Cosyne 2005 Workshops
March 21-22, 2005
Snowbird, Utah
Workshop Title
Bayesian approaches to sensory and motor processing
Organizer(s)
Konrad Koerding
Abstract
This workshop aims to bring together experimentalists and statisticians to discuss novel ways of addressing the way the nervous system deals with statistical problems. The workshop will include short talks but will focus on discussions among participants and audience.
The sensors of humans are plagued by noise, and likewise our muscles produce noisy outputs. Even if we had perfect sensors they would only tell us about the part of the world that we can currently sense. Uncertainty that stems from noise or the lack of complete knowledge places estimation of attributes of the worlds and control of our actuators firmly within a statistics framework. Bayesian statistics is the principled way of dealing with such problems. Much research in the field of machine learning addresses how artificial systems can optimally solve such problems. These algorithms now allow efficient handling of complicated real world data. Recently many counterintuitive finding, such as visual illusions, in neuroscience can be understood by considering the brain is performing Bayesian estimation. At the same time studies start to appear where the neural correlate of such statistical processing is analyzed. Therefore the Bayesian approach may provide a framework to understand and macroscopic and microscopic attributes of neural function. It suggests novel experiments as well as novel interpretations of previous experiments and places a broad range of experimental results in a coherent theoretical framework. The approach supports more powerful quantitative models of how the brain works, as well as more principled explanations of why it works the way it does.
This workshop is timely as over the course of the last couple of years many researchers started addressing Bayesian processing in the nervous system and the technical field of Bayesian processing has made impressive progress.
The workshop will bring together:
1: Psychophysicists who analyze how people behave as predicted by Bayesian statistics;
2: Physiologists who address how the brain implements a Bayesian algorithm;
3: Statisticians who develop Bayesian methods
Questions to be discussed:For which problems does the brain use Bayesian algorithms? Which kinds of such algorithms does it use? What types of probability distributions can be represented? Are the algorithms used by the brain similar in different sensory systems? Is there a fundamental difference between sensory and motor modalities? How are probabilities combined with rewards? What is the neural substrate underlying the representation of probabilities? What is the relation between probabilities and decisions?
