Sat 31 August 2024
While I'm not a particularly important person with singular sage advice, I still often find myself coaching aspiring applicants for PhD programs (and fellowship applications), and have a much better understanding of the process now than I did at my application time. So, I'm documenting the nuggets of advice I do have here to save me a bit of time sharing them with my mentees, and hopefully others will get some use out of them too!
If you are an US citizen or permanent resident entering a PhD in CS, you absolutely must apply for the National Science Foundation Graduate Research Fellowship Program (NSF GRFP) grant. Your odds are quite good. This program is basically structured as a carrot to help Americans pursue careers in science, and is not as selective as NDSEG, Hertz, or the insanely competitive fellowships from Google, Microsoft, etc. It has a roughly 1 in 8 acceptance rate---a well-written application has a great chance.
Feedback and advivce from mentors and previous awardees were instrumental in my successful GRFP app. In particular, I made use of sample statements passed around privately from prior applicants at my alma mater, ASU. I suspect the network effects of having more prior applicant peers who were sharing their statements as well as more organized adivsing explains the "elite institution" bias of GRFP award statistics. Hopefully if more of us share our advice and statements, this gulf can be narrowed!
This post contains:
Zach Portman released his statements on his desire to pursue research about bees. Consequently, he talks about his interest in them and how that inspired him to pursue research in biology. I cosign this statement from him:
Focus on the personal element of the application and demonstrate a pattern of determination and curiosity about science in general. One of the main mistakes I see in other people who apply for the GRFP is that they focus almost entirely on their research proposal and neglecting the personal elements. It’s important to remember that the purpose of the GRFP is to nurture promising scientists, not necessarily to fund the best scientific proposals (although these often go together).
Rowan Zellers similarly shared a good post with advice on GRFP apps specifically for NLP. An important point that he hammers home is that the statement does need to convincingly address a real broader impact, which may at first brush seem the reverse of Zach's advice, but they are actually well aligned. From Rowan's post:
The NSF does not expect you to actually carry out your research proposal. On the contrary, the research proposal requirement seems to test your ability to sketch out a coherent research plan for a couple of years... and to discuss its (hopefully fantastic!) *broader impact on society.
The idea is that you need to show your promise as a scientist who can plan a concrete and attainable project you demonstrably care about (grounded in your prior work) and are on board for the mission of publicly funded science---helpfully impacting society through your research---and that you understand how the former fits in to the latter. Note here that I italicized the attainability of the project and the final impact, while I bolded ideas of "caring" "having a mission" and "understanding."
You only need to show you have promise and your heart is in the right place; you are not being stack ranked against candidates on any objective measure of project plan---they don't even expect you to specifically carry out your project.
I also want to emphasize this point from Rowan's post, on soliciting feedback:
You should talk to people in your area as well as not in your area. That's good for two reasons: you want to make sure that your application is easy to understand for a non-specialist, but on the same token, if someone in your area happens to read your application, you don't want them to ding you points for feasibility or correctness.
Your application readers may be in any field of computer science (or electrical engineering, etc), with no knowledge of your research topic at all, but they might be NLP experts as well. You need to thread the needle of maximizing understandability for non-experts while minimizing the impression of oversimplification for experts.
As for the process of actually preparing your application and submitting, Alex Lang provides a useful application guide, including an annotated copy of the call for proposals and a timeline to take for preparing your statements. Also, he has a giant spreadsheet containing hundreds application samples from tons of fields; definitely take a look at some examples in computer science.
I also asked some successful applicant friends for their thoughts on applying:
All my reviewer feedback was about my background being good prep for the proposed direction + strong service/outreach. I applied as an undergrad, so I didn’t link the proposal to a specific advisor. I had a feasibility paragraph where I discussed where I’d get data, how I’d run experiments, and some concrete statements on how my training prepared me to do the proposed research etc
Once again, you really should focus on telling a story that connects to your demonstrated background and explains why you can do good research. It does not matter if you actually do the project you proposed. You don't even need to tie your application to a specific advisor (though there can be benefits to doing so, as I did, see below for more discussion).
Other friends gave me more lighthearted stories from their applications:
I personally wouldn’t dig too far in the reviews, one of them thought I went to UCLA lol (from a student from a different UC)
The panelists are really trying to breeze through these. Passing is really about meeting a 1/8 bar for excellence, not about being the very best of all your peers.
I spelled "sociolect" wrong over and over in my research statement. I became a fellow anyways, but maybe good to know that they don't check spelling.
Like I said---the panelists are reading fast.
The two statements you will submit are "Graduate Research Plan Statement" and a personal statement ("Personal, Relevant Background, and Future Goals Statement"). Additionally, you provide your CV, and solicit 3 letters of recommendation from mentors. Here's what I submitted:
After you apply, you get feedback from the reviewers (whether your application was accepted or rejected). Here's the feedback I got:
I think the following were key elements to my success:
I absolutely went to town bolding, italicizing, underlining and combination-of-those-three formatting the text in my statements. Please make sure you do this (with taste) in order to hammer home the few key points from each set of paragraphs, to highlight the key definitions, etc. I even write my papers this way, it really helps make sure readers don't miss key points.
My proposed project was a direct continuation of my demonstrated experience. Even though GRFP reviewers are explicitly told they're reviewing candidates, not project proposals, and that the proposal is a sample project, not a commitment, they still may judge your proposal like it's any other NSF proposal. Write a proposal describing what you think you can most convincingly show them that you can do right now and then pivot to what you actually want to do once you have the award.
I applied in the electrical engineering track rather than the computer science track. I did this for several reasons; I was applying to a mixture of CS and ECE programs (NLP and Speech broadly) and had more of a speech background, so applying as ECE made sense to match up to my strongest-project-I-can-execute (see 1.).
Furthermore, I think I benefited applying as a deep learning person within ECE because those reviewers were probably seeing a much more diverse range of proposals—people working on signal processing, people working on power efficiency, etc were all in my pool—had I applied as a CS person perhaps they would have not found this invocation of deep learning as novel or exciting.
My description of a research agenda of pretraining good adaptable deep learning backbones (which was very much the zeitgeist at that time) for the specific problem area of clinical speech analysis (which is an easy-to-motivate problem but not a massive research community) perhaps gave me a best-of-both-worlds outcome where readers didn't view me as yet another exhausting applicant trying to ride the DL zeitgeist. Maybe consider this before building your app around LLMs 😉. Also, GRFP panels are explicitly trying to have balance, and award grants to a wide array of topics. In a less crowded field your chances are better!
Finally, I think my application went well because I was able to convincingly motivate why I'm interested in the problem area I proposed (neurological disease) by relating it to my own personal story (having a neurological disease).
(I didn't originally plan to do this, but was suggested to do so by mentors). I think this improved my application by further convincing the panel that I'm a good pick for someone to advance that area, and it served double duty with implicitly justifying deficiencies in my CV that were impacted by my condition.
Nelson Liu and Anjalie Field have uploaded their successful statements standing alone, they're a great resource to look at if you want more examples of successful apps.
Philip Guo wrote some amazing stuff on how to apply to the GRFP (and PhD programs in general) in the past, but has unfortunately chosen to completely nuke them. Here's where I would put links to them, IF I HAD ONE.
If you have any further questions about applying to the GRFP, or if you have your own previous successful application materials that you would like to share with me so I can distribute them as examples for other prospective applicants please email me.
I also have a modest collection of other people's successful applications. While I am not comfortable just posting them all, I will share them with you privately, provided you then share your statements back to me after so I can add them to my collection.
Category: Career; Tags: PhD, Admissions, NSF GRFP, Applications