How Baidu’s deep learning framework can give China a leg-up in the AI race

Baidu’s Silicon Valley AI Lab, where PaddlePaddle was born. Photo credit: Baidu.

Chinese search giant Baidu has been on a roll of artificial intelligence investments this year, from acquisitions like Raven Tech to the opening of a new augmented reality research team, spun off from the tech giant’s Institute for Deep Learning.

Now, Baidu is looking at its deep learning framework, dubbed ‘PaddlePaddle‘ (after PArallel Distributed Deep LEarning), as a way to partner with – and capitalize on – larger Chinese companies in traditional industries, such as electronics manufacturers.

These companies “have a strong need for AI, especially in China,” Wang Yi, tech lead of PaddlePaddle, tells Tech in Asia. “All the TV shops, no matter if they’re old like Changhong, or new like Leshi – all their TVs are remoteless. They all use voice recognition.”

China now conducts more AI research than the US.

While Western companies like Google and Amazon take most of the limelight when it comes to artificial intelligence, China’s influence in the field is growing. China now conducts more AI research than the US. Its researchers have such a strong presence abroad that this year’s Association of the Advancement of Artificial Intelligence gathering – held in New Orleans – had to be rescheduled because it conflicted with Chinese New Year.

PaddlePaddle could contribute to China’s vision for powerful, homegrown artificial intelligence. For starters, Google’s services are blocked by the Great Firewall. Language barriers can also limit the reach of Western AI frameworks in China.

This makes PaddlePaddle suitable for Chinese developers who want to build and train neural networks which can learn a variety of skills, like recognizing cats and transcribing spoken Mandarin.

Also, unlike some other frameworks, PaddlePaddle is designed to work well with clusters, which are groups of connected computers that can run on-premise – and can be more tightly controlled. In contrast, cloud computing is deliberately designed so that developers don’t have to worry about the underlying hardware powering their applications.

“Traditional businesses usually do not understand cloud so they don’t dare use it,” says Wang Yi. Instead, they prefer clusters. That’s where he sees opportunity – working with companies that need AI capabilities, but are uncomfortable with storing their data on the cloud.

Already, Baidu has been approached by several Chinese companies, he says. Those in discussion include a domestic smartphone brand and an airline. Last month, Baidu also announced PaddlePaddle’s compatibility with Kubernetes, a popular open source tool that helps developers manage the workload of computers in a cluster.

Xu Wei, creator of PaddlePaddle. Photo credit: Baidu.

Deep learning via clusters is also where Baidu is hoping to make its mark – especially because cloud isn’t the company’s forte. In China, Baidu’s cloud service is eclipsed by Alibaba Cloud, the cloud computing arm of ecommerce giant Alibaba.

First China, then the world