Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

all right welcome back everyone it is October 17th and we're in our second discussion on chapter two so let's head over there and um first does anybody interesting little colors were added does anyone want to give any opening thoughts or Reflections on chapter two or just bring up some question so I could share briefly comparison to the previous sections uh it took me longer for sure and I had to do more uh search for more reference material because I think it's the first section that's really trying to introduce the the mathematics um of the framework and uh the the first Parts where we're discussing uh very free energy which is only taking into account information about the past and the present to um to compute uh free energy um this was relatively straightforward um I think the section on expected free energy is a more significant leap and I think I need to spend more time I'm um uh analyzing that and and really making sense of of that part of of the chapter awesome thank you anyone else want again just any part of two that stuck with him or a question they have or a a thought on the role of the low rad um I will just uh Echo the sentiment that um I I think I I finally you know um got the handle on free energy as a concept but the expected free energy is still completely impenetrable making that that that um uh cognitive leap is is is is still a little bit beyond beyond my Reach cool we can definitely look over it okay at the at the uh at the very small font overview we have perception as inference so it's still passive is even though we can discuss there are some ways of action is related to even what might be seen as passive inference background on basian statistics example on the background on basian statistics which we can look at again but there's an object in a person's hand and they're updating their beliefs about what they think that object is based upon what they observe it might be an apple it might be a frog and then they see it jump and they update their belief from thinking that it's probably an apple to probably a frog um that case is carried through in the exact basian setting and for smaller State spaces and smaller scale computational problems you can utilize exact base however for larger State spaces you can't always do exact Bay and so that requires an approximation method and variational free energy or the evidence lower bound is a common approximation method for doing variational basy inference um doing approximate basian inference is then mobilized in this particular partition cybernetic setting which leads to a discussion of action still we're in the um cybernetic setting continuing to discuss like perception and action as part of this unified approximation challenge then we get to the real equations equation 2.5 and 2.6 are like the two key equations in this chapter 2.5 is the variational free energy so here is where we have um the ability to build on people's work previously and also make massive massive um contributions with our studying here in 2. five previous cohorts have used the ontology to provide natural language descriptions of the equation so if this just looks like a lot of symbology then it may be helpful to read the ontology description of it 2.5 variational free energy is about beliefs and data so variational free energy is like the real time um sense making Vibe check it's very closely related to how surprising incoming observations are given beliefs some more discussion on how the kale diversions comes into play and then it goes from consideration of the real-time variational free energy into the future oriented expected free energy and planning as inference with equation 2.6 but that's not something that all cognitive systems necessarily do not all systems engage in explicit planning and then there's some discussions about special cases of expected free energy and that's the end of the low road so chapter 2 pretty much starts with a simple exact Baye setting slides over to the…