Transcript Slide 1

Neural Network Modeling
Jean Carlson, Ted Brookings
SME Review - September 20, 2006
Specific Project Objectives
•
Determine the role of optimization,
robustness, and trade-offs in the structure
and behavior of neural networks.
•
Investigate the role of “grid cells” in rat
navigation
•
•
•
•
Quantify potential information encoded by grid cells.
Potential navigation strategies employing grid cells.
The flow of information from dMEC cortical region to the
hippocampus.
Cause and governing of “place cell” formation in hippocampus.
Technical Advances
We showed that the dMEC region of the rat cortex
encodes position information in a manner
analogous to a Residue Number System (RNS).
dMEC neurons fire in a regular
grid pattern
Hafting et. al. Nature 436 (2005)
This RNS encoding allows representation of
position over vast scales with a small,
biologically realistic number of neurons.
Technical Advances
We developed a
neural net
simulation, and
modeled the
network
connecting grid
cells to the
hippocampus of
the rat
Technical Advances
Hippocampus cells learn to integrate the
various periodic signals from grid cells to
identify the rats location.
Hippocampus cell fires only
when rat is in specific location
Multiple Inputs
Cortex cell has
broad firing pattern
Highest Impact and Significance to the Army
•
Gain understanding of the strategies
employed by rats to navigate in novel,
complicated environments
•
Describe the roles of learning and network
topology in processing of sense data in
artificial neural networks.
Plan for Moving Forward
Next Steps
1.
2.
3.
Determine result of competition for influencing hippocampus between
grid cells and more localized sensory input (e.g. vision, smell).
Investigate the role of network topology and synaptic learning rules in
forming memorized sequences (i.e. navigational routes).
Model the effect of changes in the environment (e.g size, features) on
hippocampal cell activity, and connect to existing experiment.
Timeline
1.
2.
3.
6 months
6 months
12 months