ABSTRACT
It is thought that neuronal signal frequencies – firing rates – provide the physiological form of values that are used in neural network computations. The cerebellum is made up of repeating, anatomically overlapping networks. Neural network modelling has a long tradition of learning algorithms that are implemented physiologically by learned changes of synaptic transmission strength. In the cerebellum, the evidence remains inconclusive. We ask, with a focus on the input layer of the cerebellar network: (1) If the physiological instrument of the cerebellar network computation is not synaptic memory, what is it, and what is the computation? (2) What is the code? (3) How is that related to the matrix-like internal architecture of the cerebellar cortex, and the well-known functional organisation of the cortex into long, thin strips called microzones? We propose a reinterpretation of the evidence which plausibly supports controversial conclusions. Using the same numbering: (1) The granular layer computation is the functional result of anatomy. Simply passing signals through the entry layer sub-network provides the biological equivalent of mathematical operations, which run in series and in parallel. (2) Information is collectively coded in statistics of the firing rates of co-active cells. Coded this way, there are reliable statistical effects that conserve rate information when input to the cerebellum is converted to internal signals. (3) Still more unexpectedly, multi-dimensional detail contained in input signals is converted to an extremely low resolution – pixelated – internal code. This provides an otherwise elusive mechanism to coordinate firing of microzone-grouped Purkinje cells, which control the output cells of the cerebellar network. It is also a theoretical substrate for a path of rate codes that connect input to output of the cerebellar network. Since we argue that anatomy is the chief instrument of the granular layer computation, we can model it in detail, down to local anatomical noise. If this approach is sound, the idea that anatomy implements computations unaided is likely to have implications for coding in other brain regions.
Competing Interest Statement
The authors have declared no competing interest.
Footnotes
Revisions to the text of the manuscript have been made to improve clarity. The main changes are to the title, abstract, introduction and discussion.