Entwicklung funktional leistungsfähiger und biologisch realistischer Modelle der neuronalen Signalverarbeitung

01/01/2016 to 31/12/2017

How cells in the brain represent, combine and process information is still an open question. One major hurdle in answering this question is the stochasticity of neural responses: even identical stimulation causes different activity patterns. The source of this noise and its effects are difficult to assess experimentally. In addition, simplified theoretical models have so far been unable to provide a consistent picture, and so the interrelation between noise and perceptual performance is still unknown. One major shortcoming of previous models is their ignorance with respect to the hierarchical structure and performance-optimized connectivity of the neocortex. New ansatzes that take these features into account will open new possibilities to gain a complete and robust understanding of neural signal encoding and its effects on perceptual performance.

In this project we aim to exploit recent breakthroughs in machine learning and self-organised neural networks to build new population models. To this end we generalise a functional, hierarchical neural network (with exceptional perceptual performance like object recognition) and compare important information theoretic analysis with the results of established models. In a subsequent step, we adapt this model to the biophysical constraints of the neocortex, and compare model predictions with experimental measurements.