The ranking of 'high citation conversion rate' of AI papers is released: OpenAI ranks first, Megvii ranks second, and Google ranks ninth

In order to clarify this problem, the American Zeta Alpha platform counted the top 100 most cited AI papers in the world between 2020 and 2022, and came up with some very interesting results.

For example:

"popular star" OpenAI ranks 9th among the institutions with the most cited papers.

However, its name cannot be found at all on the list of institutions with the largest number of published papers.

For another example, Google Meta and Microsoft from the industry are always among the best in all data. However, in general, academia does not lag behind the industry.

In addition, the previous view that "the quantity and quality of China's AI research output may exceed that of the United States" seems to have been cracked in this report -

More and specific data, Let’s look at them one by one.

China is second, OpenAI and DeepMind win in quality

Before the specific analysis, Zeta Alpha first counted the most cited papers each year from 2020 to 2022. They are:

2022:

1. AlphaFold Protein Structure Database: Massively expanding the structural coverage of protein-sequence space with high-accuracy models

Number of citations: 1372
Institution: DeepMind
Topic: Using AlphaFold to increase the coverage of protein structure database

2. ColabFold: making protein folding accessible to all

Number of citations: 1162
Institution: Completed by multiple collaborations
Topic: An open source and efficient protein folding model

3. Hierarchical Text-Conditional Image Generation with CLIP Latents

Number of citations: 718
Institution: OpenAI
Topic: DALL·E 2

4, A ConvNet for the 2020s

Number of citations: 690
Institution: Meta and UC Berkeley
Topic: Successfully modernizing CNN during the Transformer boom

5. PaLM: Scaling Language Modeling with Pathways

Number of citations: 452
Institution: Google
Topic: Google’s 540B large language model, a new MLOps paradigm, including Its implementation process

2021

1. Highly accurate protein structure prediction with AlphaFold

Number of citations: 8965
Institution: DeepMind
Topic: AlphaFold, A huge breakthrough in protein structure prediction using deep learning

2. Swin Transformer: Hierarchical Vision Transformer using Shifted Windows

Number of citations: 4810
Institution: Microsoft
Subject: ViT Powerful variant

3、Learning Transferable Visual Models From Natural Language Supervision

Number of citations: 3204
Institution: OpenAI
Topic: CLIP

4、On the Dangers of Stochastic Parrots: Can Language Models Be Too Big?

Number of citations: 1266
Institution: Washington, DC, USA, Black in AI, The Aether
Topic: Famous position paper, on growing Be critical of the trend of language models, emphasizing their limitations and dangers

5. Emerging Properties in Self-Supervised Vision Transformers

Number of citations: 1219
Institution: Meta
Topic: DINO, revealing how self-supervision of images leads to some kind of proto-object segmentation in Transformers

2020:

1. An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale

Number of citations: 11914
Institution: Google
Topic: The first work to show how ordinary Transformer can perform well in the field of computer vision

2. Language Models are Few-Shot Learners

Number of citations: 8070
Institution: OpenAI
Topic: GPT-3

3, YOLOv4: Optimal Speed ​​and Accuracy of Object Detection

Number of citations: 8014
Institution: "Academia Sinica", Taiwan, China
Topic: YOLOv4

4, Exploring the Limits of Transfer Learning with a Unified Text-to-Text Transformer

Number of citations: 5906
Institution: Google
Subject: A rigorous study of Transformer’s transfer learning resulted in the famous T5

5. Bootstrap your own latent: A new approach to self- Supervised Learning

Number of citations: 2873
Institution: DeepMind and Imperial College
Topic: Showing that negative is not necessary for representation learning

I believe you can find a lot from it "Familiar Faces".

Next, Zeta Alpha conducted an analysis of the information behind highly cited papers in the past three years.

The first is the countries (regions) with the “top 100 most cited papers each year”.

It can be seen that the United States is the strongest, and the gap between China and China is relatively obvious.

Therefore, Zeta Alpha also believes that the previous statement that "China may surpass the United States in AI research" is not valid at least based on this data.

In addition, Singapore and Australia’s rankings are also quite unexpected, ranking fifth and sixth respectively.

"In order to correctly assess the dominance of the United States," Zeta Alpha changed another statistical method and calculated the percentage of the top 100 citations.

# Of course, the United States still ranks first, but we can see that its share has declined in the past three years.

The UK is the largest competitor outside of China and the United States, but the UK’s outstanding performance in 2022 was actually mainly contributed by DeepMind (accounting for 69%).

The following is a ranking of the top 100 individuals with the most citations by organization or institution.

Not surprisingly, Google and Meta Microsoft ranked in the top three, followed by UC Berkeley, DeepMind and Stanford.

OpenAI also received a pretty good ranking, ninth. The tenth place is MIT and the eleventh place is Tsinghua University.

Although the top three contestants are all from the industry, if we only divide them according to the type of institution, the performance of academia and its performance is basically the same.

Then followed by the ranking of the total number of papers published by each organization or institution in the past three years.

The boss is still Google. The second place is more eye-catching, which is Tsinghua University, followed by Microsoft, CMU, MIT, Stanford, UC Berkeley, Peking University (eighth), Meta...

It can be seen that the top ten are affiliated with academia Institutions or organizations occupy a large area of ​​the country.

We searched for a long time, but did not see the names of OpenAI and DeepMind -

Obviously they published a smaller number of papers and mainly relied on quality to win.

In order to verify this guess, Zeta Alpha also made a ranking of the conversion rate of highly cited papers.

As expected, OpenAI won the crown and DeepMind came third.

Of course, Meta is also good. Fourthly, LeCun even came out to "come out and say it":

We at Meta really pay more attention to quality rather than quantity.

# In contrast, Google, which has more citations but more publications, only ranked ninth, barely out of the top 10.

In addition to these, the second place is also a highlight-it is Megvii.

and domestic SenseTime are also on the list.

Attached is the complete list of the Top 100 citations in 2022

The popularity of ChatGPT has really revitalized the AI ​​industry. In what direction will the latest cutting-edge research lead? We also need to be more observant.

To this end, Zeta Alpha has also given a list of all AI papers cited in the top 100 in 2022, which may be inspiring to everyone.

1-30：

31-60：

61-90：

91-100：

So, that’s all the content of Zeta Alpha’s report.

The original text can be stamped:​​​https://www.php.cn/link/ea20aed6df7caa746052d227d194a395​​

The above is the detailed content of The ranking of 'high citation conversion rate' of AI papers is released: OpenAI ranks first, Megvii ranks second, and Google ranks ninth. For more information, please follow other related articles on the PHP Chinese website!