Neural population modeling and psychology: A review

The development of models of such psychological effects as reinforcement, attention, coding of feature detectors, and interactions between short-term and long-term memory is outlined, together with that of multi-level networks that combine modifiable synapses and lateral inhibitory anatomies.

Abstract

The development of neural population modeling as it relates to psychology is traced from the early 1940s to the present. The evolution of the field has been from descriptions of randomly connected neurons transmitting all-or-none signals to analyses of structured multi-level networks whose dynamics involve several different spatial and temporal scales. The cybernetic revolution of the 1940s led to the incorporation into digital neural models of such concepts as linear threshold logic, redundant computation, and information. Each of these concepts has more recently been synthesized with learning to generate a set of adaptive neural models. Concurrently, a variety of data from neurophysiology and from experimental psychology suggested models that incorporated continuous and nonlinear effects. Since the late 1960s there has been much activity in the design of rules for modifiable synapses in models of learning or conditioning. There has also been much activity in the design of lateral inhibitory networks that model sensory pattern storage. The development of models of these effects is outlined, together with that of multi-level networks that combine modifiable synapses and lateral inhibitory anatomies. These multi- networks model such psychological effects as reinforcement, attention, coding of feature detectors, and interactions between short-term and long-term memory.