A MODEL THAT ACCOUNTS FOR ACTIVITY IN PRIMATE FRONTAL-CORTEX DURING ADELAYED MATCHING-TO-SAMPLE TASK

A fully recurrent neural network model was optimized to perform a spat ial delayed matching-to-sample task (DMS). In DMS, a stimulus is prese nted at a sample location, and a match is reported when a subsequent s timulus appears at that location. Stimuli elsewhere are ignored. Compu tationally, a DMS system could consist of memory and comparison compon ents. The model, although not constrained to do so, worked by using tw o corresponding classes of neurons in the hidden layer: storage and co mparator units. Storage units form a dynamical system with one fixed p oint attractor for each sample location. Comparator units constitute a system receiving input from these storage units as well as from curre nt input stimuli. Both unit types were tuned directionally. These two sources of information combine to create unique patterns of activity t hat determine whether a match has occurred. In networks with abundant hidden units, the storage and comparator functions were distributed so that individual units took pal? in both. We compared the model with s ingle-neuron recordings from premotor (PM) and prefrontal (PF) cortex. As shown previously, many PM and PF neurons behaved like storage unit s. In addition, both regions contain neurons that behave like the comp arator units of the model and appear to have dual functionality simila r to that observed in the model units. No neuron in either area had pr operties identical to those of the match output neuron of the model. H owever, four PF neurons and one PM neuron resembled the output signal more closely than any of the hidden units of the model.