Realistic modeling applied to cerebellar function
Faculty of Pharmaceutical, Biomedical and Veterinary Sciences . Biomedical Sciences
NETWORKS, VOLS 1-3
IEEE International Conference on Neural Networks (ICNN)
International Joint Conference on Neural Networks (IJCNN 02), MAY 12-17, 2002, HONOLULU, HI
, p. 75-76
University of Antwerp
Realistic network and single cell modeling can be powerful tools to study the operation of real brain structures. We have used realistic simulation approaches based on fully active compartmental models to study the cerebellum. Here I summarize some of our recent results that address cerebellar learning. Assuming that long-term depression (LTD) is the basis of learning in cerebellar Purkinje cells, two questions arise: what is a parallel fiber pattern and how can the occurrence of learned patterns be decoded from the Purkinje cell spike train? We have used detailed network simulations of the granular layer using conductance based models of granule and Golgi cells to study the patterns of parallel fiber activity in response to natural mossy fiber stimulation. Based on these simulations we predict a sparse temporal coding by parallel fibers of complex spatial maps of mossy fiber input. We have studied for the first time pattern recognition by Purkinje cells in a realistic context, i.e. how to recognize the effect of depressed synapses on a spiking neuron which is spontaneously active most of the time. Using our standard Purkinje cell model we predict that parallel fiber patterns that have undergone LTD cause an increase In the Purkinje cell output instead of the decrease assumed by most cerebellar learning theories.