The most cited paper in the 21st century is a report about Deep Residual Learning (ResNet) networks published by Microsoft researchers in 2016. This groundbreaking paper, which addresses efficient learning in artificial neural networks crucial to deep learning, has significantly contributed to major AI breakthroughs like AlphaGo, AlphaFold Protein Structure Database, and OpenAI’s ChatGPT.
On Wednesday, the prestigious scientific journal Nature unveiled its list of the 25 most cited papers from millions published since 2000, based on data from five major academic citation databases.
Microsoft’s ResNet research explores machine learning implementation using a neural network with 150 layers, about five times deeper than previous models. This work tackled the longstanding challenge of signal degradation in deep neural network layers.
The paper has been cited over 250,000 times on Google Scholar and more than 100,000 times in the Web of Science database.
The list is dominated by papers that have driven AI innovations.
Ranking seventh is the 2017 paper Attention is All You Need by Google researchers, which introduced the Transformer methodology that allows machine learning models to prioritize important information. It is a key technology that has contributed to developing conversational AI like ChatGPT.
The AlexNet study, involving last year’s Nobel Physics laureate Geoffrey Hinton, ranks eighth. This paper addressed how neural networks identify and label objects in images.
Ranking twelfth is the U-Net paper, an offshoot of AlexNet research, which focuses on modifying neural network structures to reduce the data required for image processing.
Hinton notes that AI-related papers are being produced and utilized at a much faster rate than those in traditional scientific fields. He attributes the high citation counts to AI’s broad applications in healthcare, finance, robotics, and translation.
The open-source nature of most early AI machine-learning papers also contributed to their widespread influence.
Other high-ranking studies include analytical tools for life sciences and medical research, such as papers on Polymerase Chain Reaction (PCR) calculations, programs for analyzing X-ray scattering patterns to reveal atomic structures and comprehensive reports on global cancer incidence rates.