Big Data Abstracts
Compiler: Walker, Jiang Baoshan
Researchers are easily overlooked, but it is undeniable that they are the core components of all AI ecosystems.
For a long time, there has been controversy over the strength comparison of AI between China and the United States, but the relevant data support is weak.
Therefore, according to the papers published at the top AI annual meeting approved by experts, the macro policy think tank of Paulson Research Institute created an original data set in Macarpolo, which provides a sufficient data basis for evaluating the quantity and quality of AI researchers in China and the United States.
Leading AI research is relatively open source, so talent is one of the most easily quantified indicators in its ecosystem components. Paulson Institute has collected published research data, trying to obtain the development of AI research in various countries from the distribution of AI researchers.
Neural Information Processing Systems (NeurIPS) is the most important event in the field of artificial intelligence and machine learning. Paulson Institute has made a complete analysis of NeurIPS 2018 papers, and draws conclusions from the following aspects:
Here are some key points of research
1. In the top AI research, the proportion of Chinese researchers is relatively small (about 9%), but the proportion of high-quality AI research is relatively large (about 25%).
According to research fellow Joy Dantong ma's recent data analysis of top paper authors in a speech delivered at NeurIPS 2018, 10 out of 113 authors are Chinese nationals.
Notably, he found that,At present, these ten elite Chinese authors are affiliated with or about to join American research institutes (universities or companies).This echoes Jeffrey Ding's previous analysis of the authors of the NeurIPS speech in 2017, which found that:Fourteen percent of the authors are from China, but only one percent currently work in Chinese research institutions.
In 2018, we conducted the same country-of-origin analysis of high-quality (but not top-notch) publications and found that about a quarter (955) of the 3824 authors were Chinese nationals.
This discovery indicates that:Although Chinese researchers have not yet fully climbed to the top of the AI research pyramid, they account for a considerable proportion of the upper AI research.
2. Most Chinese-born researchers conduct AI research in American research institutes.
At present, the majority of senior Chinese researchers (59%) are affiliated to American research institutions, 33% are affiliated to Chinese research institutions, and about 9% are affiliated to other countries such as Canada, Singapore and Japan.
This shows that although most senior AI researchers are still flocking to American research institutes, the proportion of these researchers in Chinese research institutes is much higher than that of top researchers.
3. Most senior researchers of Chinese nationality study in American universities, and most of them work in the United States after graduation.
Nearly 60% of senior Chinese researchers study in the United States, 35% in China and 7% in other countries (Australia and the United Kingdom).
The vast majority (78%) of Chinese writers who graduated from American universities are currently working in American research institutions, while only 21% are working in Chinese research institutions.
The rise of China's technology industry over the past decade has dramatically changed the minds of many Chinese technicians working in Silicon Valley, many of whom have returned home to work in start-ups or Chinese technology giants.
Recent restrictions on postgraduate visas in the United States often result in unfair prosecutions against Chinese scientists in the United States and political statements that all Chinese students are spies, which have begun to affect the mobility and detention of Chinese AI researchers. In view of this, the data on where Chinese nationals and American-educated researchers will work may be a lagging indicator and may change substantially in the coming years.
Whether these impacts are positive (protecting the comparative advantage of the United States in top research) or negative (weakening the unique ability of the United States to attract and retain talent) remains an unknown question. This is also a follow-up article in this series. We will build new data sets on this issue and continue our research.
Annotations and research methods
1. NeurIPS is one of the most important AI meetings
two。 Based on a survey of 113 authors who spoke at NeurIPS in 2018, figures for the top 1 percent were obtained. The top 20 per cent were estimated on the basis of a random sampling of 69 of the 1087 authors with Chinese surnames (confidence interval /-7.8%, confidence level 0.95). Then, we studied each of the authors in this sample to find their country of origin, graduate school location and current job affiliation.
3. In order to match each author's country of origin, we use the location of their undergraduate colleges as the preferred alternative value. For high school educated authors, we identify their country of origin according to their high school location.
But the alternative is not perfect: for Chinese nationals who complete undergraduate studies in the United States, if they can't find information about their high school location, they will be regarded as American nationals. This may lead to a slightly lower proportion of Chinese authors. However, due to the lack of information on undergraduate education, some writers with Chinese surnames and working in Chinese research institutions are excluded, so this deviation may be partially offset.
4. When assigning affiliations to multinational research institutions, we use the headquarters of a company or university. For example, Chinese researchers working for Microsoft Asia Research Institute in Beijing will be considered affiliated to an American Research Institute because Microsoft's headquarters is in the United States. Hong Kong-based research institutes are regarded as Chinese institutions.