Robust Reservoir Generation by Correlation-Based Learning
Figure 1
Schematic of our network model. The model consists of two networks: the reservoir and readouts. The reservoir
is a recurrent network composed of excitatory and inhibitory neurons.
Excitatory neurons receive external inputs representing conditioned stimuli
(CS, (4), (5)). Four types of connections, excitatory to excitatory, excitatory to
inhibitory, inhibitory to excitatory, and inhibitory to inhibitory, are
assumed. Connection weights between excitatory neurons are modifiable by CBL
(see (3)), whereas the weights of other connections are fixed as in
conventional RC models. On the other hand, a readout neuron receives inputs
from excitatory reservoir neurons (, ) and the input representing unconditioned
stimuli (US, (12), (13)). The connection weights between excitatory reservoir neurons
and the readout neuron are modifiable by supervised learning rule (see (11)).