A biologically plausible model of associative memory which uses disinhibition rather than long-term potentiation.

Mammalian memory is commonly "explained" in terms of long-term potentiation (LTP) of excitatory synapses. However, depotentiation of inhibitory pathways (disinhibition) is also a known phenomenon in the brain. Artificial neural networks which are offered as partial models of the cerebrum traditionally encode memory by the potentiation of excitatory "synapses" in a manner which is thought of as analogous to LTP. Analysis shows that such models have seriously limited storage capacities. The models also depend on mechanisms which do not appear to be biologically plausible. This paper demonstrates that these difficulties are avoided by encoding memory by means of disinhibition rather than LTP. The resulting models are simple and plausible, though unconventional.