rose.blog

I thought this was quite interesting. It’s the slides from the 1998 Turing lecture, given by Jim Gray talking about what enabling technologies would have the most value for IT. Remarkably prescient given the date of the talk and how fast things change.

Here is a particularly interesting screen grab from p.47:

Gray sees such a system “discussing” the system with the designer…and mentions that this is a type of Turing test–the automated programming system is imitating a programmer.

I wonder how close to this vision a good declarative language would be?

What I’m going to try to do is list a bunch of applications that require the solutions of NP-complete or NP-hard problems. These will be found under the Applications tab in the sidebar.

I’m going to try to restrict the list to applications that have the following properties:

1. Problems people really want to solve are too big for exact solvers
2. People are not happy with the level of approximate solutions they can obtain with the best known heuristics
3. The people in 1. and/or 2. would see significant value in being able to solve bigger problems, faster and would make or save a lot of money if this were possible; I’ll try to include only if customers might be willing to pay in excess of $1B/year for satisfactory solution of their problems

Note I’m not saying these problems ACTUALLY have $1B markets, I’m saying that it might be possible. If anyone out there in blogland has actual market numbers to any of these problems that would be very useful.

Any suggestions for problems like this would be greatly appreciated (existing apps or new ones, doesn’t matter); any info on desired customer solution characteristics (size, accuracy, speed, etc.) would be great also.

The protein folding problem is this: given the primary structure of a protein (its linear sequence of amino acids), compute its three dimensional shape. Here is a good introduction.

One approximation that is often made to simplify the problem is to restrict the aminos to a lattice. A further restriction is to label all aminos as either H (hydrophobic) or P (hydrophilic).

Restricting to a lattice and H & P aminos gives the HP model, invented by Ken Dill. Finding the global energy minimum of this model of protein folding is NP-complete.