 Computer Scientist Gets to the Bottom of Financial Scandal
started 4/1/2002; 4:07:29 AM - last post 4/12/2002; 10:47:08 AM
|
|
Ehud Lamm - Computer Scientist Gets to the Bottom of Financial Scandal 
4/1/2002; 4:07:29 AM (reads: 2796, responses: 5)
|
|
|
|
Brent Fulgham - Re: Computer Scientist Gets to the Bottom of Financial Scandal
4/1/2002; 12:38:55 PM (reads: 1309, responses: 0)
|
|
|
Doh! Got me.
|
|
Oleg - Re: Computer Scientist Gets to the Bottom of Financial Scandal
4/2/2002; 10:44:29 AM (reads: 1250, responses: 1)
|
|
Everything in this article is true or plausible, except this one phrase:
Instead of generating a value, the program returned `bottom'.
If we had such a system that can return bottom, we can perhaps get it to find a constructive proof of the Banach-Tarski paradox and thus double our salary.
|
|
Ehud Lamm - Re: Computer Scientist Gets to the Bottom of Financial Scandal
4/3/2002; 4:34:52 AM (reads: 1278, responses: 0)
|
|
|
I am not sure how the system is suppsoed to work, but theorem proving capabilties can allow you in some cases to reach such a conclusion.
|
|
Paul Snively - Re: Computer Scientist Gets to the Bottom of Financial Scandal
4/12/2002; 10:47:08 AM (reads: 1163, responses: 0)
|
|
|
I haven't looked at Peyton-Jones' combinators yet, and I might not have sufficient theoretical information to judge even were I to look, but it seems perfectly plausible to me that the particular set of combinators in question can make two desirable claims:
- The combinators represent a sufficient model of a sufficient variety of financial instruments to have wide applicability.
- The combinators have the property that, given the execution model of their environment (Haskell, in this case), they are guaranteed to terminate.
Again, I'm not even a Haskell programmer, so I could easily be dead wrong, but these both seem like perfectly plausible claims to me, and the assertion that a Haskell program using the combinators evaluated some combination as "bottom" just tells me that a piece of software detected cycles in the graph, which we know software can do. The fact that the graph is represented as function calls is incidental to the cycle detection. And Peyton-Jones' other writing on the combinators makes quite clear that lazy evaluation is critical to the process, as I would expect in a system in which the value of f(g(x)) may depend upon the value of g(f(g(x))). :-)
Am I naïve?
|
|
|
|