Puzzle program does 10,000-piece jigsaw in a day

1 hr.

Devin Coldewey

It's not enough that our robots clean our rooms and cook our meals ? soon they'll do our puzzles for us, too. Andy Gallagher, a researcher at Cornell University, has made a program that can assemble extremely difficult puzzles with great speed. It sounds frivolous at first, but there are actually some interesting applications.

The puzzles this program ??or artificial intelligence (AI) ??attempts to solve are more difficult than your average jigsaw: Instead of that saw cutting in zigzags and wavy lines, it cuts straight, eliminating the shape-based cues we use to match pieces. Gallagher's program can take a 10,000 piece puzzle of a high-resolution photograph and solve it in 24 hours ? which, as puzzle lovers will tell you, is no mean feat.

Check out this video of the program at work:

Gallagher's program solves the puzzle by studying the edges of every piece and putting together the ones which have?colors that?match best. But with 10,000 pieces and a lot of data for each piece, that's a long process. So Gallagher's model uses a strategy based on how humans solve puzzles: Finding groups of pieces and building on them, then sticking the groups together. Using this technique brought the solving?time?down to a day, three times the speed?of earlier models.

This type of algorithmic problem solving isn't just for puzzles. Although the exact method can in fact be used to reassemble shredded documents, something the Pentagon is always keen on, there are also?lots of applications for medical research, which needs to consider and compare the contours of molecules and medications.

Devin Coldewey is a contributing writer for msnbc.com. His personal website is coldewey.cc.

lenny dykstra super tuesday epidemiology total eclipse of the heart jionni lavalle earthquake san francisco donald payne