John Hertz (NORDITA, Stockholm and Niels Bohr Institute, Copenhagen)

Abstract:
An important current problem in neuroscience is inferring the connectivity in neuronal networks from recordings of spike trains of many neurons. To do this, we choose a (stochastic) dynamical model for the neuronal network and find the connection strengths that maximize the likelihood that this model could produce the measured data. In this talk I will describe and show some examples of how to do this, both exactly (but often very slowly) and approximately (but much faster). I will pay particular attention to (1) dealing with nonstationarity in the data and (2) how to take into account the many unrecorded neurons in the region where the spike trains are recorded.