| Polyhomeostasis is taking place when a dynamical system tries to achieve a non-trivial distribution function for its time-averaged output. Polyhomeostasis is a form of meta-learning generalizing homeostasis, the regulation of a single scalar quantity. We consider a neuron driven by white noise of constant strength and polyhomeostatically adapting its firing rate in order to achieve a maximal information entropy for its output. For the case of a single neuron and sparse coding we observe the emergence of self-organized tipping transitions between competing transient attractors. For the case of networks of autonomously adapting neurons self-organized chaotic and intermittently bursting dynamics emerges. |
|