Viola Priesemann: Learning in Living Neural Networks - from homeostatic regulation to dendritic predictive coding

Max-Planck-Institut für Dynamik und Selbstorganisation Göttingen

Homeostasis is ubiquitous in living systems, from basic physiological regulation to societal organization. For neural systems, homeostasis is key to maintaining an intermediate level of activity. We will introduce the basic principles of homeostasis and show how it shapes network dynamics and modulates information transfer, depending on the statistics of network input. Subsequently, we will derive local, unsupervised learning rules for efficient coding and demonstrate how they may be implemented in dendritic compartments. By utilizing the local membrane potential - representing the encoding error - the learning rules lead to sparse, independent representations, even if input features are highly correlated. In sum, we derive homeostatic regulation and local learning rules from first principles and finally provide a brief outlook on how similar mechanisms shape society. In doing so, we shed light on the generality and robustness of such self-regulating mechanisms - and their limitations.

Guests are welcome!

 

Organized by

Benjamin Lindner / Margret Franke