Short term memory phenomena in an autosynaptic neuron

Addresses the question of why some kinds of connected neural net can store temporary information (the time representation problem). One line of research uses results of biological investigations where it appears that time representation involves a (maybe adaptive) short-term memory mechanism in the neural net. Another is based on hysteresis effects caused by multistable states in the neural population. However, the authors' previous works have shown that even single neuron models including self-interactions or complex activity functions may exhibit multistable levels of activity. They extend this previous work and study the dynamic behavior of a simple neuronal circuit formed by a single neuron with an autosynapsis, that is, a neuron that has its own output fed back as one of its input signals. Additionally, they relate the neural dynamics to short-term memory phenomena. Their neuron model is built with three elementary neuronal operations, two linear and one nonlinear. The two linear operations mimic the dendritic and somatic operations found in the biological neurons, meanwhile the nonlinear operation or activity function mimics the axonic and terminal operations of the neuron.