Recognition
Boffins from MIT and Harvard have worked out a way to
build better artificial visual systems with high-performance gaming hardware.
One of the most important things about visual systems is
that they need to be able to recognise what they are seeing. This is difficult
because much of the inner workings of biologically based systems remain a
mystery. However using GPUs MIT and Harvard researchers are now
making progress faster than ever before.
They built a powerful computing system that delivers over
hundred fold speed-ups relative to conventional methods. According to Nicolas Pinto, a PhD candidate in James
DiCarlo’s lab at the McGovern Institute for Brain Research at MIT the extra
power means that it is possible to discover new vision models that traditional
methods miss.
The boffins used PS3s armed with dozens of
high-performance Nvidia GPUs and then designed a high-throughput
screening process to tease out the best parameters for visual object
recognition tasks. The have managed to accurately identifying a range of
objects on random natural backgrounds with variation in position, scale, and
rotation. The project would have taken two years with conventional computing
but this time only took a week.
The next stage is to try things like face identification,
object tracking, pedestrian detection for automotive applications, and gesture
and action recognition.