Dr. Georg Keller
Education:
| PhD | ETH Zurich | Richard Hahnloser | How songbirds listen to themselves sing. | ||||||||
Post-doctoral:
| Postdoctoral Fellow | Max Planck Institute of Neurobiology | Mark Hübener & Tobias Bonhoeffer | Visuomotor integration in visual cortex. | ||||||||
Talk title: Predictive processing in cortical circuits
A defining aspect of our brains interaction with the world is the coupling between movement and the resulting sensory feedback. With experience the brain learns to associate specific movements with their sensory consequences and thus builds an internal model of the world. Based on this we speculate that much of what we perceive is not the result of what our sensory organs transmit to our brains but either the result of what we expect to perceive or the result of a large deviation from these expectations. Our work aims to understand the computational contribution of neocortex to this process. Our research focuses on mouse visual cortex and is guided by the ideas of predictive processing. In visual cortex, visual input is compared to predictions of visual input based on these internal models to compute prediction errors. Experience with self-generated visual feedback establishes a finely tuned circuit in visual cortex capable of computing prediction errors between top-down predictions and bottom-up visual input. Our results describe the cortical microcircuit that implements this computation, as well as contributing to our understanding of the molecular markers of the neurons with defined computational roles. Understanding and manipulating this circuit will be instrumental in advancing our understanding of perceptional disturbances, such as those observed in schizophrenia.
About Dr. Georg Keller:
The aim of Dr. Georg Keller and his lab is to understand the computational algorithm of neocortex. Their research is based on the central hypothesis that cortical function is governed by internal models of the environment. Sensory input is compared to predictions of sensory input based on these internal models to compute prediction errors. This would mean that much of what we perceive is not the result of what our sensory organs transmit to our brains but either the result of what we expect to perceive or the result of a large deviation from these expectations.
To investigate cortical processing, they use a combination of imaging and electrophysiological techniques to record activity of individual cortical neurons during a behavioral task performed in a virtual reality environment. They use molecular techniques to target our recordings to specific cell populations in order to identify the different functional elements of the circuit. Based on these experiments and theoretical considerations, they then attempt to infer the basic functional principles underlying cortical processing.