Dynamic Programming Chips

As anyone who has dabbled in computer science can no doubt tell you, the best route to solve any large-scale problems correctly is to break it down into smaller chunks. Turns out that this technique is called dynamic programming. It makes for very efficient problem solving not just in coding, but also in fields ranging from genomic analysis to economics to physics. Unfortunately, however, to adapt dynamic programming to computer chips with multiple “cores,” or processing units, your average genomic analyst or economist would have to be practically an expert programmer, too.

To make it so more people will be able to utilize dynamic programming chips, researchers at MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) and Stony Brook University have been working on a new system, Bellmania. The system lets users decide what they want a program to do in broad terms--the kind any non-programmer who doesn't need to worry about how particular computers are would. Bellmania then automatically creates versions of these new programs that are optimized to run on multicore chips.

According to MIT News, in order to test out Bellmania, the researchers "parallelized" several algorithms that used dynamic programming. In keeping with the whole point of dynamic programming, they split up the algorithms into smaller chunks so that they would run on multicore chips. The new programs were between three and eleven times as fast as those produced through earlier parallelization techniques. They were also, on average, just as effective as those manually-parallelized by computer scientists. So--voila! Expert researchers whose endeavors would be aided by dynamic programs no longer need to trouble themselves about becoming experts in another field altogether.

This is pretty interesting, I think. I'm glad that some of the best minds in computer science are working toward letting the best technology be easily-accessible to the best minds in other fields. Where would we be without cooperation?

Sources

• "Faster programs, easier programming"--MIT News
• Computer chip picture