Cosyne 2008 Workshops
March 3-4, 2008
Snow Bird, Utah
Data sharing and modeling challenges in neuroscience - a first step towards predictive neuron models?
Arnd Roth (UCL)
Wulfram Gerstner (EPFL)
Fritz Sommer (UC Berkeley): email@example.com
Data sharing in neuroscience is currently pushed by modelers, experimentalists, and public agencies, for achieving various -not always compatible- goals:
- to test the predictions of models
- to help with parameter settings in large-scale simulations
- to overcome the limitations of the traditional "single lab" approach
- to expose experimental data to the full spectrum of available analysis methods
- to make the results of publicly funded science available to everybody
To achieve these goals, the sharing of raw data as such seems not sufficient. Rather, data sharing in neuroscience requires additional efforts to organize the raw data. For example, it is typically required to link raw data to detailed protocol descriptions, to stimuli or behavioral responses.
A discussion of questions of how to organize neuroscience data for sharing and what to expect from such activities seems timely since public support for data sharing activities has been recently brought into place, for example, a new NSF program for funding efforts in experimental labs to make data publicly available.
One specific topic of this workshop is to discuss recent experiences to organize neuroscience data by defining "modeling challenges". A modeling challenge puts data sets in the context of particular questions to be addressed, for instance by specifying a task that a model for the data should be able to achieve. First steps in this direction have been taken by the Berkeley "Neural Prediction Challenge" or by the Lausanne "Competition: Predicting Single-Neuron Behavior".
This discussion-oriented workshop aims to bring three different groups of people together: theoreticians interested in quantitatively predictive models; experimentalists interested in data sharing; and proponents of data sharing initiatives and challenges.
General questions to be addressed:
- How can we quantify the predictive power of current neuron models?
- How can we make use of publicly available data?
- What type of data would modelers like to see publicly available?
- What type of data would experimentalists like to see publicly available?
Format: Relatively short talks (15 minutes) with enough time for discussion after talks and in one or two general discussion sessions.