Recently, a European PhD candidate has stirred up the machine learning community. He couldn't help but wonder why his American counterparts seem to have cheat codes, each holding ten top-tier conference papers, five of which are first-authored. During his own PhD journey in Europe, he spent the first year figuring out what research entails, published a paper at CVPR in the second year, and only began to grasp the nuances of project management and grant applications by the third year.
Now, his resume has only added two more papers, one in a journal and one at a conference, both as the first author. It sounds impressive, but compared to the achievements of his American peers, he is left astounded.
"How do they manage it? Do they not need sleep?" He is baffled by the efficiency of these American PhDs. He believes he is no less intelligent than them, yet whenever he has a new idea, he finds that a Stanford or DeepMind PhD has already published similar work. Truly understanding the depth of these papers requires considerable time and effort, how can one complete it in just a few months?
Image source note: The image is generated by AI, authorized by the service provider Midjourney
This PhD candidate is filled with questions about what factors enable these individuals to be so efficient. He is not one to engage in comparisons, as everyone's environment is different. But in his view, for someone with only three years of research experience, continuously producing high-quality research results within a short year is truly unbelievable.
Some netizens have also expressed their views, believing that the competition in the American academic world is incredibly intense. In the US, the work culture is to push hard. A person who once studied in a top computer science program in the US revealed that graduate students work more than 10 hours a day, almost year-round. Once, he went to the lab at seven in the evening and found his classmates still hard at work, not leaving until one in the morning.
Although it's not an official requirement, the environment creates immense pressure. Needless to say, labs in the US are filled with top global talent. For example, a top project at Tsinghua University has an acceptance rate of only 0.1%. Attracting such outstanding students to work ten hours a day, it's no wonder they produce results.
Of course, this phenomenon is not limited to the AI field; it's prevalent in almost all STEM (Science, Technology, Engineering, and Mathematics) fields. Some have experienced this state during their physics PhD, where there was nothing else to do.
Moreover, resource disparity is a significant factor. The resources for top PhDs are vastly different. Some labs are equipped with numerous expensive GPUs, significantly accelerating research progress. Those without such resources can only look on in envy. Even among different universities in the US, the availability of GPUs varies greatly.
Even a top-tier machine learning PhD posted about how his lab didn't have a single H100, leading to a scramble for computational resources. Compared to the "GPU elite" like Princeton and Harvard, PhD students with insufficient resources naturally cannot quickly achieve research results.
Finally, the endorsement of renowned institutions is an intangible boost. The close ties between top universities and large tech companies not only inspire innovative projects but also provide additional resource support. It's clear that the reasons behind this disparity in academic output are indeed worth pondering.