Cosyne 2007 Workshops
February 26-27, 2007
The Canyons, Utah
Workshop Title
Asking why - normative models in neuroscience
Organizer(s)
Alan A. Stocker: alan.stocker@nyu.edu
Konrad P. Kording: homepage
Abstract
We expect that over the course of evolution many properties of the nervous system became close to optimally adapted to the statistical structure of problems the nervous system is usually faced with. Substantial progress has been recently made towards understanding the nervous system on the basis of optimality: asking "why" the nervous system is solving problems the way it does.
Normative models typically start with an analytical formulation of which problem the nervous system has to solve, and propose an answer: how the nervous system "should" optimally solve this problem given its limited amount of neural resources. Such a principled framework seems particularly important for understanding complex systems, where pure descriptive models often cannot provide satisfying answers. Normative models have had great success in explaining a wide range of aspects of neural processing, from neural response characteristics to human social behavior. For example, the way sensory neurons encode information is frequently phrased as a problem of most efficiently representing sensory information. As another example, the way people combine cues within the same modality or between modalities is often formulated as optimal inference given the constraint of sensory uncertainty. Such models, by specifying the problem to be solved in a meaningful way, often explain counterintuitive properties of neurons as well as the behavior of humans in perceptual, action and cognitive tasks.
The goal of this workshop is to bring together theoretical and experimental neuroscientists interested in understanding why the brain is behaving the way it does. Experts from different fields, including neurophysiology, perception, cognition and behavior, will present their newest research results. Yet, the focus of the workshop is on discussions among participants and audience.
In particular, the workshop should lead us toward answers of the following questions: What is the state-of-the-art in applying normative models in neuroscience? Where are the limitations of normative models in providing answers in neuroscience? Can we link and unify some of the different approaches (e.g. efficient coding and optimal estimation)?
Program
8.30-8.30 introduction
8.40-9.15 tom griffiths: "Normative models of human inductive inference"
9.15-9.50 pascal mamassian: "Priors and cost functions in visuo-motor psychophysics"
9.50-10.10 break
10.10-10.45 david knill: ""Bayesian models for robust cue integration and learning"
10.45-11.20 wrap-up
__
4.30-5.05 alan stocker: "The importance of addressing both perceptual bias and variability"
5.05-5.40 daeyeoul lee: "What we should do, what we do, and how we do it"
5.40-6.00 break
6.00-6.35 barak pearlmutter: "A Rant about Names and Theories in Neuroscience"
6.35-7.10 bruno olshausen: "Generative models and visual cortical function"
7.10-7.30 final discussion
