DID YOU KNOW? DTIC has over 3.5 million final reports on DoD funded research, development, test, and evaluation activities available to our registered users. Click
HERE to register or log in.
Accession Number:
AD1105877
Title:
Principles of Robust Learning Derived from the Structure and Function of the Cortical Column
Corporate Author:
NORTHEASTERN UNIVERSITY BOSTON United States
Report Date:
2020-06-17
Abstract:
The goal of this project is to uncover the effects of learning and long-term memory storage on synaptic connectivity, thus,creating the basis for quantitative analyses of these fundamental brain functions. To that end, we developed a model of a biologically-inspired recurrent neural network, capable of storing and reliably retrieving temporal sequences of network states. The model incorporates many basic elements of local connectivity in the mammalian neocortex, including excitatory and inhibitory neuron classes, neuron morphologies, the homeostatic constraint on connection weights, and several types of errors and noise in signal transmission. The model was solved analytically and numerically with the replica and convex optimization methods. In addition, a perceptron-type learning rule was developed to load associative memory sequences into the network in a biologically-plausible online manner. Our results revealed that when individual neurons are robustly loaded with a near-maximum amount of memories they can support, the network develops many structural and dynamical properties that are consistent with experimental observations.
Descriptive Note:
Technical Report,30 Sep 2015,29 Sep 2018
Pages:
0017
Distribution Statement:
Approved For Public Release;
File Size:
0.97MB
