Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

all right it's February 5th 2024 we're in the first discussion on chapter 6 for cohort 5 so Andrew thank you for facilitating everyone feel free to raise your hand and everything and Andrew go for it sure so chapter six we're beginning part two of the textbook um this is basically yeah it's describing a recipe for Designing active inference models we get as we know from part one uh specifically chapter 4 especially we've already been introduced to partially observable markof decision making processes as a discret Time version of um kind of one of the the sort of Exemplar active inference models that's used um in a variety of Behavioral experiments as well as the continuous models um in continuous time and so this is kind of just getting into more of like how do we actually begin like given in part one we've been introduced to a lot of the main principles and kind of the framework of active inference now we're being introduced to how to actually construct the models ourselves um so let's see yep um yeah the first part we have the introduction uh crucial defined like when designing an active inference model you want to specify your system of interest um for me personally it was easy coming into active inference have big ideas of what to apply active inference to um because of course it can be applied to a variety of living organis organisms or other systems but um really whenever you want to design your own project you do need to specify a system of Interest what is the problem um some other thoughts are like what are the expectations you have of what kind of Behavioral um like emergent behaviors you might see or expect to see versus what ends up happening whenever you put your model together um so really uh one one take on it is to really just start from basic principles like we have part one of the book we know much more about something like how active inference works now start to use that as a tool for yeah kind of shaping your your project or your experiment you want to put together um so you kind of simplify and grow outwards from [Music] there um section 6.2 there's a very nice brief like um designing an active inference model in four steps again and number one immediately it's which system are we are we modeling uh in the context of like statistical marov blankets like how do we Define the the generative model versus the generative process like what what is what is the agent you're trying to model and what what defines it and how do you distinguish it from the environment or the process that's occurring around it in which it is operating um what is the most appropriate form of the model are you looking at for example some you know experiment that invol uh involves I decades um like continuous movements Arc reflexes those kinds of things in which case you might be um going for more of a continuous time model or are you looking at like are you attempting to um model uh an animal or a human uh operating in its environment in which case it might be um it might be making discrete choices um in which case you're looking at more of a a discrete time model or a more than likely a PDP um if the agent's going to be choosing its actions and has a Time Horizon and you know you're including things like expected free energy um not just variational free energy then step three how to set up the generative model choosing what are the appropriate variables where the priors it's whenever you kind of get into the you know the the meat of the of the modeling process like what are all the specifications here we need to make in order to simulate your experiment that you're going for and number four again the same thing but now how do we set up the generative process because in the same way that you have to set up the model you also have to set up the the process in which it's it's operating kind of intera AC in with so 6.3 deeper discussion uh what defines a model what defines the process where do we draw the line um what are the the in…