Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

all right welcome thank you everyone it's August 15 23 and word cohort 4 in our first discussion of chapter five so soon we will screen share but first any remarks that anybody wants to make on chapter five just anything that stayed with them after reading it or feelings that they had or anything else about chapter five I will say uh I found this to convert and conversely to chapter four being rather mathematically against this one is a little bit more of neurobiologically dense at least for those who maybe like are new to um thinking about the relationship between this and areas of the brain but I did appreciate in this chapter there was much more on Precision which we didn't get too much before um even though it was brought up many a Time preceding chapters and I really appreciated the relationship between uh that was made between different forms of precision and different neurotransmitters neuromodulation we found that to be very interesting and helpful and maybe kind of reinforcing the idea of like they're not just probability distributions here but also uh Precision of each or like confidence in each probability distributions so my big takeaway from this chapter cool thank you anyone else is there any thoughts on five any other thoughts or any any question that somebody wants to lead with or we can just explore the chapter and look at some written questions um I did I did want to ask uh what everyone makes of uh the relationship made here I mean in preceding chapters we're talking a lot about um you know kind of the differences between viewing variables as categorical versus continuous and it's a big kind of bifurcation throughout many of these chapters like we have continuous models versus um you know Palm DPS and um just different ways of looking at things but it at one point in this chapter they point to the idea that um there's actually a big relationship a connection between the two that um and maybe what's going on in the upper levels of say the the brain cortical regions and so on is that they're working with categorical variables but in a way it's not simply categorical variables but rather these are like um looking at discrete trajectories that the continuous variables could take that was a nice way of maybe putting making some kind of thread between the two ways of looking variables that is there's still a continuous aspect within the categorical States um it's just it's looking at it as a discrete trajectory I didn't know if anyone made anything of that found it to be useful for anything or anything they're working oh thanks um Darius than anyone else yeah it's not so much a comment on the previous Point um about discrete versus continuous variables I wanted to ask really a question more uh more concerned with the mechanisms behind prediction coding at least as it's embodied in the brain um so you I gotta understand this idea of sort of the shuttling up of prediction errors and the shuttling down of expectations I guess Daniel my question is we kind of just take this idea that prediction errors arise for granted um something I've been trying to kind of wrap my head around is to what degree is that a kind of just um immediate mismatch between the priors at one layer and the incoming sense data or is it that those there's some Bayesian inference required to even compute the prediction error so you know I have a prior x i compute variable base to work out the probability of my prior given y is that in a sense that distinction between the prior and the posterior is that what is constituting surprise or is it just a pure sense data coming up the chain which then is discordant with my prior X if that if any of that makes sense yeah what a lot of uh a lot of angles of this just first very briefly yeah this uh hierarchical hybrid type of model where the higher levels of the system are discretized and then the lower levels are kind of like where the rubber hits the road is being more continuous that's a motif that…