“Close enough,” says one hippie CPU

When smart folks get together, there’s no telling what crazy ideas they’ll come up with. Inventing a calculator which gets math wrong, for example. Like a hippie. Like a math hippie.

Deb Roy, a researcher at MIT’s Media Lab, outfitted his home with 11 video cameras and 14 microphones to do research on speech acquisition of human children. He wanted to monitor his kid at all times to, for example, collect data on how many times the baby hears its caregivers say “mama,” and how often it sees its mother—hopefully to draw conclusions on how the baby associates the two and eventually learns the word.

Herein lies a problem: monitor something at all times while it takes the time necessary to begin learning a language and you’re going to end up with a lot of data. Say, something like 200,000 hours of video. Okay, step two: go through it and analyze the data.

Obviously this is a job for a computer; it’s a relatively mundane task which is very tedious. Jobs like this are arguably what computers were made to do. But video files are notorious for being very big, and somewhat difficult for even computers to process. So Deb had a problem.

So Deb met a guy named Joseph Bates, and they started talking about this. Long story short, they came up with a pretty genius, if zen hippie, solution: use sloppy calculations, dude. Just get as close as you need to get.

At first that doesn’t really make sense. Why would a sloppy calculation help make the job quicker? Well, a lot of extra wiring and computing power is used to make a processor’s calculations precise. Strip all that down, and you’re left with calculations which are nearly as good (something like 1% off) which take significantly less power, and take up significantly less space.

Here’s what that means: Deb and Joseph could do something like make a 1000-core processor (because they generate so much less heat, and are so much smaller, it was easy to do) which could go through and process video files at a rate of about 5 gigabytes per second; that’s about 20 times faster than a modern high-end CPU. In terms of accuracy, the processor they made was wrong about 14 pixels for every million. When playing video, you’re not going to see that.

Long story short, these guys came up with a Zen solution to a task which would otherwise take a very long time: when you’re designing a bridge, you’ve got to get your numbers spot-on or else bad things happen. But when you’re doing something where being a little off isn’t a big deal, don’t squander your resources!

Another Engineering heuristic brought to you by the folks at MIT. Gotta love ’em.

Links to the articles I read:


One thought on ““Close enough,” says one hippie CPU

  1. Pingback: Get Real - Link Dump 3-14-11

Leave a Reply

Fill in your details below or click an icon to log in:

WordPress.com Logo

You are commenting using your WordPress.com account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s