Scene Gist

Computers are able to perform more human-like function all the time as technology improves. Although I think (or is it hope?) Chobits and Blade-Runner are a long way off; surprising new technologies continue to appear with great frequency.

For this week's Blue Monday, I decided to delve into the computer world to interview Geoff Basore and Zach Maier on their biologically inspired computer research. Unfortunately, they were not yet of age so my preferred “scientific” atmosphere of the local Garden Lounge was out of the question, but I did catch the boys on lunch for a photo opportunity:



Geoff (left) and Zach (right) have been working on a program that can identify a scene in milliseconds. The idea for this program was drawn from the human visual system, which can get the "gist" of a scene in a tenth of a second. For example, when flashed a picture of a beach, forest, or portrait a person can usually give the image a category even if it was only displayed for 1/10th of a second (although rapid succession of images greatly increases error).

The program they are working on will classify entire images using color channel values as well as Fourier transforms, which detect the frequencies in a scene giving the program insight on the orientation and diversity in the image.

"Other aspects we hope to explore include contours, shapes, and textures," explains Geoff, who is looking into broadening the abilities of the program.

The program even derives some of its training techniques from biological systems. Particles in the system act like swarms of bees, moving toward better solution to identify a scene. The particles are trained in teams and are randomly swapped at points in training, like chromosomes are swapped during gamete formation and regrouped at fertilization.

"We're not modeling biology," says Zach, "rather, we draw inspiration from it."

With biology as their muse, Zach and Geoff already have their system running at 30% accuracy after only a month and a half on the project.

Geoff is looking forward to the military application of scene gist. Of course, the first thought that came to my mind when they explained this new technology to me was how much better google image search could be with "scene gist" turned on. No more bountiful bosoms or pop stars coming up when I'm trying to search for an image to include with my research (or better yet for my precious blog!). As far as Zach is concerned though, the applications are "pretty much limitless."

The Drink: Even though they're not yet of age, Zach and Geoff wanted to feature pina colada as their drink (the guys will have to have theirs sans rum).
1/3 rum 1/3 coconut milk and 1/3 crushed pineapples