Versatility and adaptive performance

Abstract

We construct a class of adaptive networks whose purpose is to function in a volitile environment. The characteristic structure is a sparsely connected network with a low-activity internal control mechanism that functionally keeps the network adaptive and with a local-activity-dependent synaptic response to rewarded and penalized actions. The resulting dynamics, called adaptive performance, adjusts easily to changes in parameters induced by input or feedback, and it shows fluctuations over a wide range of time scales. Our specific studies concern conditioning and control. We speculate that adaptive performance may be an underlying mechanism in brain function, as well as provide a basis for new technology.