Models of Learning
Hanneke den Ouden
This part has the following subsections, it is best if you work through them in order:
Here we will look at how we can define a model that can 1) learn about values of stimuli and 2) translate these values into choices.
Before we delve into the modelling, let's have a good look at the data we acquired from the gambling task.
Before we fit a model to data, we need to get to know our model. Here we will define a basic reinforcement learning model, and have a look at how this model behaves for different parameter values. In other words, we will let the model 'play' the task and assess what sort of behaviour it can and cannot capture. Next we will estimate the optimal parameter values for the model we specified, based on the data. This step allows us to quantify the 'identifiability' of the parameters.
Modelling real data
Finally, we can look at our data! Just like in the simulated data we estimate the parameter values from the data that we collected from 'real' subjects. We will briefly touch upon looking at the parameter values, and think about the differences in behaviour from the simulated data and our real subjects.
We review the most important steps when modelling behavioural data.