A neural network model of serial order recall from short-term memory

The authors present a neural network that converts a spatial pattern encoding the serial order of a list of items into a corresponding sequence of output events. Performance can be initiated, interrupted, and continued under control of a single, independent signal. The network uses recurrent connections to normalize and store node activity in short-term memory, to choose the items in sequence, and to modulate the output rate during readout. The latency and duration of output events can model certain empirical trends of human performance in recalling lists of items