Remember all the time you spent as a kid trying to find Waldo (or Wally, if you’re outside the U.S.) in the Where’s Waldo books? Well, like almost everything else these days, computers have turned us humans into chumps when it comes to that activity. We can now thank a doctoral student in computing for developing an algorithm that optimizes the search process and identifies the best places on the page to find the striped shirt and glasses-wearing character.
Randal Olson used a snowed in weekend to apply machine learning to the process of finding Waldo. He began by gathering all 68 locations from the seven main books in the Where’s Waldo series and finding commonalities. First it was determined that Waldo is almost never in the top-left corner, almost never along the edges of the page, and never in the bottom-right area.
It was found that the best place to start is the lower half of the left page. From there, Olson used an iterative genetic algorithm, which continues to make slightly different attempts until finding a better solution, and ended up with a graph that provides the most efficient search paths.
Every so often Waldo may be located in the very last spot of the page that you look, but the graph offers the fastest way to find him. As Olson puts it, “if we followed this path exactly, we’d most likely find Waldo much faster than someone following a more basic technique.” That makes it a great tool to have when it comes to challenging your friends to a Where’s Waldo race.