Chinese universities have significantly overtaken their American rivals in research output at one of the world’s leading artificial intelligence (AI) conferences, according to an analysis of more than 5,000 accepted papers that went viral on social media.
Among the top 50 institutions contributing to the International Conference on Learning Representations (ICLR), held in Rio de Janeiro last month, mainland China accounted for about 44 per cent of the total, with Tsinghua University, Shanghai Jiao Tong University, Zhejiang University and Peking University taking the top four spots globally.
Some social media users pointed out that if Hong Kong’s 7.7 per cent share was counted together with mainland China’s, the country’s contribution would exceed half of all papers presented at the conference.
“If you count mainland China, Hong Kong and Singapore as mostly Chinese, and add the many Chinese-American researchers [in the US], AI research is basically a Chinese racket,” one user identifying himself as a Silicon Valley hi-tech entrepreneur wrote, partly in jest.
Tsinghua University alone had 332 accepted papers, based on an analysis by Ukrainian computer scientist Dmytro Lopushanskyy, who is now an AI technical lead at Seattle Children’s Hospital.
The United States accounted for about 32 per cent of the papers, led by Stanford University, Carnegie Mellon University and Massachusetts Institute of Technology, which each produced roughly half as many papers as Tsinghua University.
Chinese and American tech companies were neck and neck in the rankings, with Alibaba, Huawei, ByteDance and Tencent competing alongside Microsoft, Meta, Nvidia and Google. Meanwhile, Europe – home to prestigious universities in Britain and Switzerland, among others – accounted for less than 10 per cent of the papers, according to Lopushanskyy’s analysis.
China’s rise in AI research has been building for years. It has already surpassed the US in the volume of AI papers, citations and patent filings, although the US still leads in frontier AI models and private investment, according to the AI Index Report released by the Stanford Institute for Human-Centred Artificial Intelligence last month. -- SOUTH CHINA MORNING POST
The viral chart used a “unique affiliation” method – meaning each institution received one full count for every paper it appeared in, regardless of how many co-authors or partner organisations were involved.
Lopushanskyy also compiled a separate “first author” ranking, counting only the institution affiliated with a paper’s lead author. Under that method, Tsinghua University still ranked first with 174 papers, followed by Zhejiang University, Shanghai Jiao Tong University and Peking University.
But the picture changed under a third “fractional” method, where credit was split among all participating institutions for each paper, according to his full dataset published on GitHub.
Using that approach, mainland China accounted for 28.1 per cent of the total, narrowly behind the US at 28.3 per cent, suggesting the two countries were effectively neck and neck in overall research contribution.
Other social media users argued that paper output alone did not determine leadership in AI. The US still dominates key parts of the industry, from advanced chip supply chains and venture capital to some of the world’s most powerful industrial AI labs.
ICLR is widely considered to be one of the “Big Three” AI and machine learning conferences in the world. It focuses on areas such as neural networks, large language models and reinforcement learning. -- SOUTH CHINA MORNING POST
