Chinese universities have surpassed their American counterparts in research output at a major international artificial intelligence conference. An analysis of over 5,000 accepted papers revealed that at the recent International Conference on Learning Representations (ICLR) in Rio de Janeiro, institutions from mainland China were responsible for about 44% of the contributions. Leading the charge were Tsinghua University, Shanghai Jiao Tong University, Zhejiang University, and Peking University, which occupied the top four global spots. If contributions from Hong Kong are included, China’s presence at the conference exceeds half of all presented papers.
Tsinghua University alone had 332 accepted papers, while the United States contributed around 32%, with Stanford, Carnegie Mellon, and MIT leading the American institutions. Chinese and American tech giants like Alibaba, Huawei, ByteDance, Tencent, Microsoft, Meta, Nvidia, and Google were similarly matched in their contributions. Europe, despite its prestigious universities, accounted for less than 10% of the papers.
China’s growth in AI research has been evident for years, having surpassed the US in the number of AI papers, citations, and patents. However, the US maintains a lead in cutting-edge AI models and private investments. The “unique affiliation” method used in the viral analysis credited each institution with one full count per paper. In contrast, a “first author” ranking, focusing on the lead author’s institution, still placed Tsinghua University at the top. A “fractional” method, which divided credit among all participating institutions, showed China and the US nearly tied in contributions, with China at 28.1% and the US at 28.3%.
Some argue that the sheer number of papers doesn’t equate to leadership in AI, as the US continues to dominate in advanced chip supply chains, venture capital, and leading industrial AI labs. ICLR is recognized as one of the top three AI and machine learning conferences globally, focusing on neural networks, large language models, and reinforcement learning.
