Sloppy maths surprisingly useful
6 Jan 2011 by Evoluted New Media
If your head is still a little fuzzy after the New Year and you can’t keep up with the calculations needed on the job, don’t worry because sloppy arithmetic is actually useful according to researchers at MIT.
If your head is still a little fuzzy after the New Year and you can’t keep up with the calculations needed on the job, don’t worry because sloppy arithmetic is actually useful according to researchers at MIT.
Imagine an arithmetic circuit that gives imprecise answers: it would be smaller than those used in today’s computers; consume less power and more of them could fit on a single chip, increasing the number of calculations performed at once. But how useful would these chips be – the answer is surprisingly useful.
Joseph Bates, an adjunct professor of computer science at Carnegie Mellon University has designed a chip that could perform tens of thousands of simultaneous calculations using sloppy arithmetic and was looking for applications that leant themselves to it.
The chip has a thousand processing units – compared to the four or eight usually found in commercial computer chips – and communicate locally meaning they can be much smaller, and time and energy efficient.
Bates joined forces with Deb Roy – a researcher at MITs Media Lab – and graduate student George Shaw to see if video algorithms, which are fairly error-prone, could be retooled to tolerate sloppy arithmetic. They used an object recognition system algorithm designed to distinguish foreground and background elements in video frames.
Shaw rewrote the algorithm so the results of the numerical calculations were increased or decreased by a randomly generated factor of between 0 and 1%. He compared its performance to the standard implementation of the algorithm.
“The difference between the low-precision and the standard arithmetic was trivial,” said Shaw. It was about 14 pixels out of a million, averaged over many, many frames of video.”
However, the algorithm has to do more than that to be considered useful, and the researchers are exploring what to do next. The chip may be particularly compatible with image and video processing, and although it hasn’t been manufactured yet, it should work as anticipated. It might also find uses in “nearest neighbour searches” in computer science, or computer analysis of protein folding.
