I liked the way you formulated different scales in systems mathematically!
I am reading donella meadows and she describes systems as elements + interconnections + purpose; and also the notion of stock, in-flow and out-flow.
Here, you considered training data and errors both as input to network but what if we consider training data as coming in, and errors as outflow. Using errors we drain the unnecessary entropy. Training data brings in lots of information and weights acts as stock(kind of buffer) till some future error decides to let go of that information to replace with something more general.
I liked the way you formulated different scales in systems mathematically!
I am reading donella meadows and she describes systems as elements + interconnections + purpose; and also the notion of stock, in-flow and out-flow.
Here, you considered training data and errors both as input to network but what if we consider training data as coming in, and errors as outflow. Using errors we drain the unnecessary entropy. Training data brings in lots of information and weights acts as stock(kind of buffer) till some future error decides to let go of that information to replace with something more general.