Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

Yeah, [Music] okay. It's recording. Um, so for for Joseph, for Daniel, for the rest of the institute who's uh following along remotely uh or via YouTube, um welcome to the Active Inference Institute. Uh we do a uh weekly textbook group readings uh twice on Thursdays, once in the morning, once in the evening uh if you're if you're roughly in the Western Hemisphere. And um yeah, so so today we're going over chapter 7 in cohort 7. Um and we'll be talking about essentially the exemplar discrete state space models and active inference. That would be hidden markoff models and uh partially observable markoff decision-m processes. We have a nice T-ise example which is pretty classic in in the field as well as in kind of behavioral and and and and neuroscience and neur neuroiming studies in general. And uh again we're in the second half or part two of the book which is highly dedicated to to constructing uh essentially active inference models as opposed to part one which is much more focused on theory and and definitions. Um, and so it's for anyone who's interested in computational modeling, whether it be doing it themselves or getting a clearer grasp of kind of what are these computational models or cognitive models that we're putting together in order to say collaborate with a uh team of of people who do do cognitive modeling, then it it's very worthwhile to spend time with the second half of the book. And then furthermore, there are just there are many discrete state spaces in in a in a very general sense, in a in a physical sense, in a kind of decision- making and in in and human belief sense. It's referring to scenarios where essentially if we think about it in data, it's sort of like categories. It's thinking about binaries. uh it's not so much thinking about uh fully continuous number series such as a range over one through a million where we could have so many different uh different values including with the decimal points uh potentially an infinite number of values instead we're looking at say RGB color scales on old television so red blue green um we could have a distribution between the three of them in the sense of how much red how much blue and how much screen. Uh but nonetheless, it's it's a kind of um discrete uh categorization process. So, um yeah, normally I give a rather long uh kind of chapter summary, but I'm happy to just kind of open up the floor if uh Joseph, you have any particular questions or things you'd like to discuss here. Um I I would like to hear your summary. Um I also do have some questions. Um I haven't fully been able to understand um for example figure 72. So it if we could like I I I guess I just don't understand what's plotted and and where the lines are coming from. But um yeah, I I'm I'm very interested in in trying to build these models and understand how to build them and um and so this seems like a really critical chapter for for me. Yeah. Yeah. No, that's great. Um yeah, figure 72, which I I personally have not spent a lot of time with recently, so coming back to it is rather interesting. Yeah. Yeah. So the the this is where the chapter it essentially opens with hidden Markov models uh before we jump up to uh partially observable Markov decision-m processes. So it's like we have um I believe my is my screen still being shared appropriately or can you not see anything? Uh no I don't see your screen. Okay, got it. Thanks for that. Just realized I doing a little bit of uh display gymnastics. Um that should have done it. So this is and a quick reference. This is the KOD for the textbook group. And so we have this available to all members and we have u not just a overview of the textbook group and you know dedicated uh meetings and and time availability and and all those sorts of things for the separate cohorts. We also have a pretty good breakdown of the active inference textbook especially since it became open source. So, um, so yeah, figure 71, I know it's not the one you referenced, but just…