June 28, 2012
Search giant Google along with researchers from Stanford University have made an interesting discovery based on an X labs project. After being fed 10 million images from YouTube, a 16,000-core cluster learned how to recognize various objects, including cats. Earlier this week, the New York Times detailed the program, explaining its methods and potential use cases.
The project began a few years ago when Google researchers planned to make a human brain simulation. The compute cluster acted as the brain’s neural network, tasked with learning on its own and using the Internet as a source of information. After processing millions of unlabeled YouTube thumbnails, the system taught itself how to recognize a cat. While the video website is known for its comprehensive collection of user submitted feline antics, project researchers were focused on the simulation’s ability to learn objects without human input.
A 1,000-node cluster was the basis for the neural network, representing more than 1 billion connections. The unlabeled images were collected randomly and processed by machine-learning algorithms. Similar to IBM’s Watson, the technology relies on “deep learning” techniques, using previous outcomes to inform future decisions. Machine learning has also become integral in other applications, including speech recognition.
The researchers removed all identifying labels from the images because they wanted to see if the network was able to create the concept of an object. Dr. Jeff Dean of Stanford University explained how the project differs from other recognition technologies. “We never told it during the training, ‘This is a cat,’ ” he said “It basically invented the concept of a cat. We probably have other ones that are side views of cats.”
As a result, the software generated a vague image of a cat on its own. The simulation has also introduced possible evidence of “grandmother neurons”, which some believe are specialized cells, trained to recognize an individual face or concept.
It’s not all about cats, of course. The system was also able to identify human faces and bodies to some degree. Compared to previous attempts to identify unlabeled images, the simulation fared much better at learning and recognition. According to the researchers’ findings, they were able to deliver 15.8 percent accuracy in recognizing 20,000 object categories. They claim that’s 70 percent better than what had been achieved previously.
According to David Bader, executive director of high-performance computing at the Georgia Tech College of Computing, the simulation represents an improvement of “an order of magnitude over previous efforts.” He believes this work could lead to a complete model of the human visual cortex before the end of the decade.
Full story at The New York Times
10/30/2013 | Cray, DDN, Mellanox, NetApp, ScaleMP, Supermicro, Xyratex | Creating data is easy… the challenge is getting it to the right place to make use of it. This paper discusses fresh solutions that can directly increase I/O efficiency, and the applications of these solutions to current, and new technology infrastructures.
10/01/2013 | IBM | A new trend is developing in the HPC space that is also affecting enterprise computing productivity with the arrival of “ultra-dense” hyper-scale servers.
Ken Claffey, SVP and General Manager at Xyratex, presents ClusterStor at the Vendor Showdown at ISC13 in Leipzig, Germany.
Join HPCwire Editor Nicole Hemsoth and Dr. David Bader from Georgia Tech as they take center stage on opening night at Atlanta's first Big Data Kick Off Week, filmed in front of a live audience. Nicole and David look at the evolution of HPC, today's big data challenges, discuss real world solutions, and reveal their predictions. Exactly what does the future holds for HPC?