E13 Neural coding

Topic

Neural coding studies how information about a stimulus or external variable is represented in the neural response—that is, in the activity patterns of one or more neurons. The instantaneous firing rate summarizes this neural response by transforming a train of discrete action potentials into a single continuous variable that encapsulates the response at any given moment.

Without assuming a specific coding mechanism, information theory tools—such as entropy and mutual information—quantify in general terms how much uncertainty regarding the stimulus is reduced by observing the neural response: the greater the mutual information between the stimulus and the neural response, the more the uncertainty about the stimulus decreases upon observing that response.

Many sensory neurons exhibit bell-shaped tuning curves in their neural response relative to a stimulus variable: the response of such a neuron peaks when the stimulus takes on the neuron's preferred value and decreases as the stimulus value deviates from that preference in either direction.

The coefficient of variation of the inter-spike intervals in the neural response quantifies the regularity or irregularity of the firing pattern: a higher coefficient indicates a more irregular firing pattern, while a lower coefficient indicates a more regular one.

At the population level, the stimulus can be reconstructed from the simultaneous neural responses of many sensory neurons—either through optimal linear decoding or the population vector method, which weights each neuron's contribution based on its tuning preference and its instantaneous firing rate at that moment. Taken together, it is this very neural response—summarized by the instantaneous firing rate, characterized by its tuning curve and the regularity of its temporal firing pattern, and quantified using information theory—that makes it possible both to understand how much a neuron knows about a stimulus and to reconstruct that stimulus from the activity of an entire population of neurons.

ID:4271