 1997
This announcement is from the MIT EECS 1996-97 archive.
|
Current events
To MIT EECS home page
|
Your comments
and inquiries are welcome.
