Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

hello everyone it's October 7th 2022 we are in meeting 19 for cohort one and we're discussing chapter eight so there's a bunch of questions and also a lot of things to talk about let us just begin with looking at chapter eight so does anybody want to uh give any first thoughts on active inference and continuous time everything flows nothing stands still Etc yes um I not on that specifically okay but I will have something to say about this chapter I I worked on this topic um during my my graduate time so what should we get to about 8.3 I'll weigh in excellent um any comments like discrete continuous time I know this is something we'll touch on again and again it's kind of like another low road High Road duality with continuous Indiscreet time and I think the textbook tries to give equal time and highlight the hybridization of these models in a way where realistically in the literature most are pure discrete models and continuous time models um have been almost like domain bound certain areas like motor reflexes where there's more of a tradition of continuous models are heavily represented in them whereas decision making models tend to use the discrete formalisms and that difference between the sort of like continuous time traditionally motor Associated and discrete categorical traditionally decision oriented models was addressed in the um folk psychology paper and live stream with Ryan Smith at all okay um also I just started kind of an overview got up through eight three of just understanding and just trying to lay out chapter eight so for any chapter everyone's always welcome [Music] to just make summaries and and add their own notes so we can look at the questions many of which are on the earlier part of the chapter and so maybe we can look at the earlier part of the chapter more this week and of course then have later time uh okay so 8.1 it's a single paragraph it's a complement to chapter seven which was a discrete time approach towards generative model construction the focus here is on continuous State space models which are well suited for physical fluctuations sensor receptors and continuous motion of effectors they start with the case of movement control 8.2 so let's get familiar with the notation the states hidden states are going to be X the data the observables are going to be y and then how States evolve over time depends on a static variable V so that could be a hidden or slower changing cause or factor in the environment then we have the Omega terms associated with each of the data and the hidden States change Through Time and that variability um is like the noise and this is like the signal or the flow so it's kind of the lengthen decomposition or language approach towards dynamical systems modeling um one interesting thing to note is first that this is hugely foresaged in um SPM then um also that the data are a function of the states at that time whereas the equation for the hidden States is a change in Hidden States x dot note that action is absent from equation 8.1 this is because action is part of the generative process not the generative model thought that was kind of an interesting um question what does that mean that action is part of the generative process and not the generative model surely action selection is part of the generative model the cognitive process of policy inference active States are part of the particular States active states are blanket States particular states are the generative model the generative process is referring to the external States the hidden States so is action part of both the generative process and the generative model because it kind of sits in between intermediates or is this referring to the consequences the generative model only deals with those variables directly influenced by States external to a Markov blanket so is that saying that the generative model is kind of like the complement of the autonomous States action internal States or is this even more extreme it's…