Machine learning researchers often face the "publish or perish" dilemma, with top conferences like ICML, Neur IPS, and ICLR being primary outlets. Publishing requires navigating a competitive peer-review process with low acceptance rates, typically 20-30%.

While attending these conferences offers valuable networking and exposure to new research, the travel itself incurs significant hidden costs. Beyond the actual conference days, preparation, long flights, jet lag, and organizational overhead can consume an additional week or more. The author highlights their experience attending ICML in South Korea, which, combined with travel and recovery, disrupted their work rhythm.

Upon returning, researchers face a backlog of emails and tasks, alongside potential project stagnation. This interruption can easily lead to a loss of one to two weeks of progress on research projects, impacting crucial tasks like developing proofs or writing drafts. The author emphasizes that while conferences are valuable, their true cost extends far beyond the scheduled event days, demanding a more holistic accounting of time and energy.