The 2023 Artificial Intelligence Index Report is released! This report shows that China ranks first in the world in AI top conference papers, but the number of citations is lower than that of the United States. In addition, among the world's top ten AI paper publications, China has 9 seats, catching up with MIT one after another.
Recently, Stanford released the 2023 AI Index Report.
It is worth noting that the Stanford AI Index report lists the top ten institutions in the world for "AI paper publication volume", and 9 of them are from China, catching up with MIT one after another.
They are: Chinese Academy of Sciences, Tsinghua University, University of Chinese Academy of Sciences, Shanghai Jiaotong University, Zhejiang University, Harbin Institute of Technology, Beijing University of Aeronautics and Astronautics, University of Electronic Science and Technology of China, Peking University, and MIT.
This year's report is divided into eight main sections: Research and Development, Technology Performance, Ethics of AI Technology, Economics, Education, Policy and Governance, Diversity, and Public Perspectives.
The following extracts several key points from the report.
The cooperation between China and the United States ranks first in the world
From 2010 to 2021, although the pace of cross-border cooperation in AI papers has slowed down, since 2010, the number of artificial intelligence research cooperation between the United States and China has increased by about 4 times , which is 2.5 times more than the total number of cooperation between China and the United Kingdom.
However, from 2020 to 2021, the total number of Sino-US cooperation will only increase by 2.1%, the smallest year-on-year growth rate since 2010.
Furthermore, the total number of AI papers has more than doubled since 2010. From 200,000 in 2010 to almost 500,000 in 2021 (49,601).
In terms of the type of AI papers published, in 2021, 60% of all published AI papers will be journal articles, 17% will be conference papers, and 13% will come from repositories.
While journal and repository papers have grown 3-fold and 26.6-fold, respectively, over the past 12 years, the number of conference papers has declined since 2019.
Pattern recognition, machine learning, and computer vision are still hot topics in research in the field of artificial intelligence.
China continues to lead in the total number of journals, conferences and repositories.
The U.S. still leads in terms of AI conference and repository citations, but those leads are slowly being eroded. Nonetheless, most of the world's large-scale language models and multimodal models (54% in 2022) are produced by US institutions.
China dominates the list of AI top conferences, but the number of citations is lower than that of the United States
In the publication of AI journal papers, China has always maintained its leading position, with 39.8% in 2021, followed by the European Union and the United Kingdom (15.1%), and then the United States (10.0%).
Since 2010, the proportion of citation frequency of Chinese artificial intelligence journal papers has gradually increased, while the European Union, the United Kingdom, and the United States have all declined. China, the European Union and the United Kingdom and the United States accounted for 65.7% of the total global citations.
So, what about the publication of papers by the world's top conferences?
In 2021, China will account for the largest share of the number of papers published at the world's top AI conferences with 26.15%, followed by the European Union and the United Kingdom with 20.29%, and the United States with 17.23%.
Judging from the citations of top conference papers, although China is highly productive, its citations are lower than those of the United States. The citations of the top conference papers in the United States are 23.9%, and 22.02% in China.
It can be seen from the side that China publishes the largest number of papers, but the quality is not as high as that of the United States.
The United States leads the world in AI paper repository submissions with 23.48%. The lowest in China, 11.87%.
9 institutions in China, AI papers published surpassed MIT
In 2021, China will account for 9 of the world's top ten institutions for the total number of papers published. The total number of papers published by different institutions is shown in the figure below. MIT ranks tenth, with 1,745 papers published.
In terms of computer vision (CV), ten institutions in China rank among the top ten in the world. They are Chinese Academy of Sciences, Shanghai Jiaotong University, University of Chinese Academy of Sciences, Tsinghua University, Zhejiang University, Beihang University, Wuhan University, Beijing Institute of Technology, Harbin Institute of Technology, and Tianjin University.
In the field of natural language processing (NLP), things are different.
The top ten institutions/companies in the world are: Chinese Academy of Sciences, Carnegie Mellon University, Microsoft, Tsinghua University, Carnegie Mellon University-Australia, Google, Peking University, University of Chinese Academy of Sciences, Ali, Amazon.
Speech recognition ranks as follows:
Industry leads academia
Among the important artificial intelligence machine learning systems released in 2022, the language system accounts for the most, with 23, which is 6 times the number of multimodal systems.
Industry leads academia in paper output.
Until 2014, most important models were published by academia. Since then, the industry has turned around. By 2022, 32 important machine learning models will be born in industry, while only 3 in academia.
It can be seen that building state-of-the-art artificial intelligence systems increasingly requires a large amount of data, computing power, and financial resources compared with non-profit organizations and academia, while industry players certainly have more financial resources to do this. matter.
In 2022, the United States will produce the largest number of significant machine learning systems with 16, followed by the United Kingdom (8) and China (3).
Furthermore, since 2002, the U.S. has surpassed the U.K. and the European Union, China, in terms of the total number of significant machine learning systems created
Looking at the distribution of researchers behind these important AI systems, the United States has the most researchers, 285 people, more than twice that of the United Kingdom, and nearly six times that of China.
LLM is getting bigger and bigger, and the computing power is more expensive
Large language and multimodal models, sometimes referred to as base models, are an emerging and increasingly popular type of AI model that are trained on large amounts of data and are suitable for a variety of downstream applications.
Large-scale language and multimodal models such as ChatGPT, DALL-E 2, and MakeA-Video have demonstrated impressive capabilities and started to be widely deployed in the real world.
Analyzing the country affiliation of the authors of these models, the majority of these researchers were from US institutions (54.2%).
The Stanford AI Index report also lays out a timeline for the release of large language and multimodal models.
Large language models are getting bigger and more expensive.
The first large-scale language model, GPT-2, was released in 2019, with 1.5 billion parameters and a training cost of about $50,000. Google PaLM is one of the large language models launching in 2022, with 540 billion parameters and a cost of up to $8 million.
From the perspective of parameters and training costs, PalmM is 360 times larger and 160 times more expensive than GPT-2.
Not just PalM, but in general, large language and multimodal models are getting bigger and more expensive.
For example, DeepMind's large-scale language model Chinchilla, which will launch in May 2022, is estimated to cost $2.1 million, while BLOOM's training costs about $2.3 million.
Progress of GAN in face generation over time, the last image was generated by Diffusion-GAN, a model that achieves state-of-the-art state-of-the-art on STL-10.
Last year, with the release of models such as OpenAI's DALL-E 2, Stability AI's Stable Diffusion, Midjourney, Meta's Make-AScene, and Google's Imagen, text-to-image generation models have gradually entered the public eye.
Below, enter the same prompt, "A panda playing the piano on a warm Paris night", with images generated by three publicly accessible AI text-to-image systems, DALL-E 2, Stable Diffusion, and Midjourney, respectively.
Of all the recently released text-to-image generation models, Google's Imagen performs the best on the COCO benchmark.
This year, the Google researchers who created Imagen also released DrawBench, a harder text-to-image benchmark designed to challenge increasingly powerful text-to-image models.
In addition, the report also introduced some biases in the current generative AI model. For example, when DELLE-2 prompted the CEO, everyone seemed to adopt a confident posture with crossed arms.
In Midjourney, when prompted to generate "influencers," it generated 4 images of older-looking white males.
For the full report, see:
https://aiindex.stanford.edu/wp-content/uploads/2023/04/HAI_AI-Index_Report_2023.pdf