Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

all right hello everyone this is the first meeting of cohort five of the textbook group on September 19th 2020 23 and thanks everybody for saying hi um I will share my screen and we will be looking at the cohort five live meeting page so this is kind of like the starting place for participating or joining cohort five you can always click on somebody's icon on the top right of the platform and jump to the exact page they're at if it's not clear from looking at the um sidebar so here we are in the first meeting and after today the recording will be here just like in cohort for there's recorded videos so I guess there's many places to begin but is there any question that anybody wants to ask about any aspect of the textbook group thank you how do we even start and evolve it each time and and make it fit for everybody who's doing this on their own time and who's coming at it from different angles Varun and anyone else you can raise your hand or however well uh just a quick question um does the textbook um include going through some of those uh extra notebooks some like active inference coding like active influences from scratch or some of the like programmatic implementations or does this mainly stay only on the theory great question um the textbook itself all of the code is in Matlab it's in appendix C and it is not a major part of the textbook itself however in the textbook group people are bringing resources and other references um in the page called code you'll find several dozen implementations of active inference code and we can certainly arrange sessions like just us informally or trying to have the authors or other experts join to look over some code um features if people are interested in this but it's not a necessary or central part of the textbook however because the second half of the textbook is oriented towards application especially than talking about how you go from like a sketch of a system of Interest and then use the recipe in chapter six towards actually designing the generative models so for people who are already and wanting to actually be working in code then it's there and we can work on it um s mail yeah I asked question I wanna know um clearly about one very important term supervised and the relationship with free energy that can help me for good understanding this textbook I think these two tell me is so important surprise yeah yes great question so look many questions will arise about all kinds of terms and about relationships so here's just one way that that this can be approached so I went to the questions page in the search box I'll type in Surprise now this is going to bring up a lot of questions because it's a very common term but um we could review the questions that have been asked and see if if one of them directly asks what you're looking for and if not then it's like a huge Act of service to add the question because it's something that we're all curious about and so this is how we kind of build our repertoire of asking questions and also Everyone is always welcome on their own time to go into the answers and discourse section and um develop the answer whether by adding material or by reorganizing shortly to answer your question directly surprise is about how expected or unexpected a given sensory observation is so if you were least surprised by a sensory observation you would have had like the most accurate belief and so surprise minimization is equivalent to having the model with a maximum evidence surprise in and of itself though is often very hard to calculate and so instead of calculating surprise directly attractable optimizable bound on surprise is used and that's called free energy so that's definitely one of the key pieces which is like if we had the minimum surprise model we'd have the maximum evidence model in the exact Bayesian case that would be totally fine however in the high dimensional setting it's not tractable to compute surprise and so we have this optimizable heuristic with free…