In the past two days, the media has made a lot of hype about this report; Personally, I am particularly interested in the two trend charts in this report. The report shows that from 20 13 to 20 15, the number of articles referring to "deep learning" in SCI has increased by about six times, and it is emphasized that "from the perspective of the number of articles, the United States is no longer the first in the world."
America is not the first in the world? Who is number one in the world? Please look at the two trend charts in the report:
1.
The above figure shows the changing trend of the number of articles cited by "deep learning" or "deep neural network" in SCI every year, and distinguishes them by country. The data itself is queried from the core database of the Science Network.
2.
The second picture is the same as the other conditions in the first picture, except that an additional condition "the article must be quoted at least once" is added. In other words, the number of articles counted in the second figure is the "effective number of articles cited", which is relatively more reasonable.
In these two pictures, China surpassed the United States (and all other second-tier countries) at 20 14 and 20 15, and was in a leading position.
The question is, in the field of AI research, are China people really that strong?
Intuitively, over the years, there are many people from China/China in the field of artificial intelligence, such as Andrew Ng, Sun Jian politician, Yang Qiang, Huang Guangbin, Ma Yi and Zhang Dapeng. I can tell you a lot. I have seen it with my own eyes in Google's research team, Microsoft Research Institute and other places, and people from China and face of china are everywhere. But this is just an intuitive feeling. On the whole, how important is the scientific research contribution made by China people/China people, and how big is the promotion of artificial intelligence? Are the statistics in the White House report scientific and reasonable?
In fact, I have several questions about the statistics in the US national strategic plan, mainly including:
Directly searching for the two keywords "deep learning" and "deep neural network" can really cover all the scientific research progress in the field of artificial intelligence in recent years? In the fields of robotics, intelligent control, machine vision and driverless driving, there are still many articles that do not mention deep learning. How many related articles will be missed by strictly matching keywords? Does it affect the statistical results?
"The article has been quoted at least once" is scientific, but it seems far from enough. Do you really need to consider the influencing factors of SCI for this kind of statistics? If you don't consider it, how many papers published in poor journals will be mixed in? Will these data be polluted by the behavior of domestic research institutions to cite SCI to evaluate professional titles?
In the era of globalization, according to national statistics, will there be obvious deviation? The White House report did not mention how statistical articles were classified into different countries. According to the institution where the author works at the time of publication, will a large number of China scholars who visit foreign institutions be counted as foreigners? How to count the articles written by China people and foreigners? If the journals belong to different countries and regions, will the differences of journal level (impact factors) in different countries make the statistical results biased?
Based on this, I also want to do my own statistics.
Before the statistics, I set myself several principles:
Starting from the impact factors of journals, only the top journals with high impact factors are counted.
Starting from the keywords of Web of Science, it covers all scientific research fields related to artificial intelligence, not just articles on deep learning.
It's my own choice to focus on China people instead of comparing China with foreign countries by country classification, because in today's academic circles, cooperation and exchanges between countries can't be ignored. Overseas Chinese like Andrew Ng are actually contributing to the development of artificial intelligence in China. Instead of restricting the country, it is better to take a look at China scientists and researchers on earth as a whole.
The time range of concern is from 2006 to 20 16 years, spanning about 10 years-because many people say that this wave of artificial intelligence really rose in the years after 2006.
My statistical results
Let's take a look at the average contribution of China's authors in 2006-20 16:
3.
During the period from 2006 to 20 16, among the nearly 20,000 top artificial intelligence articles, Chinese contributions and citations accounted for 29.2% and 3 1.8% respectively. In the past ten years, China people contributed an average of 30% of the top AI research articles and citations with about one fifth of the authors. Statistically speaking, this is an above-average scientific research contribution.
However, the average level does not show the strength of scientists and researchers in China in recent years. From 2006 to 2065438+05, the changing trend of the number of articles submitted and cited by China people;
4.
From 2006 to 20 15, the proportion of top AI papers participated by China authors increased from 23.2% to 42.8% year by year. However, the number of cited AI top papers participated by China increased from 25.5% to 55.8% year by year (the data of 2016 is less, so it is not used for trend comparison).
That is to say, even if only the top articles in top journals are counted, the contribution of China people/China people in the field of artificial intelligence is exactly the same as that revealed by the White House report-from any perspective, China people/China people play a decisive role in the field of artificial intelligence, and from 20 14 and 20 15, China people/.
case analysis
Will anyone think this statistic is incredible? Will this result surprise many people? We can also give a more specific example for in-depth analysis.
In the top artificial intelligence journals and conference proceedings, I give a famous example: IEEE Transactions on Pattern Analysis and Machine Intelligence (PAMI), with a 20 15-year impact factor of 6.077, which is so high that there are no friends. Students who want to contribute here may know how difficult it is to be admitted and published.
According to the number of citations, I selected the top 500 papers of IEEE Transactions on pattern analysis and machine intelligence in 2006-20 16. The following is the citation distribution of these 500 papers:
5.
It's actually horrible. The top 500 articles have the highest citation number of 27 15 and the lowest citation number of 41-what a top journal! Ordinary periodicals are hard to compare with.
So, in these 500 top artificial intelligence papers, what are the contributions of scientists and researchers in China? Let's start with a few figures: 500 top authors 1220, among which 3 16 are China scientists and researchers, accounting for 25.9%. The total cited times of all authors are 23 136 1, among which the total cited times of China scientists and researchers are 63,846, accounting for 27.6%. If we only look at the data of 20 14 years (at that time, the number of articles and citations in Chinese was very high), scientists and researchers in China accounted for 5 1.8%, more than half.
6.
If you only look at the details of 10 China authors and 10 non-China authors who are most cited in IEEE Transactions on Pattern Analysis and Machine Intelligence 2006-2016, it is also a very interesting table:
7.
The top 10 China people and the top 10 non-China people cited by IEEE Transactions on pattern analysis and machine intelligence have almost the same number of citations per capita. Indeed, the contribution of top AI scientists, Chinese people/China people, is no less than that of other scientists.
In addition, if you are not familiar with the research field of artificial intelligence, especially pattern recognition, then remember the names of the top 20 scientists in the table. If you are interested, you can search their resumes to see where they all work and do research, and who their students and colleagues are. I believe there will be many discoveries.