The most likely value of the latent variable naturally defines two global population states.

2017-09-19T17:24:52Z (GMT) by Jan Humplik Gašper Tkačik
<p><b>A)</b> For every repeat of the stimulus and for every time bin we estimate the most likely value of the latent (<i>h</i>*) given the population response at that time, as well as the total number of spiking neurons in that response (<i>K</i>). The plot shows the trajectories of <i>h</i>* and <i>K</i> averaged across repeats. Error bars correspond to 1 SD. <b>B)</b> Probability density of <i>h</i>*, i.e. the most likely value of the latent given the population response. <b>C)</b> A scatter plot of the total network activity vs. the most likely value of the latent. <i>h</i>* naturally divides the population responses into two clusters. <b>D)</b> Probability distribution of the total network activity given this global population state. While the most likely value of <i>K</i> for low <i>h</i>* is zero, the distribution has a tail that extends to <i>K</i> ≈ 5.</p>