Neuronal noise

Background

Neuronal activity at the microscopic level has a stochastic character, with atomic collisions and agitation, that may be termed "noise." While it isn't clear on what theoretical basis neuronal responses involved in perceptual processes can be segregated into a "neuronal noise" versus a "signal" component, and how such a proposed dichotomy could be corroborated empirically, a number of computational models incorporating a "noise" term have been constructed.

Single neurons demonstrate different responses to specific neuronal input signals. This is commonly referred to as neural response variability. If a specific input signal is initiated in the dendrites of a neuron, then a hypervariability exists in the number of vesicles released from the axon terminal fiber into the synapse. This characteristic is true for fibers without neural input signals, such as pacemaker neurons, as mentioned previously, and cortical pyramidal neurons that have highly-irregular firing pattern. Noise generally hinders neural performance, but recent studies show, in dynamical non-linear neural networks, this statement does not always hold true. Non-linear neural networks are a network of complex neurons that have many connections with one another such as the neuronal systems found within our brains. Comparatively, linear networks are an experimental view of analyzing a neural system by placing neurons in series with each other.

Initially, noise in complex computer circuit or neural circuits is thought to slow down and negatively affect the processing power. However, current research suggests that neuronal noise is beneficial to non-linear or complex neural networks up until optimal value. A theory by Anderson and colleagues supports that neural noise is beneficial. Their theory suggests that noise produced in the visual cortex helps linearize or smooth the threshold of action potentials.

Another theory suggests that stochastic noise in a non-linear network shows a positive relationship between the interconnectivity and noise-like activity. Thus based on this theory, Patrick Wilken and colleagues suggest that neuronal noise is the principal factor that limits the capacity of visual short-term memory. Investigators of neural ensembles and those who especially support the theory of distributed processing, propose that large neuronal populations effectively decrease noise by averaging out the noise in individual neurons. Some investigators have shown in experiments and in models that neuronal noise is a possible mechanism to facilitate neuronal processing. The presence of neuronal noise (or more specifically synaptic noise) confers to neurons more sensitivity to a broader range of inputs, it can equalize the efficacy of synaptic inputs located at different positions on the neuron, and it can also enable finer temporal discrimination. There are many theories of why noise is apparent in the neuronal networks, but many neurologists are unclear of why they exist.

More generally, two types of impacts of neuronal noise can be distinguished: it will either add variability to the neural response, or enable noise-induced dynamical phenomena which cannot be observed in a noise-free system. For instance, channel noise has been shown to induce oscillations in the stochastic Hodgkin-Huxley model.