Evolution of the cerebellum as a neuronal machine for Bayesian state estimation

Area of Science:

• Neuroscience
• Computational Biology
• Evolutionary Biology

Background:

• The cerebellum's evolution is linked to the electric and vestibular senses in early vertebrates, crucial for predatory orienting behaviors.
• Accurate sensory information is vital for precise control of movement and perception.

Purpose of the Study:

• To construct a computational neural model of cerebellar function based on electrosensory and vestibular inputs.
• To explore how neural circuits approximate Bayesian inference for prey localization and state estimation.
• To propose a generic model of cerebellar computation applicable to motor control, perception, and cognition.

Main Methods:

• Developed a computational neural model interpreting individual neuron spikes as measurements of prey location.
• Modeled spatial spike distribution in an electrosensory map as a Monte Carlo approximation of Bayesian posterior distribution.
• Extended the optimal filtering mechanism to handle dynamic targets and platforms, creating a spiking neural state estimator.

Main Results:

• The emergent neural circuit resembles cerebellar-like hindbrain electrosensory filtering circuitry found in sharks and other vertebrates.
• The model demonstrates how spiking neurons can implement Bayesian state estimators efficiently using fractional-order dynamics.
• The computational models function as a novel particle filter, utilizing spikes as particles.

Conclusions:

• The proposed model offers a plausible explanation for the cerebellum's role in motor control, perception, and cognition.
• The findings suggest a generic computational principle for cerebellar circuitry based on Bayesian inference and particle filtering.
• The study makes testable predictions about neural mechanisms within cerebellar circuitry.