Adaptive filter models

Abstract

Adaptive-filter models are widely used to understand cerebellar function. Major features include the resemblance between their basic structure and that of the cerebellar cortical microcircuit, their use of a simple decorrelation algorithm known to be effective for processing physical signals, and their capacity to be connected appropriately for predicting the sensory consequences of movement commands, and for learning accurate movements. One method for evaluating adaptive-filter models is to examine the input-output signals of individual cerebellar microzones: current evidence suggests that the flocculus can act like an adaptive filter, but the relevant information is available for few other regions. A second method is to examine microcircuit features for compatibility with the model. A key task for this form of evaluation is deciding how far the apparently incompatibility of features signifies (1) experimental artifact, (2) the unsuspected requirements of an adaptive filter when it is implemented neurally, or (3) the operation of a new algorithm. Adaptive-filter models have thus the potential for guiding both future experimental and modeling studies.