Low-resolution neuronal code: a theory

Abstract

ABSTRACTIt is a long-standing puzzle how insentient brain structures work ‘blind’, that is, seemingly with little or no information about the data they process. A plausible solution may be that, in some cases, they work in low resolution, i.e., sacrifice detail by disregarding it. That way, there is no need to read individual signals (respond in a way that reflects source and modality, for example). This feels counterintuitive because it seems wasteful, and also because it challenges the assumption that detail is desirable because it makes brain computations more powerful. Worse, physiological implementation faces problems. There must a threshold (data loss must be somehow limited or bounded), a way for lossy data to code information, a way to read it, a substrate and functional context. This is not a compromise but a strategy. The cerebellar cortex is a plausible substrate. The main source of excitatory input to the cerebellum, mossy fibres, terminate in the inner layer of the cerebellar cortex. We propose a neurophysiologically-detailed mechanism that recodes input to the cerebellum into internal signals. Resolution has spatial dimensions set by topography at the scale of long strips. The code is contained in collective parameters of firing rates, read by random sampling. We model the mechanismin silicoto quantify and test the ideas. The detail of the hypothesis and the simulation is paramount because it is crucial that biological messiness at local scale does not compromise performance. We find that low-resolution code is an excellent candidate to explain cerebellar neurophysiology.NEW AND NOTEWORTHYIt is a long-standing puzzle how brain structures work with so little information about the data they process. A seldom-considered but plausible solution is to work in low resolution, i.e., sacrifice detail by disregarding it. The problemthenis to coordinate the behaviour of functionally-grouped cells which may individually receive an entirely different mixture of eclectic inputs. We propose a biologically-detailed mechanism which solves that problem and may have applications in other brain regions.