Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

when um man that's distracting distracting okay I just won't share screen it's what's causing it where I was getting at was okay so on the one hand we we have this discretization and this continuity and on the other hand uh we can we have this embedding of the discreetness within the continuity or the continuity uh within the discreetness um or there's like nestedness right where you like you have like discreetness continuity continuity within the continuity and so on and so forth but also uh while we have the prioritization of what's Happening there's also kind of like a tuning aspect right like a tune like there's tuning me mechanics happening right like when you think of stuff like hyperplanes and or hyper parameterization um I think a beautiful paper to look at um is biological cognition by uh I think is hubner and schulen and um they bring up this term called heterarchical processing right and heterarchical has to do with like selective tuning or you think of like a like you're an audio engineer and you're mixing you have a you have a mixing board and you're trying to get the right trying to mix the channels and all that together the right way like this tuning or this adaptive tuning rather but it's like there's like a mix of there's layers of the hierarchical and then n tuning happening right so so it's not even though I would argue it's mostly hierarchical you still have this tuning happening right you still have this sort of like this degrees of tuning happening but I'm not sure if that I like I'm I already am like about I'm on chapter five right now TR it's chapter four but I'm not sure if the book really addresses like this hyper parameterization issue but I guess you can kind of go over that if you can yeah yeah yeah that's um I think that's a really interesting point too so whenever I don't want to find the the figure I don't have it top of mine but it's I'm it's in the second half of the book and it might be in the section of learning but what are your what everything you're saying about like tuning and hyperparameters th those are absolutely like those are compon like directly components not just figuratively and in active inference models and so what what you can have is like like I mentioned Precision earlier that came up at the end of the the chapter this chapter um but you can actually like Precision can be modulated such that it impacts the rest of the model once you have a hierarchical model it's like it's not just you know every once in a while something happens here but most of the time it's over here it's like the whole model is integrated right and so you do have this really strong constant interplay um between the different layers um so so that that that hopefully at least figure figuratively gives you some sense of like the the way you describe like a continuous and the discret the discreet and the continuous it's like well as far as like coding a simulation you do have to more clearly Define one versus the other at least at each layer um but then that said they're so tightly integrated it's as if you know the discret is impacting the The Continuous in given moment and vice versa so that's there and then somewhere in the in the in the textbook there's um there's another figure showing the PDP we were looking at earlier but it also adds on hyperparameters and those very hyperparameters are what can be adjusted or tuned and they relate to Precision uh for example you could have agent learn its a matrix meaning the the likelihood Matrix which figures in this chapter and so they might learn to you know let's say we we have our Mouse in the t- Maze the agent right let's say they repeatedly go through the maze over and over and through trial and error um end up learning that the stimulus is always on the left let's say we're we're the experimenters we to choose that let's say the stimulus is is always on the left for an agent who's who's who who who undertakes or engages in learning um it's not only going…