Walter Senn, Clinical Neuroscience Bern: Is there a Newtonian law for neurons and synapses? - A least action principle for neurobiology

Clinical Neuroscience, Institut für Physiologie, Computational Neuroscience Lab, Bern, Switzerland

When	Feb 02, 2021 from 05:15 PM to 06:00 PM
Where	Zoom Meeting. Meeting ID and password will be sent with e-mail invitation. You can also contact Fiona Siegfried for Meeting ID and password.
Contact Name	Fiona Siegfried
Add event to calendar	vCal iCal

Abstract

Computational neuroscience spans the wide range from biophysical modeling of neurons to cognitive theories and intelligence. There are successful examples in the related disciplines of physics and artificial intelligence in building comprehensive theories. In physics, the principle of least action offers an axiomatic approach to most physical disciplines, and in artificial intelligence the paradigm of deep learning has unprecedented impact on modeling cognition. I will present a neuronal principle of least action that is formulated in the spirit of the theoretical physics, and that connects the biophysics of neurons with cognition. The central notion of this principle is the one of an error. Errors appear on the behavioral scale, and we claim that they also appear within the smallest scale of the brain, the neurons. A pyramidal neuron, for instance, is a local predictor that tries to predict more informed input to the apical region by less informed input to the basal region. Cortical microcircuits transform global information to a local error representation that drives learning in the various cortical layers. The neural least action principle suggest that cortical circuits do not implement vanilla backprop, but instead a data-driven pseudobackprop — that still has to investigated in terms of performance.