入职新加坡国立大学之前,哥大计算机视觉博士总结五年研究生涯(7)
2023-10-27 来源:旧番剧
2.2 单篇Paper选题
前面说的PhD选题是大方向上的,具体到每一篇paper,选择的principle和重点则不太一样。来Facebook后从马爷爷那知道了一个著名的Heilmeier问题系列,是指导老师们申项目的,我觉得稍微改改,便很适用于我们考虑,某一篇paper的选题,合不合适:
What are you trying to do? Articulate your objectives using absolutely no jargon.How is it done today, and what are the limits of current practice?Who cares? [Support other’s research? Shape research landscape? Power applications in industry?]What"s new in your approach and why do you think it will be successful?If you"re successful, what difference will it make? [e.g. Contributions in theory/modeling? Improve accuracy by 5% on dataset A, B, C…
?]What are the risks and the payoffs? [Further, how would you mitigate the risks? If your proposed method does not work, what could be alternative design? These can end up as discussions such as ablation studies in your paper.]How much will it cost? [e.g. How many GPUs do your experiments require? How long is each training process? How about data storage?]How long will it take? [How many hours are you going to work on this per week? When is the submission DDL? Can you make it?]What are the midterm and final "exams" to check for success?