Over the past year, the AI world has entered a new stage of development led by large models.
While people marvel at the powerful capabilities of large models, they are also concerned about security risks such as the compliance of their training data sources and biased data use.
Similarly, in the context of the increasingly standardized industry regulatory environment and the increasing emphasis on information security and privacy data, federated learning research and application trends are gradually moving towards trusted federated learning.
The "Federated Learning Global Research and Application Trends Report" is a non-profit project that tracks the dynamics and progress in the field of federated learning. The 2023 annual report is the third issue of this series and aims to update and showcase the latest developments in federated learning scientific research results and technology applications.
The report mainly introduces the technical research and application progress of federated learning from its birth in 2016 to 2022 in a relatively comprehensive and in-depth manner from the aspects of technical research, scholar portraits, mainstream frameworks, industry applications, and development trends, and looks forward to the future of this technology. Development direction and prospects.
Federated Learning Knowledge Tree
Based on the key technologies and related technologies of federated learning, as well as the research topics of highly cited academic papers in this field, this report unearths the important technical points of globally active federated learning and represents them as a knowledge tree structure, as shown in Figure 1.
Figure 1 Federated learning knowledge tree
Report Highlights
1. The number of global federated learning papers published is growing rapidly
There were a total of 6,861 papers related to federated learning from 2016 to 2022. Since federated learning was proposed in 2016, the number of research papers has increased year by year. In 2019, it entered a stage of rapid growth, with an average annual growth rate of 38.6%.
Figure 2 Trend of federated learning research papers (2016-2022)
2. “China and the United States” lead the development of global federated learning
China and the United States are far ahead of other countries in publishing federated learning papers. More than 60% of highly cited papers come from China and the United States, and China and the United States have the largest number of collaborative papers in the world; more than 70% of the best papers come from China and the United States. Among them, Google and Carnegie Mellon University have the largest number of highly cited papers, while in China, Beijing Post and Telecommunications, Hong Kong Science and Technology, Sun Yat-sen University and Shenzhen Big Data Research Institute.
Figure 3 Distribution of highly-copied papers on federated learning by country (2016-2022)
3. The popularity of applied research is gradually increasing
Industry applications are becoming more and more mature. In recent years, the application research interest of federated learning in the Internet of Things, edge computing, healthcare, databases, vehicle interaction and recommendation has gradually increased.
Figure 4 Research hotspot trends in federated learning applications (2016-2022)
4. Analysis of National Natural Science Foundation of China Project Funding
According to the public data on the official website of the IMF, obtain the funding status of federal learning fund projects in China (including mainland China, Hong Kong and Macao) and cooperation between China and foreign regions. Judging from the overall situation of the data obtained so far, although the number of related fund projects has increased slightly from 2016 to 2022, the total number is small. A total of 156 federal learning approved fund projects in various regions were found, including 102 There are 14 projects each including the National Natural Science Foundation project NSFC, 26 Hong Kong innovation and technology fund projects, the Macao Science and Technology Development Fund project FDCT, and the National Natural Science Foundation of China international cooperation projects.
Figure 5 Distribution of approved federal learning-related fund projects in several countries and regions from 2016 to 2022
5. Federated learning framework and system status
Open source frameworks mainly come from China and the United States. Among them, Pysyft launched by OpenMined and FATE in the FATE open source community have a popularity of more than 4,000, ranking in the first echelon; FedML.AI's FedML, Adap's Flower, Google's TFF and other frameworks are also very popular, with popularity exceeding 2000, and the two frameworks FATE and FedML have currently launched LLM modules.
Figure 6 Federated learning framework open source trend chart
Report directory
To view the full report content, you can click on the link to view.
Report link: https://lfs.aminer.cn/misc/pdf/FL-Report-23-Chinese.pdf
About Us—Zhicause
AMiner is an intelligent auxiliary analysis and interpretation platform for scientific and technological documents under Zhipu AI.
The platform is based on scientific research big data such as papers, scholars and institutions, and
comprehensively uses big data, artificial intelligence, cloud computing and other technical means
to provide full-field technical insight and trend analysis services to governments, enterprises, scientific research institutes and other institutions.
Want to know more technical information in the field:
Follow AMiner ( https://vip.aminer.cn/analysis/ ) to customize special research reports!