Associative anticipatory learning and control of the cerebellar cortex based on the spike-timing-dependent plasticity of the parallel fiber-Purkinje cell synapses (Q6072432)

The present article is in a certain sense a completion and extension of two other earlier works of the author in connection with the cerebellar cortex modelling and these are: [\textit{M. Fujita}, Neural Comput. 25, No. 6, 1440--1471 (2013; Zbl 1415.92016)], and [\textit{M. Fujita}, Neural Netw. 75, 173--196 (2016; Zbl 1415.92015)]. Results of both papers are extensively referenced in the present article. The concept in the development here is to integrate the long-term depression (LTD) of the synapse into the usual learning process of the cerebellar cortex model. The proposed associative anticipatory learning (AAL) theory integrates the temporarily asymmetric plasticity of the parallel fiber (PF)-Purkinje cell (PC) synapses. However the basic network structure is the same as in [Fujita, 2016, loc.cit.]. In the second section one reviews some basics on universal function approximation theory, explains the way AAL works and presents theoretical results on the associative universal function learning. Two types of AAL are considered and results on existence, convergence and stability are proved. Further the theory is applied to stabilize by AAL, a time-delayed inverted pendulum. Simulation strategy as well as simulation results on learning with both types of associative anticipatory learning: AAL Type 1 and AAL Type 2 are provided in the third section. Discussions and Conclusions are to be found in the fourth and fifth sections. The Appendix contains additional statements to a proof of a result on stability and to applied simulation strategy.