A Neuropsychologically-Inspired Computational Approach to The Generalization of Cerebellar Learning S. D. Teddy 1 , E. M-K. Lai 2 , and C. Quek 3 Centre for Computational Intelligence, School of Computer Engineering, Nanyang Technological University, Nanyang Avenue, Singapore 639798. 1 sdt@pmail.ntu.edu.sg, { 2 asmklai, 3 ashcquek}@ntu.edu.sg Abstract. The CMAC neural network is a well-established computational model of the human cerebellum. A major advantage is its localized generalization prop- erty which allows for efficient computations. However, there are also two major problems associated with this localized associative property. Firstly, it is difficult to fully-train a CMAC network as the training data has to fully cover the entire set of CMAC memory cells. Secondly, the untrained CMAC cells give rise to unde- sirable network output when presented with inputs that the network has not pre- viously been trained for. To the best of the authors’ knowledge, these issues have not been sufficiently addressed. In this paper, we propose a neuropsychologically- inspired computational approach to alleviate the above-mentioned problems. Draw- ing inspirations from the psychological aspects of the generalization of motor skill learning, the proposed ”patching” algorithm strive to construct a plausible memory surface for the untrained cells in the CMAC network. We demonstrate through the modeling of human glucose metabolic process that ”patching” of un- trained CMAC cells offers a satisfactory solution to incomplete training data. 1 Introduction The human cerebellum is a brain region in which the neuronal connectivity is suffi- ciently regular to facilitate a substantially comprehensive understanding of its func- tional properties. It constitutes a part of the human brain that is important for motor con- trol and a number of cognitive functions [1], including motor learning and memory. The human cerebellum is postulated to function as a movement calibrator [2], which is in- volved in the detection of movement error and the subsequent coordination of the appro- priate skeletal responses to reduce the error [3]. It has been established that the human cerebellum functions by performing associative mappings between the input sensory in- formation and the cerebellar output required for the production of temporal-dependent precise behaviors [4]. The Marr-Albus-Ito model [5] describes how the climbing fibers of the cerebellum perform this function by transmitting moment-to-moment changes in sensory information for movement control. The Cerebellar Model Articulation Controller (CMAC) [6] is a neural network in- spired by the neurophysiological properties of the human cerebellum and is recognized for its localized generalization and rapid algorithmic computations. As a computa- tional model of the human cerebellum, CMAC manifests as an associative memory network [7], which employs error correction signals to drive the network learning and