TECHNOLOGY FOR CHANGE

How to Harness Future Personal Supercomputers

R. Colin Johnson | Date: 12-16-09 | Comments

In anticipation of mobile devices reaching supercomputer speeds by 2018, Intel is crafting pattern-recognition algorithms for use today that can be scaled up to take advantage of such future speeds.

At Intel, its People and Practices Research Group is crafting pattern-recognition algorithms that it hopes will gracefully scale-up, as more and more consumer electronic devices approach supercomputer speeds.

According to Intel, by the year 2018 mobile phones will have somewhere between 4 and 10 teraflops of processing power, surpassing the speeds of room-filling supercomputers of not so very long ago. In anticipation of these advances, Intel's "Everyday Sensing and Perception" study aims to answer the simple yet relevant question: "What are people going to do with these systems when they achieve supercomputer speeds?"

Richard Beckwith, a research psychologist at Intel's People and Practices Research Group, explained the group's decision to develop an application now, one that uses the same algorithms he believes will be used in the future, "but which we can actually deploy today."

To get under way today, Intel chose an educational application that was both practical and immediate. By observing grade school teachers and their students working, Intel was able to compose pattern-recognition algorithms that can be scaled up, for eventual use by adults, by introducing computing power in increased increments. Pattern recognition is relatively easy in a restricted domain. In this case, using data from a video camera to distinguish between types of "math manipulatives"�plastic coins, pyramids and colored blocks, which are already being utilized in about 50 percent of U.S. classrooms to help teach mathematical concepts to children.

The first algorithm the group crafted was one that assigned children the task of sorting coins into piles by value. For this, the group developed an application for Intel's ClassMate PC, a low-cost netbook-like computer whose internal camera was modified to aim downward in front of the child to "watch" as they performed their tasks. The ClassMate PC watched the child sort plastic coins into like piles of pennies, nickels and quarters. If they got it right, it gave them positive reinforcement with spoken feedback by saying that they did a good job.

Besides the pattern-recognition algorithm, Intel is crafting classroom management software for the teachers that will allow them to monitor the progress of each student even if they are working within a larger group. The students will also be able to take their ClassMate PCs and their math manipulatives home to perform sort and other similar tasks as homework, "something that hasn't been possible before," said Beckwith.

The People and Practices Research Group is also creating a pico-projector augmentation to the system that will directly project positive and negative feedback images onto the coins themselves, such as putting a red spotlight on coins that are sorted incorrectly.

Beta testing will begin in early 2010 and, if successful, will be scaled up to work on adult applications, such as searching through video streams for photos of friends, by the time mobile devices have reached supercomputer speeds circa 2018.

Watch a video here of how Intel aims to harness pocket supercomputers.