Baidu Builds Largest Computer Brain for Online Queries

Sept. 4 (Bloomberg) -- Bloomberg’s Mia Saini examines the battle for internet dominance in China and what these companies may mean to e-commerce in the United States. She speaks on “Bloomberg Surveillance.”

Baidu Inc. (BIDU) is building the world’s largest and most powerful computer cluster to improve image recognition as online queries move away from text.

With about 100 billion digitally simulated neural connections, Baidu’s computing cluster will be 100 times more powerful than the 2012 Google Inc. (GOOG) project dubbed “Google Brain,” Andrew Ng, chief scientist at the operator of China’s biggest search engine, said in an interview yesterday. Engineers at a Baidu lab in Silicon Valley are designing the project, which will be built in Beijing and completed in about six months, according to Ng.

Within five years, online searches based on voice and pictures will account for more than half the total as users of mobile devices seek easier ways to find information, Baidu Chief Executive Officer Robin Li said yesterday. Better image recognition will require more computing power, according to Ng.

“The bigger you build these things, the better they perform,” said Ng, who was hired by Baidu in May. “Our initial task is to recognize images better, to create computer vision.”

The Baidu system will be about 10 times more powerful than a 2013 computer cluster at the Stanford Artificial Intelligence Laboratory, Ng said. The chief scientist was involved in both the “Google Brain” and Stanford projects before joining Baidu.

Photographer: Nelson Ching/Bloomberg

The Baidu Inc. logo hangs in the lobby of the company's headquarters in Beijing. Close

The Baidu Inc. logo hangs in the lobby of the company's headquarters in Beijing.

Close
Open
Photographer: Nelson Ching/Bloomberg

The Baidu Inc. logo hangs in the lobby of the company's headquarters in Beijing.

The human brain has about 100 trillion neural connections.

Baidu’s network is using the same kind of more powerful servers with graphics process unit computing that was applied by Stanford researchers, rather than the central processing unit technology deployed for Google’s network, Ng said.

Deep Learning

So-called deep learning capability by computers configured to simulate human brain function has already reduced speech recognition error rates by 25 percent, Ng said.

Mobile technology is driving online searches, shifting demand for how users query for information, Li said yesterday at the company’s Baidu World conference in Beijing.

China had 632 million Internet users as of June, with 83 percent using mobile phones to access the Web, according to the China Internet Network Information Center.

About 10 percent of Baidu search queries are done by voice, Li said yesterday. Within five years, voice and image searches will surpass text queries, he said.

“In the mobile era, consumer behavior is changing,” Li said at conference. “For search, we can now use voice. Voice and speech is a more natural way to express our requests.”

To contact Bloomberg News staff for this story: Edmond Lococo in Beijing at elococo@bloomberg.net

To contact the editors responsible for this story: Michael Tighe at mtighe4@bloomberg.net Aaron Clark, Terje Langeland

Press spacebar to pause and continue. Press esc to stop.

Bloomberg reserves the right to remove comments but is under no obligation to do so, or to explain individual moderation decisions.

Please enable JavaScript to view the comments powered by Disqus.