More and more China-based research teams have been taking over global AI competitions.
In April, a joint laboratory between the Harbin Institute of Technology in Northeast China's Heilongjiang Province and Hefei-based AI firm iFlytek in East China's Anhui Province took the lead on Stanford reading comprehension tasks, Hu Yu, executive director of iFlytek, noted during the conference.
In a recent Microsoft Common Objects in Context dataset studying object recognition on October 29, Beijing-based Megvii Technology Inc beat its foreign competitors such as Facebook and Google, ranking first.
"China, with its huge population that can generate a tremendous volume of data, has now become the only one country that can be compared with the U.S. in terms of AI development," Shao Yang, president of strategic marketing at Huawei's consumer business diion, said at the conference.
China was ranked as the top country in terms of numbers of AI publications cited worldwide in 2015, followed by the U.S. and India, worldwide management consultancy McKinsey said in a report released in April.
Although China does not yet have the same kind of vibrant AI ecosystem as the U.S., the country is on a par with others in terms of algorithm development, particularly in voice recognition and targeted advertising, the report noted.
Optimizing core chips
While Google's tensor processing unit (TPU), a cloud system combined with software and hardware designed for machine learning, has been rattling the AI industry over the past year, more questions have been raised regarding which cutting-edge AI platforms, such as graphics processing unit (GPU) or field-programmable gate array (FPGA), are more suitable for deep learning.
"It's hard to design a specific platform for the purpose of 'training' AI systems, we always expect maximized flexibility in an algorithm. So the best model is likely to be equipped with both CPU [central processing unit] and GPU," said Hu Leijun, vice president of Chinese information technology firm Inspur.
The core technology in the AI sector is based on computing capacity, but sustainability should also be built on a friendly ecosystem, Wang Zai, vice president of Chinese AI chipmaker Cambricon Technologies Corp, said at the conference.
The company was the first "unicorn" in China's AI semiconductor sector to be worth more than $1 billion. After it released the first AI chip in the country in 2016, it teamed up with Huawei and found a way to commercialize its core technologies.
"As the smartphone maker's flagship product Mate 10 is equipped with Cambricon's neural processing unit, we have to constantly upgrade our technologies to improve user loyalty," Wang said.
Over the next five years, China's AI will make even more major breakthroughs in empowering industries, Yang noted.
"But it might lag behind in developing open-source software as well as general chips," she said.
Also, the U.S. has a more robust AI ecosystem than China, industry representatives noted during the conference. In terms of basic research, U.S. scholars demonstrate much deeper study and understanding of fundamental fields, for example, the study of math.
"Chinese scientists are very smart, but some are too eager to turn their research into profits," said Micree Zhan, CEO of Beijing-based custom chip manufacturer Bitmain.