Jakob Macke (TU München, Department of Electrical and Computer Engineering, Neuroengineering) will speak about
“Bayesian inference on mechanistic models of neural dynamics.”
One of the central goals of computational neuroscience is to understand the dynamics of single neurons and neural ensembles. While advances in experimental techniques are making it possible to measure neural activity at unprecedented scale and resolution, most analysis methods for neural data are based on generic statistical models, for example GLMs or maximum entropy models. While undoubtably useful, these models do not incorporate knowledge about underlying mechanisms, and interpreting their parameters can therefore be challenging. How can we make statistical inference tractable for mechanistic models of neural dynamics?
I will present a general method for Bayesian inference on mechanistic models of neural dynamics, building on recent advances in likelihood-free inference. Our approach can be applied in a ‘black box’ manner to a wide range of neural models without requiring model-specific modifications. In particular, it extends to models without explicit likelihoods (e.g. spiking networks): The key idea is to adaptively simulate multiple data-sets from different parameters, and then to train a probabilistic neural network which approximates the mapping from data to posterior. Our approach will enable neuroscientists to perform Bayesian inference on complex models of neural dynamics without having to design model-specific algorithms, shrinking the gap between theory-driven and data-driven approaches to neural dynamics.