The Learning of Insertions by the Cerebellum

A computational model of the cerebellum is developed to learn how to guide the learning of insertions in syntax or motor responses by means of a teacher (assumed in the hippocampus). A short simulation is shown, in which LTP/LTD is used for synapses from the granule cells onto the Purkinje cells, attaining suitable effectiveness. Introduction The cerebellum is well known to be important for learning temporal aspects of control, and its features in developing temporal sensitivity have been explored by numerous experimental studies of conditioned leaning [1, 2]. It has also been recognized as important in supporting cognitive processes, both from the effects of cerebellum deficits on language processing as well other cognitive task areas. There are also effects of interaction and combination of the cerebellum and hippocampus which prove of importance, as trace conditioned learning shows [3]. More generally we expect the cerebellum to be the basis of error− based learning driving the learning of action responses (including motor planning and covert language use) from being under attention from parietal and prefrontal control to be housed more securely at an automatic response level in prefrontal circuits crucially involving the basal ganglia. The purpose of our work is to develop a model of the cerebellum based on the known architecture it possesses in the brain, and then attempt to extend the ability it has been shown to possess computationally [1, 2, 3], to learn to be sensitive in response to timing in conditioned learning to similar sensitivity that we suspect is at the basis of syntactic analysis in language and in planning of motor actions. In particular we will attack the problem of insertions of endings (or also pre−fixes, although that will not be considered specifically but should also be amenable to our treatment). We extend the architecture modeled in [1, 2, 3] by considering feedback from the deep cerebellar nucleus (DCN) to the pons so as to guide the insertion process. However our paradigm needs a teacher throughout, acting as an unconditioned stimulus (US). We assume this arises from the hippocampus or nearby parahippocampal cortex, to be used later to help improve the production, under cerebellar guidance, of the initially incorrect but later corrected responses/stem−ending. Thus our simulation is in two stages: 1) We consider the learning of a stem representation in the cerebellar cortex; 2) We then develop the learning of the stem−ending temporally defined response, assuming that this total sequence is presented as the US in a conditioned−type of learning process. The paper consists of a description of the model, a description of the simulation results, and a final discussion. Description of Cerebellum The Cerebellar cortex is composed of golgi cells (GoC), granule cells (GrC), basket inhibitory interneurones (BK) and purkinje (PK) cells, as in figure 1. There are 2 routes of input to the cerebellum: the climbing fibres and mossy fibres. The climbing fibres originate in the Inferior Olive (IO) and are excitatory to the PK cells, with a single synapse on each PK node. The mossy fibre input, also excitatory, originates in brain stem nuclei such as the PONS, and contacts the GrC and GoC, and (external to the cerebellar cortex) the deep cerebellar nuclei (DCN). It has been suggested that GoC inhibition on GrC forms time−windows from GrC firing [4]; another effect of the connections between these neurones is the setting up of 2 populations of GrC: an on set and an off set [3] when there is an input via the PONS. PK neurones receive excitatory input from GrC and IO, and inhibitory input from the BK nodes; PK neurones have intrinsic currents which lead to spontaneous firing rates, which when combined with its inputs leads to resting firing rates of 50 −100Hz. PK cells inhibit 2 populations of DCN neurones: excitatory (DCN+) and inhibitory (DCN−); further inputs to the DCN come from the pons. Reciprocal connections from DCN+ to the pons allows for the reintroduction of excitatory activity into the cerebellar cortex, whereas BICS 2004 Aug 29 Sept 1 2004