The Roadrunner supercomputer simulates the human visual system

It takes a supercomputer to really simulate one of the most complex parts of the human brain, the visual cortex, and the scientists at Los Alamos have one operating for only one week but doing more than any other computer in operation today. The supercomputer nicknamed Roadrunner can execute 1 million billion instructions per seconds (petaflops) making it the fastest supercomputer in operation today; it more than two times faster than the previously fastest offering by IBM. Roadrunner is built using AMD Opteron dual-core processors each with four PowerXCell 8iTM processors for an aggregate of more than 100,000 cores to the tune of $133 million; the supercomputer like many others in the past was engineered by IBM.

So what are the scientists using this computational workhorse for? Apparently, one of the many programs they had running on it this past week was PetaVision which is a huge simulation of the human visual system.

PetaVision models the human visual system—mimicking more than 1 billion visual neurons and trillions of synapses. Neurons are nerve cells that process information in the brain. Neurons communicate with each other using synaptic connections, analogous to what transistors are in modern computer chips. Synapses store memories and play a vital role in learning.

Synapses set the scale for computations performed by the brain while undertaking such tasks as locomotion, hearing or vision. Because there are about a quadrillion synapses in the human brain, human cognition is a petaflop/s computational problem. (source)

Lots of effort in computer vision research today goes in creating image processing algorithms that try to achieve performance at the human level with as little computation as possible; the reason for this is obvious: Computers are not fast enough. In fact, I am actually working on an object recognition system and all my efforts go on minimizing the number of pixels in the image that I consider for the recognition task. If I process all the pixel data then I'd have time to get another PhD before a single trial is done. With faster computers (okay, so you won't have a desktop machine supercomputer with petaflop performance in your home any time soon,) such issues will essentially become less important. It reminds me of what has happened in the computer graphics field after the popularization and performance increase in GPUs. Eventually, the simple and easy to code algorithms will become the focus or computer vision research given the abundance of CPU cycles while the more mathematically advanced algorithms will be put aside.

In conclusion, I have to say that I am glad that one of the problems the scientists are actually experimenting with on Roadrunner can give us better insight about how the most complex part of the human brain works. I say this because it is probably obvious that most cycles on this machine will go on computing more effective ways to 'dispose' of people in the many current and future wars. It is sad that such great machines are used to both discover new knowledge and at the same time destroy fellow human beings.

Roadrunner photo credit: LeRoy N. Sanchez, Records Management, Media Services and Operations