Alexandre’s poster submission “Inferring mesoscopic population models from population spike trains” has been accepted at COSYNE this year. There were 704 submissions to COSYNE this year, of which only 56% were accepted.
How does the interplay of single-neuron dynamics and neural connectivity give rise to the rich dynamical properties of neural populations? To tackle this question, it is desirable to have models which exhibit a wide range of population dynamics but remain interpretable in terms of connectivity and single-neuron dynamics. However, many commonly-used statistical models of neural population dynamics are based on generic models of dynamics (e.g. in Macke et al. 2011). Conversely, it has been challenging to link mechanistic spiking network models to empirical population data. To close this gap, we propose to model such data using mechanistic, but low-dimensional and hence statistically tractable models. We approximate neural populations as being composed of multiple homogeneous `pools’ of neurons, and model the dynamics of the aggregate population activity within each pool. We derive the likelihood of parameters (both single-neuron parameters and inter-pool connectivity) given this activity, which can then be used to either optimize parameters by gradient ascent on the log-likelihood, or to perform Bayesian inference using Markov Chain Monte Carlo (MCMC) sampling.
We illustrate this approach on a model based on generalized integrate-and-fire neurons (Schwalger et al., 2017). Using micro- and mesoscopic simulations of multiple neuron pools, we demonstrate that both single-neuron properties (membrane and adaptation constants) and connectivity-parameters (excitatory vs inhibitory connections and connection strengths) can be recovered on simulated data. Moving beyond point estimates, we compute the Bayesian posterior for combinations of parameters using MCMC sampling. Finally, we investigate how the approximations inherent to a mesoscopic population model impact the accuracy of the inferred single-neuron parameters. Ultimately, our method ensures compatibility between experimental multi-population data and mesoscopic dynamical models, by providing methods for statistical inference of low-dimensional mesoscopic models.