The effects of top-down connections in computational multilayer two-way hebbian networks: Abstract representation and temporal context
South Hall A
Sunday, Oct 18, 2009, 4:00 PM - 5:00 PM
*M. D. LUCIW
, J. WENG;
Michigan State Univ., East Lansing, MI
A great mystery is how more abstract representation is developed in later cortical areas. It is also unclear how motor actions might alter lower-level cortical representation. The computational studies reported here indicate that top-down connections from a later layer (e.g., a motor area) to an early layer enable the early layer to exhibit two properties of abstraction: (a) responses from developed neurons are insensitive to irrelevant sensory information (invariant information) but are sensitive to relevant sensory information for classification (discriminant information), (b) neurons form topographic cortical areas according to abstract classes. Both properties transform meaningless (iconic) raw sensory inputs into internal representation with abstract meanings. We also show how top-down connections provide temporal context and how such context assists perception in a continuously changing physical world, with which an agent interacts during its developmental learning. In experiments in object learning and recognition, using small toy objects, we show the developmental effect of top-down connections: responses of neurons to visual stimuli of rotating object sequences will be clustered in the same area on the neuronal plane for each single class, which is the first computational account, as far as we know, of this widely recognized cortical phenomenon. And the recognition performance is also greatly increased when compared with a network where top-down connections were not enabled. It is also shown how the temporal context greatly improves performance, to nearly 100% after the transition periods from one object to the next.
Partially supported by General Motors Research and Development
[Authors]. [Abstract Title]. Program No. XXX.XX. 2009 Neuroscience Meeting Planner. Chicago, IL: Society for Neuroscience, 2009. Online.
2009 Copyright by the Society for Neuroscience all rights reserved. Permission to republish any abstract or part of any abstract in any form must be obtained in writing by SfN office prior to publication.
When adding items to your Itinerary, please click "Add Checked Selections to My Itinerary" on
page of your search results.
About the Meeting
Fellowships, Awards, and Prizes
Frequently Asked Questions
Download Abstracts and
Daily Books to Your Kindle
OASIS Technical Support.
Monday - Friday, 9 am - 5 pm CT
The Online Abstract Submission and Invitation System
© 1996 - 2015 Coe-Truman Technologies, Inc. All rights reserved.