Neural Coding: How Neurons Represent Information

Rate codes and timing codes, population vectors, sparse coding and 'concept cells' that respond to Jennifer Aniston: how patterns of spikes carry information about the world and our actions.

Every sight, sound, memory and movement corresponds to patterns of action potentials. Neural coding asks how those patterns carry information — and how other neurons read it.

Rate coding

In the 1920s Edgar Adrian found that sensory fibres signal stimulus strength by firing rate: a heavier weight on a muscle produces more spikes per second. Rate coding is robust and common, especially in sensory and motor systems.

Temporal coding

But timing can matter. In the auditory system, neurons fire locked to the phase of sound waves, letting the brain detect microsecond differences between the ears. In the hippocampus, phase precession relative to the theta rhythm adds information about position. The very first spikes after a stimulus can carry a surprising amount of information — enough for rapid recognition. Synchrony among neurons may also bind features together, though this remains debated.

Population coding

Single neurons are noisy, so information is usually carried by populations. In 1986 Apostolos Georgopoulos and colleagues found that motor cortex neurons are broadly tuned to the direction of arm movement; adding each neuron's preferred direction, weighted by its firing rate, produced a population vector that accurately predicted the movement. Modern brain-computer interfaces decode intended movements and speech from populations of neurons in a similar spirit.

Sparse coding and concept cells

How many neurons represent one thing? At one extreme is the hypothetical "grandmother cell" that fires only for your grandmother. In 2005 Rodrigo Quian Quiroga, Itzhak Fried, Christof Koch and colleagues recorded from neurons in the medial temporal lobes of epilepsy patients and found one that fired for different photos of the actress Jennifer Aniston — and not for other people — and another for Halle Berry, even responding to her written name. These concept cells suggest highly selective, sparse coding in memory regions, though each probably responds to several related concepts and each concept is represented by many cells.

Representational geometry

Increasingly, researchers analyse the geometry of population activity — how patterns for different stimuli relate in a high-dimensional space — and find that the brain often represents information on low-dimensional "manifolds".