Two-stage computational model of the entorhinal-hippocampal loop
A. Lõrincz, L. Csató, Z. Gábor, and Gy. Buzsáki
Collection of Abstracts, Annual Meeting of the Neural Society, Los Angeles
1998
Abstract
The model is based on the following postulates.
(1) The entorhinal-hippocampal loop optimizes information transfer and
trains long term memories to learn the optimal representation
(2) These operations are performed in two, temporally discontiguous steps
so that the networks of the EC and the hippocampus can carry out both linear
an non-linear functions, albeit at different times.
(3) In the first phase the hippocampal network operates as a non-linear
error-compensating control architecture. The entorhinal cortex (EC) compares
the difference between neocortical representations (primary input) and feedback
information conveyed by the hippocampus (reconstructed input). The resulting
difference ("error") initiates plastic changes in the hippocampal networks in
an unsupervised fashion.
(4) Alteration of synaptic connections in the hippocampus gives rise to a new
hippocampal output that represent independent components under linear operation.
(5) In the second the hippocampus operates in linear mode and generates separated
(independent) outputs which, in turn, train the long-term memory traces in the EC
in a supervised fashion.
Each hippocampal field and EC layer perform distinct roles (dentate gyrus:
deconvolution; CA3: whitening; CA1: separation). Each network as well as their
interactions contribute uniquely to the overall performance of the EC/hippocampal
architecture (EC to CA1 connections as well as CA3 recurrent collateral system
serve predictions). The result of these operations is the formation of long-term
memories with minimized mutual information (independent components) in the
EC/neocortex. Because only the reconstruction error enters the hippocampus,
relatively limited computational networks (hippocampal fields) are required
for training long-term memories in the EC/neocortex.
Supported by NINDS, NIMH, US-Hungarian Joint Program.