Doctoral Research Seminar on: "Adversarial Training of Neural Encoding Models"


Doctoral Research Seminar on "Adversarial Training of Neural Encoding Models", by Poornima Ramesh Computational Neuroengineering, TUM (Prof. Macke)

Abstract: Neural population responses to sensory stimuli can exhibit both nonlinear stimulus-dependence and richly structured shared variability. We show how generative adversarial networks (GANs) conditioned on the input stimulus, can be used to optimize neural encoding models to capture both deterministic and stochastic components of neural population data. We illustrate our approach on simulated data and population recordings from primary visual cortex. We show that adding latent noise-sources to a convolutional neural network (CNN) yields a model which captures both the stimulus-dependence and noise correlations of the population activity.

3 July 2019, 16:30h, ICS Karlstraße 45, 6. Floor, room 6009