Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

Hello, thanks for all joining. It's ACTIV textbook group, cohort one, meeting 22, chapter nine on October 28th. Okay, we're in the second discussion on chapter nine. Model-based data analysis. Would anyone like to add anything they want or mention how the generative model discussion from the previous one hour relates to model-based data analysis? How does chapter four relate to chapter nine? Okay. What does chapter nine address? What do chapter six and nine address that chapter four doesn't? I mean, six is about thinking about applying it in a, like, incredibly kind of simplified, step-by-step kind of way. And nine is very specifically about answering scientific questions with it and doing that with a sort of... sort of schema of active inference. So, I mean, that's... So, I mean, that's... the through line, I guess? But, um... I'm not sure more than that what to say. No, that's really insightful. Ali? Ali? Yeah. Actually, I would compare it to a kind of learning how to solve problems in, I don't know, mathematics or physics or other areas. For example, in chapter four, we acquired some necessary knowledge or prerequisites about how to... how to think about the phenomena or the problem we're dealing with. And chapter six kind of outlines a roadmap, a roadmap about how exactly we should proceed to solve that problem in a kind of formulaic way. And chapter nine, actually, in chapter nine, we get our hands dirty and try to solve some of the actual or empirical problems with the tools or techniques we had learned in chapter four and six. So, in this case, I somehow think of chapter nine as a kind of case study. Well, a kind of case of studies or a kind of solved problem section of this textbook. And, yeah, I don't know how much that captures their nature. That's great. Wow. Great thoughts. I totally agree. Let's look at the table of contents. Chapter one, special overview chapter. Mirror symmetry, same but different with ten. So, now in the middle eight chapters. Two chapters. Two chapters on the low road and the high road. Helping to two different approaches which can be developed like in a lot more detail to approach active inference from two fascinating and non-controversial starting points. Bayesian inference from the low road, free energy principle, and the repeated measurement persistence imperative. Chapter four is the first chapter on active inference. Active inference is about the generative models that are being specified and how they're being computed. After just one chapter on active inference. It goes to the area where the most research and modeling has been done in active inference. Which is computational man alien neuroanatomy. One chapter on active inference. One chapter on active inference. One chapter on the primary domain. The first half of the book. The second half of the book. The second half of the book. The second half of the book. The second half of the book. The part where you're going to be in parallel, perhaps on your first, but perhaps not on your first reading, you're going to be playing through these applied aspects of modeling. First is the recipe for designing active inference models in chapter 6, as both of you have very nicely said. Then, the bigger 4.3 dialectic discrete continuous time is revisited in two chapters. Because it probably, it's one key difference, it's a minimum of two, so it prevents any like, oh, well, it has to be a PMDP. It's like, well, obviously not. So it keeps the space of generative modeling open, and it allows for discussion of the very interesting topic of hybrid active inference nested models with continuous actuators, for example, but discrete decision-making apparatus. So, it just is like wanting to cover these two key types of models in a bit more detail than was addressed in chapter 4. But this was the first possible moment to address it after 4.3. Just kidding. Maybe not. Then, chapter 9. Or, yeah, where we are now. Which is just, okay. You've built your pure…