Open-source AI infrastructure to boost innovation in China

Source: Xinhua| 2020-03-26 18:03:55|Editor: huaxia

BEIJING, March 26 (Xinhua)-- From smart fever-screening at subway stations to scan-reading diagnosis, artificial intelligence (AI) is on the frontline of China's battle against the novel coronavirus.

Behind the smart systems are deep-learning frameworks that emulate the way the human brain learns, like recognizing patterns and coping with ambiguity.

Megvii, developer of the fever-screening system and one of China's major AI unicorns, announced on Wednesday it will make its core deep-learning framework MegEngine open-source to the world.

As more leading Chinese AI enterprises embrace open-sourcing, experts expect AI development in China to shift into overdrive, paving the way for more innovation.

Tang Wenbin, co-founder and chief technology officer of Megvii, said in an online press conference that AI infrastructure is composed of AI chips and production platforms. The latter includes data, algorithms and computing power.

Tang used a cooking analogy to explain deep-learning. The data "ingredients" are cleaned and sorted. The training algorithm "cooks" the data so the computer can identify trends and make predictions on unseen data, and the deep-learning framework is like the pot.

With the help of a pre-built and optimized framework, engineers can focus on the high-level structure of their model, without getting into the details of underlying algorithms.

Meanwhile, the government is to expedite construction of "new infrastructure" projects such as 5G networks and data centers, shoring up information services for the growing market.

As digital transformation is expected to boost innovation and underpin economic and social development, it is a good time to release its open-source framework, said the company.

Tang said Megvii hopes its deep-learning framework will become a cornerstone of China's AI development and an integrated part of China's AI infrastructure.

Gao Wen, an academician of the Chinese Academy of Engineering, said open-source frameworks are a trend in global AI development.

As the source codes are freely available and can be modified and developed in new directions, more opportunities and products are expected to emerge from an open AI ecosystem.

Silicon Valley tech giants have released several deep-learning frameworks, such as TensorFlow by Alphabet's Google and Facebook's PyTorch.

In China, Baidu announced the open-sourcing of its PaddlePaddle framework in 2016. Tsinghua University has released Jittor and Huawei is expected to release MindSpore this year.

Harry Shum (Shen Xiangyang), a former executive vice president of AI and research at Microsoft, who recently assumed a Dual-Professorship position at Tsinghua University, said building responsible AI is a challenge and requires participation by multiple parties.

"Only when a good framework is open-source, more developers, researchers and enterprises can join in, and more people can have a deep understanding of AI," he said.