The cerebellum converts input data into a hyper low-resolution granule cell code with spatial dimensions: a hypothesis.

We present a theory of the inner layer of the cerebellar cortex, the granular layer, where the main excitatory input to the cerebellum is received. We ask how input signals are converted into an internal code and what form that has. While there is a computational element, and the ideas are quantified with a computer simulation, the approach is primarily evidence-led and aimed at experimenters rather than the computational community. Network models are often simplified to provide a noiseless medium for sophisticated computations. We propose, with evidence, the reverse: physiology is highly adapted to provide a noiseless medium for straightforward computations. We find that input data are converted to a hyper low-resolution internal code. Information is coded in the joint activity of large cell groups and therefore has minimum spatial dimensions-the dimensions of a code group. The conversion exploits statistical effects of random sampling. Code group dimensions are an effect of topography, cell morphologies and granular layer architecture. The activity of a code group is the smallest unit of information but not the smallest unit of code-the same information is coded in any random sample of signals. Code in this form is unexpectedly wasteful-there is a huge sacrifice of resolution-but may be a solution to fundamental problems involved in the biological representation of information.