Information encoding by individual neurons and groups of neurons in the primary visual cortex

摘要

How is information about visual stimuli encoded into the responses of neurons in the cerebral cortex? In this thesis, I describe the analysis of data recorded simultaneously from groups of up to eight nearby neurons in the primary visual cortices of anesthetized macaque monkeys. The goal is to examine the degree to which visual information is encoded into the times of action potentials in those responses (as opposed to the overall rate), and also into the identity of the neuron that fires each action potential (as opposed to the average activity across a group of nearby neurons). The data are examined with techniques modified from systems analysis, statistics, and information theory. The results are compared with expectations from simple statistical models of action-potential firing and from models that are more physiologically realistic. The major findings are: (1) that cortical responses are not renewal processes with time-varying firing rates, which means that information can indeed be encoded in the detailed timing of action potentials; (2) that these neurons encode the contrast of visual stimuli primarily into the time difference between stimulus and response onset, which is known as the latency; (3) that this so-called temporal coding serves as a mechanism by which the brain might discriminate among stimuli that evoke similar firing rates; (4) that action potentials preceded by interspike intervals of different durations can encode different features of a stimulus; (5) that the rate of overall information transmission can depend on the type of stimulus in a manner that differs from one neuron to the next; (6) that the rate at which information is transmitted specifically about stimulus contrast depends little on stimulus type; (7) that a substantial fraction of the information rate can be confounded among multiple stimulus attributes; and, most importantly, (8) that averaging together the responses of multiple nearby neurons leads to a significant loss of information that increases as more neurons are considered. These results should serve as a basis for direct investigation into the cellular mechanisms by which the brain extracts and processes the information carried in neuronal responses.