Reddit帖子质疑中国机器学习研究的发表实践
一名Reddit用户询问中国的机器学习研究是如何开展的,并称研究者和学生能够发表大量论文并参与广泛的合作网络。该帖还声称,ChatGPT和Claude在部分论文中发现了可疑段落,而这些论文后来被ICLR、AAAI等会议接收。作者询问,发表与招聘网络以及学术任人唯亲是否可能解释这些现象。
该帖为审查发表产出、合作网络、潜在的AI辅助写作以及发表实践对研究质量的影响提供了社区层面的讨论线索。相关说法仍是个人观察,并以问题形式提出,而非已经确立的事实。
In this post I am not trying to bash Chinese researchers by any means but I have gotten VERY curious in the past few years as to how research is conducted in China.
It is no secret to anyone with a pair of eyes that Chinese researchers in ML (especially Chinese researcher based in China) are extremely productive to the point of defying common sense. We are talking about 100-200 papers while an undergraduate/graduate student. 1 paper a week is not unheard of. Can easily pull up Google scholar profiles of random Chinese researcher or PhD student with hundreds of papers submitted to dozens of different conferences. At the same time they have co-authors in the range of dozens and sometimes hundreds, especially when it comes to benchmarks or surveys.
The research seem to concentrate heavily in the LLM/Multi-Modal space, with very few novel mathematical derivation/arguments but tons of experiments, sometimes ranging almost hundred page of experiments. I ran some of these work in ChatGPT/Claude to check for possible LLM generated artifacts and usually it quickly returns a list of suspicious sections, but the paper were subsequently accepted in ICLR/AAAI and similar conferences anyways.
I am confused as to what exactly is happening. I understand that Chinese folks have entire publishing ecosystem and hiring ecosystem that is insular to the west. It is well known that the Chinese faculties working in the states would routinely advertise for academic positions in places like Xiaohongshu (RedNote) or Douyin or BiliBili, which creates a certain level of unfairness. But I wonder if similar network exist for people who want to publish papers to get together to "scratch each other's back", or give someone the proper credentials (some kind of academic nepotism).
Would appreciate it if someone can give us an inside look into how this all work.