I'd like to get a discussion focused on this topic started in the MPC +
NuPic thread, as my head's been spinning along the same circles regarding a
project involving game worlds and agents in the world.

>From the previous thread via Ian:

So basically you have a representation of a goal that doesn't change, but
> how you coordinate your muscles during the movement has to be relearned to
> get to that goal. Some cells will always be in charge of dynamically
> reacting to the world, and others will learn about these new world dynamics
> and remember them for next time.


> To properly implement this in CLA I think there needs to be a goal SDR, a
> series of predictions given the current input and the goal, and a mechanism
> by which the goal and the predictions are compared moment to moment to the
> input from muscles and the world. Moment to moment corrections would
> probably be random at first, then gross movements reinforced for reducing
> the error between input and prediction, and finally refined.


> One aspect I *really* don't have a good answer for though is how does goal
> + world state == action? Is it the intensity of the goal representation? Is
> it a goal plus some global trigger that says "go?"


>
The way I've started to plan is to construct ASMs from the cambryan
intelligence model that actually make the motor commands. The inputs to
these would be the prediction and sensor results from models. The machines
would then compute to favor some logical behavior based on the predictions
and return onto the same bus the sensors do, which the model feeds from on
each each step.

Goal would be some scaler value that would change somehow.. based on world
state and the agent's current state. An ASM somewhere down the line will
signal on the "do what I say!"  interrupt bus for the ASMs based on the
scaler value and what the model is predicting..

That's as far as I've gotten in my head space.
_______________________________________________
nupic mailing list
[email protected]
http://lists.numenta.org/mailman/listinfo/nupic_lists.numenta.org

Reply via email to