Adaptive filters and internal models: multilevel description of cerebellar function

Area of Science:

• Neuroscience
• Computational Neuroscience
• Systems Neuroscience

Background:

• Cerebellar function is increasingly conceptualized through engineering frameworks for motor control and signal processing, often involving internal models.
• The 'chip metaphor' views the cerebellum as a homogeneous microcircuit with unique microzones, requiring knowledge of both the algorithm and connections to understand function.

Purpose of the Study:

• To investigate the relationship between the cerebellum and internal models using the chip metaphor.
• To assess the biological plausibility of implementing specific engineering schemes, such as adaptive filters, within the cerebellum's structure.

Main Methods:

• Utilized the adaptive filter algorithm as a candidate model for cerebellar microcircuit operations.
• Analyzed the biological plausibility of various engineering applications (e.g., Smith predictor, Kalman filter, feedback-error-learning) within the cerebellar framework.
• Examined plausible schemes like forward models and recurrent architectures for adaptive inverse control.

Main Results:

• Some proposed cerebellar applications, including state estimation (Smith predictor, Kalman filter) and feedback-error-learning, appear biologically implausible.
• Even plausible schemes like forward models and recurrent architectures may not have a simple mapping between microzone function and internal model structure.
• Cerebellar involvement in behaviors is unlikely to fit neat classifications like 'forward model'.

Conclusions:

• Cerebellar microzones likely implement task-specific adaptive filter operations, integrating diverse signal-processing roles.
• The cerebellum's function is better understood as a flexible adaptive filter rather than a dedicated implementation of specific internal models.
• This suggests a more generalized computational role for cerebellar microzones in motor control and signal processing.