Baidu goes all in on AI at dev conference

Self-driving cars, Intel and cameras, oh my!

China's Baidu has made a host of AI announcements at its Baidu Create developer conference in Beijing this week, including a new chip (joining the likes of Google and Nvidia); an Intel partnership; and a retail camera.

The Kunlun chip is designed specifically for artificial intelligence purposes, for both device-based edge computing and in the cloud through data centres. The Kunlun 818-300 model is being built to train AI and the 818-100 for inference.

Baidu began working with field-programmable gate arrays (FPGAs) for deep learning in 2011, and says that Kunlun is about 30 times faster than its first efforts in the field. It can reach 260 tera-operations per second with 512 GBps memory bandwidth, the company told VentureBeat.

‘Traditional chips limit how much computing power is available and thus how far we can accelerate AI technologies. Baidu developed this chip, specifically designed for large-scale AI workloads, as an answer to this demand', the company said in a statement.

Baidu has not announced a release date for the chip yet.

Self-driving into the sunset

The company also made announcements around its Apollo self-driving vehicle programme. It described a partnership with Chinese bus maker King Long to create autonomous buses called Apolong, which will be deployed in cities including Beijing, Shenzhen and Pingtan in China and Wuhan and Tokyo in Japan, early next year.

Also in the Apollo sphere, Baidu is working with Intel to deploy technology from Israeli developer Mobileye into its self-driving vehicles. Baidu will build Mobileye's Responsibility-Sensitive Safety (RSS) model into the codebases of the commercial Apollo Drive and Apollo Pilot (the deployment version of Project Apollo) programmes - making it the first autonomous driving solution provider to publicly adopt RSS.

Baidu will also take on Mobileye's Surround Computer Vision Kit hardware to detect surroundings.

Got my Xeye on you

Baidu is using more Intel technology in the new Xeye (no, we're not sure how that's meant to be pronounced, either) camera, combining the Movidius Myriad 2 vision processing units (VPUs) with the Chinese firm's machine learning algorithms.

The camera will use the combination of technologies to analyse objects and gestures and detect people to provide ‘personalised shopping experiences' in retail environments.

Finally, we heard about Baidu Brain 3.0, an upgrade to Baidu's suite of AI services. The drag-and-drop training platform for AI models now has 110 services, from computer vision and natural language processing to face recognition.

A spokesperson said that Brain had 60 AI services at the last Baidu Create.