73 Building a model for aggression
Recall in Chapter 8 that we considered the experiment of Remedios, et al. in which they monitored activity of 115 neurons while simultaneously assessing if the mouse attacked the other mouse in its arena. In looking at Figure 8.1, we might be able to connect neuronal activity to attack behavior. Your task is to develop a model that, at a given point in time, takes the calcium imaging intensity of the neurons and from these gives a probability of being in attack mode. The model will be informed by all time points in the experiment. That is, you are not producing a generative model for the neuronal activity. You are taking it as given (quantified by the intensity of the calcium imaging). You are producing a generative model for the binary variable, attack. This is another example of a decoding problem, where the goal is to decode how neuronal activity translates into behavior. (And naturally, you should do prior predictive checks.)
This is a challenging problem, and there are “standard” models (those quotes around standard are loaded, and we will talk about what it means for a model to be standard later on) for this, but I think it is best for you to come up with your own. This really helps to get you thinking about generative modeling.