Synthesizing realistic neural population activity patterns using semi-convolutional GANs


Nov 07, 2017 (modified: Nov 07, 2017) ICLR 2018 Conference Blind Submission readers: everyone Show Bibtex
  • Abstract: The ability to synthesize realistic patterns of neural activity is crucial for studying neural information processing. Here we used the Generative Adversarial Networks (GANs) framework to simulate the concerted activity of a population of neurons. We embedded a semi-convolutional network architecture in a Wasserstein-GAN to facilitate the generation of unconstrained neural population activity patterns while still benefiting from shift invariance in the temporal domain. We demonstrate that our proposed GAN, which we termed Spike-GAN, generates spike trains that match accurately the first- and second-order statistics of datasets of tens of neurons and also approximates well their higher-order statistics. We show that, when applied to a real dataset recorded from salamander retina, Spike-GAN produces population spike trains that match the second-order statistics as well as state-of-the-art maximum entropy models. In addition, Spike-GAN faithfully reproduces the temporal dynamics of the retinal population's activity, an aspect that is typically ignored by maximum entropy models and most existing methods. Finally, we show how to exploit a trained Spike-GAN to construct 'importance maps' to detect the most relevant statistical structures present in a spike train. Spike-GAN provides a powerful, easy-to-use technique for generating realistic spiking neural activity and for describing the most relevant features of the large-scale neural population recordings studied in modern systems neuroscience.
  • TL;DR: A method to generate realistic neural activity patterns using Wasserstein-GANs and a novel semi-convolutional approach
  • Keywords: GANs, Wasserstein-GANs, convolutional networks, neuroscience, spike train patterns, spike train analysis