I remember John saying something like this at CDC 2012 in Hawaii, over drinks before the banquet:
“That first sentence of Anna Karenina — ‘Happy families are all alike; every unhappy family is unhappy in its own way.’ I finally realized that it’s a statement about system architecture!”
One example of "architecture" that comes to mind, different from the examples you give, is the structure of weakly coupled MDPs that comes up in operations research, more commonly used in old-school approximate DP. To quote from Brown&Zhang who have some recent results on tightness, weakly coupled structure arises where "a set of linking constraints in each period couple an otherwise independent collection of subproblems." (the paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3849573)
Typical relaxations appeal to Lagrangian relaxation of the linking constraint, to then focus on local optimization within subproblems. This ends up being a nice model for local planning with global resource constraints.
Interesting to think about in relation to broader notions of architecture in the control of complex systems!
One layer developed by the human society to handle the problem of sequential decision making under uncertainty seems to be the market mechanism, particularly the derivative market. The price, theoretically based on Black Shores formula, carries the information of risk, uncertainty, and expected return.
I remember John saying something like this at CDC 2012 in Hawaii, over drinks before the banquet:
“That first sentence of Anna Karenina — ‘Happy families are all alike; every unhappy family is unhappy in its own way.’ I finally realized that it’s a statement about system architecture!”
Like the idea of architecture a lot!
One example of "architecture" that comes to mind, different from the examples you give, is the structure of weakly coupled MDPs that comes up in operations research, more commonly used in old-school approximate DP. To quote from Brown&Zhang who have some recent results on tightness, weakly coupled structure arises where "a set of linking constraints in each period couple an otherwise independent collection of subproblems." (the paper: https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3849573)
Typical relaxations appeal to Lagrangian relaxation of the linking constraint, to then focus on local optimization within subproblems. This ends up being a nice model for local planning with global resource constraints.
Interesting to think about in relation to broader notions of architecture in the control of complex systems!
One layer developed by the human society to handle the problem of sequential decision making under uncertainty seems to be the market mechanism, particularly the derivative market. The price, theoretically based on Black Shores formula, carries the information of risk, uncertainty, and expected return.