Our brains can learn a model about the world by continuously making predictions, and comparing the prediction from this inner model with the outcome from the world. In case of a mismatch, the error can be used to update the model in principle. These conceptual ideas date back to Helmholtz. But how is such learning realized on the neuronal level? We propose that dendritic branches play a central role in representing the encoding error – implementing “dendritic error computation”, where the encoding error is represented as a deviation from resting potential. This concept arises from analytical derivation of optimal synaptic learning rules for efficient coding. In the second part of the talk we discuss how, on the network level, homeostatic regulation implements a fine balance between excitation and inhibition, enabling for rapid and flexible changes of computational properties. Overall, spanning the arch from analytical principles of neural computation to physiological implementation on the synaptic and neural level, we contribute to a fundamental understanding of learning in living neural networks.

Certificate of attendance: Please contact team assistant serenella.brinati.1(at)hu-berlin.de

This talk is hosted by SFB1315 Speaker Matthew Larkum (Sub-projects A04, A10 and Z).

Lecture Poster

Series Overview