UCSD Cogsci Tech Report 97.01

[Javier R. Movellan]

UCSD.Cogsci.TR.97.01

AUTHORS: Javier R. Movellan and Paul Mineiro

TITLE: Modularity and Catastrophic Fusion: A Bayesian Approach with
Applications to Audiovisual Speech Recognition.



ABSTRACT:

  While modular architectures have desirable properties, integrating
  the outputs of many modules into a unified representation is not a trivial
  issue.  In this paper we examine catastrophic fusion, a problem that occurs
  when modules are fused in incorrect context conditions.  This problem has
  become especially apparent in the current research on automatic recognition
  of multimodal signals and has practical as well as
  theoretical relevance.  Catastrophic fusion arises because modules make
  implicit assumptions and thus operate correctly only within a certain
  context.  Practice shows that when modules are tested in contexts
  inconsistent with their assumptions, their influence on the fused product
  tends to increase, with catastrophic results.  We propose a principled
  solution to this problem based upon Bayesian ideas of competitive models.
  We study the approach analytically on a classic Gaussian discrimination task
  and then apply it to a realistic problem on audiovisual speech recognition
  (AVSR) with excellent results. For concreteness our emphasis is on
  applications to AVSR but the problems at hand are very general and touch
  fundamental issues about cognitive architectures.



ELECTRONIC COPIES: http://cogsci.ucsd.edu 
                    and follow the link to  "Tech Reports" 

PHYSICAL COPIES:  Available for $7.00 within the US, $10.00 outside the US.

For physical copies send a check of money order payable to UC Regents
and mail it to the following address,

TR Request
Javier R. Movellan
Department of Cognitive Science
University of California San Diego
La Jolla, Ca 92093-0515