Textbook Group meeting for Parr, Pezzulo, Friston 2022 .

all right welcome back um we're in the second discussion on chapter 3 so does anyone just have any quote or question or anything to kind of begin with any thoughts or we can just start to look at the the written questions from before or some page or figure from three a a basic sort of disorienting thing uh I guess in the uh diagram at the top of page 44 um where it shows external States internal States action States and sensory States I'm confused uh about what are realistic examples of like influence on that direct Arrow between active States and sensory States like again I understand active State or the point of active State generally is to impact upon the world and like eventually we see the effects of that in sensory States but I guess the direct Double Arrow between the active States and sensory States seems confusing to me yeah it doesn't have to have any influence in the real system but it's it it it could but I think in a if we if the sensors and actuators of the system are separate then that edge may have no effect but in principle it exists because the the the states are not like pre-tagged it's always just based upon the choice the internal state so a causal edge here in sketching out the specific system You' quickly determine like whether it's something causal in the system or not but we'll see if it comes up in any of the examples Susan uh yes um so can you I'm really unclear in what the what are the boundaries between um the um inference the the external and internal States in other words specifically what is action in this context and I I'm wrestling with speech acts um you know not doing something is still an action on your environment if you think about it so yes well one aspect of action that is um not really looked at in the textbook is deciding when to act like in the kind of simplest format of the action perception Loop there's just a perception and an action every single time step so the like the choice to like wait versus to do something at all um sometimes that's brought in as like a null action okay yeah just like an action that kind of like keeps you where you are okay there could be a nested model where the top is a binary decision you know make a move or not within make a move there's just the four moves um okay is is that have anything to do with the um transition function the um the the transition function is describing how the state the hidden States change through time so it's the B Matrix we're getting an observations okay and then the a kind of maps the observations to yeah the hidden States so from the thermometer readings to the temperature in the room and then the transition function is the temperature of the room changing per delta T just per time interval the temperature in the room underlying changing and the formulas on the right there you know that's actually for the transition functions yeah so then what okay three here yep which is B it's saying it's the distribution of the state at the next time point okay conditioned upon the state at the current time point and the policy so say okay the temperature of the room at the next time point is conditioned upon the current temperature in the room and whether I'm turning on the air conditioner or not okay okay and then which policy is like selecting which slice of b gets applied as the transition function and one last question um so is this is this model just saying that there was a move made and there was a and that's a SE is that is that you know as I'm seeing two different is that perception action or is that just a different action repeating in the picture you know just looking at the yeah like the fact that there's two threes here no actually because you got the upper and the lower so is that the higher high road low road no um so 4.3 top graph is the discrete time model and the lower graph is the continuous time model okay and then in um here's just the discrete time model okay just the top half of figure 4.3 okay so here again we…