Many labs world-wide use routinely integrate-and-fire or Hodgkin-Huxley models for describing single-neuron properties in network simulations.
What is the best variant of integrate-and-fire models, leaky, quadratic, or exponential? Do we need an adaptation variable?
Should we take into account refractoriness? Should we include a noise-term for intrinsic variability?
Will biophysical neuron models of the Hodgkin-Huxely type perform better than integrate-and-fire models?
If so, how many compartments and how many ion currents?
To answer these questions we made experimental single-neuron current injection and conductance injection data publicly available and posed a competition:
Who is the best in predicting the spike-times of the experimental neurons?
Who is the best to predict subthreshold voltage?
How can we quantify the performance of neuron models?
I will argue that simply putting data on the WEB without posing a well-defined question is not enough to attract modeling efforts, but we need well-formulated tasks.
