Max Planck Institute for Dynamics and Self-Organization -- Department for Nonlinear Dynamics and Network Dynamics Group
Personal tools
Log in


Tuesday, 15.01.2013 17 c.t.

Computing with neural synchrony: an ecological approach to neural computation

by Prof. Dr. Romain Brette
from Département d'Etudes Cognitives, Ecole Normale Supérieure, Paris, France

Contact person: Andreas Neef


Ludwig Prandtl lecture hall


What is synchrony good for? The main paradigm in standard neural network theory of perceptual systems is pattern recognition. A given object can produce many different sensory patterns, so a key computational issue is invariance to these different "views" of the object. But a number of philosophers and psychologists have argued that perception is not about recognizing patterns but about discovering the laws that govern sensory signals, which are indeed highly structured in a natural environment. The invariance problem disappears because laws are precisely invariants. James Gibson coined the terms "ecological approach" to describe this change in focus. I will describe an attempt to develop a neural network theory that follows this ecological approach. The key insight is that neural synchrony reflects the structure of sensory signals, which can then be detected by coincidence detection. Spike-timing-dependent plasticity then implements structure learning. I will illustrate these concepts with examples from hearing and olfaction.

back to overview