Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

okay it's July 29th 24 and we're in the first of the applying active inference sessions for the third week of the cohort 6 and Andrew is going to give some overviews and share from his recent tutorial so go for it perfect um thanks Daniel and um as I mentioned to the the other session attendees today um I did have a bit of a full work day so not fully prep to to run through all this but um I am happy to present what I have and just kind of bear with me a little bit um I presented this tutorial um about a week and a half ago July 17th at University of Pennsylvania for the 10th International Conference on computational social science um wonderful experience doing that by the way super cool to see some other folks doing active inference research uh at that conference um and uh super great to to meet some folks there in including Brennan kleene who I believe has had some degree of affiliation with the institute in the past um and he is also a pi mdp uh package developer one of many uh but I got into contact with them while I was developing this tutorial so it's just excellent having him in the room um it seems I did their pack package some amount of Justice uh in putting together this tutorial so um the intention of this tutorial was multifold uh one uh I myself is am an internet The Institute and so I'm still learning a lot um that said I've been involved long enough to where I've been itching to uh learn much more about coding uh more directly as opposed to reading the the text book inside and out as I have been for for at least a year now I've been involved with the textbook group in facilitating meetings um so so a learning experience for me um a teaching experience in the sense of presenting it at a conference as well as hopefully developing it as a resource for uh other folks at The Institute who might be coming at active inference from you know maybe an angle that they don't have a lot of Prior experience with say reinforcement learning the logic of which can is is very relatable to to active inference and and building simulations um that way and uh and finally the the theme of the conference was social science and so I uh wanted to develop a multi-agent model and I spent a little bit of time looking at at kind of popular uh agent-based modeling paradigms uh in that that people still kind of use and study in social sciences today uh so agent-based modeling being um again this is one of those moments where I'm GNA be kind of adaptively figuring out how to go about this um but um yeah let's let's jump to the slides maybe so there we are um yeah I want to develop a multi- agent simulation that basically recreates a popular social science Paradigm uh in this case it's creating in number of Agents however you prefer to set it um it it it groups them together it connects them uh via like Network logic um the code I wrote has two different options you can either make say in uh groups of M agents so you could have like four groups of five agents uh or otherwise however you want to set that um as I was discussing briefly uh with with one of the other attendees today um I work in education so I particularly like this Paradigm because uh I think about students in our classrooms at the the schools that I work for and the idea of like okay what's what's a good way of facilitating them learning learning to cooperate with one another uh but at the same time not not get too dependent upon relying upon each other for like you know putting together a group project or something that is it that they should be able to learn in such a way that they develop both a p a sense of like personal efficacy but also being able to work within a group um and so the the Paradigm has to do with something like Collective problem solving where where you you have agents and and networks they're collectively solving a problem uh the problem is uh based in um what's called an ink landscape and uh it's basically just um just basically agents seek to find the best…