New model of information processing at granule cell layer makes cerebellum as biological equivalent for ANFIS and CANFIS: Sharing of processing resources and generalization

Puzzling about cerebellum is its wide range of functional involvement with its seemingly uniform structure. Late views based on connectivity with other parts of the brain implicate it, from traditional lower level role in balance and coordination of movements, up to cognition and emotional level. Cerebellar models are used in attempts to explain this functional involvement, with possible benefits in solving similar technical problems. Information processing in the input layer used by these models can easily reach biological and practical limits in number of neurons required, even though cerebellum hosts more than half of total number of neurons in the brain, most of them in the input layer. To relax these limits for some representative tasks, new model of information processing at input layer, named granule cell layer, is used. This paper presents information processing at the basic granule cell-Golgi cell (GrC-GoC) building block of the first cerebellar processing layer, and overall cerebellar processing becoming similar to fuzzy models of Adaptive-Network-Based Fuzzy Inference System (ANFIS) and CoActive Neuro-Fuzzy Inference Systems (CANFIS) types. Simulink spiking model illustrates functionality of GrC-GoC block. It generates as a result multidimensional receptive fields from traditional group of mossy fibers with population code that drive GrC, which are gain modulated by other input of GoC driven by second group of mossy fibers with rate code. Simulink behavioral model of a cerebellum with new structure is used in two robotics applications. Model shows wider sharing of processing resources, and generalization capabilities that it offers, with more biologically plausible learning.