Dynamical Properties of Self{Regulating Neurons

The phenomenon of learning is still an open problem. The present work performs one step towards a model of learning that is inspired by early ideas of cybernetics and constructivism and makes use of the mathematical formalisms of dynamical system theory and articial neural network theory. Homeostasis is proposed as principle of synaptic learning in recurrent neural networks, and an analysis of stabilisation properties of the suggested neuron model are investigated, by means of phenomenological analysis of system dynamics. The model of synaptic regulation is placed in a larger framework of an arti cial life approach to adaptation and cognition. It is conjectured that the proposed mechanism of synaptic regulation, if employed in a structurally appropriate nervous system, can realise adaptive behaviour. Acknowledgements I want to thank the INDY group at the Fraunhofer Institute for Autonomou