Gradschool advice - writing a statement of purpose

Don’t just write a statement; make one.

Disclaimer: I am no expert in admissions. This series of blog posts is a hastily organized braindump resulting from abundant introspection, and from discussions I’ve had over the last few years with prospective grad students. Much of this is my view of the process and holds (if at all) only for North American PhD programs in ML/CV/Robotics. Take the advice in this post with a grain of salt.

This post is for you if you are writing a research statement / statement of purpose (SOP) for grad school applications.

This is a very opiniated view on statements of purpose (SOPs). In this post I will NOT talk about personal statements which may be required by some schools.

For someone that’s a grad student, I read a lot of SOPs every year. Over the last 6 months, I’ve read more than 50 statements (and helped edit nearly two-dozen). I often find myself giving the same piece of advice across most of these statements, so I thought “why not save myself the time by curating the most frequent bits here?"

If you’re wondering why I read so many statements, it’s partly cause I help process internship/graduate applications at our group, and most mentees that seek mentorship from me are grad school applicants looking for a pair of eyes on their statements.

I often spend 20 or 30-minutes when reading a first-draft statement – I have plenty of redundant feedback across most first-drafts. Over the years I’ve started curating canned responses/guidelines, and I will attempt to organize them in this post.

Be you

The first bit of trivial, yet essential advice: be you. Every statement is unique because it outlines your purpose. Do not optimize a statement to cater to your prospective reviewers’ interests, or due to fear of competition. Be transparent – this will allow your potential reviewers to assess your application’s needs in the fairest manner possible. Do not feign interest in a particular research area or direction you have no intention of pursuing or just because it’s trendy. People can usually tell these apart, particularly if you do not come off as having an in-depth understanding of the area.

The “purpose” of the statement

A number of statements fail to convey the motivation behind pursuing a particular program. This is a crucial aspect of your SOP – describe why you are interested in computer science in general, and your sub-field in particular. Reasons could be goal-driven (e.g. accelerating drug discovery using reinforcement learning) or curiosity-driven (e.g. why does stochastic gradient descent work so well for training neural nets) or something else. Stating them upfront provides reviewers with a pretext for everything that follows in your SOP. See this tweet for an endorsement. Again, “be you”.

Not a personal statement

This is specifically for PhD programs

A common pitfall I’ve seen with most SOPs is that they over-emphasize personal traits and qualities. Such personal claims about diligence, creativity, and enthusiasm suffer a credibility problem because the applicant has incentive to brag or exaggerate. One way to overcome this is by letting the other parts of your application package establish such personal traits. Your letter writers are the best people to do this – they do not suffer the aforementioned credibility problem. Rather than emphasize personal traits, it helps to think of your SOP as a mini research proposal. Talk about why your research direction is important and exciting. Imagine you were successful; what kind of new applications would then be enabled?

No cliché

At this point I’ve lost count of the number of SOPs that start with a childhood anecdote having to do with science fairs or robot competitions or TV shows. As aptly summed up by Kris in this Tweet thread, “Please don’t start your SOP with the word Ureka!"

Avoid chronology unless “absolutely” necessary

Courtesy Vikram Nitin

A very common failure mode is to organize all your projects chronologically. Worse still, many statements begin with a description of undergrad, high-school, or middle-school. This usually leaves very little room to talk about research agenda. Also admission committees are interested in knowing the set of your research themes, not the sequence. Adopt a “most impactful first” ordering. Make your opening project count – better to open with your latest-and-greatest project than a project you did early on in your undergrad.

Emphasis on future research

Another silent failure mode is when your statement does not talk enough about future research plans. Prospective advisors aren’t just interested in knowing your background; they are keen to know what you would like to work on if you are offered a spot. It is thus crucial to discuss in detail your future research directions. I have seen many good SOPs devote equal space for past and future research. Just describing past projects does not give your potential advisor enough signal to gauge your research trajectory. Note, however, that noone will hold you to executing on the research problems outlined in your statement. Most people indeed change their research trajectory (at least secondary subject areas) upon admission or as they progress in a research career.

Generic-Specific-Generic organization

A good design idea for research statements is the generic-specific-generic pattern. Begin with a somewhat broad characterization of your interests (Hypothetical example: My research revolves around thinking how we might leverage advances in deep learning to rethink the geometry processing pipeline ground up …). Do not be too broad (Hypothetical example: My research interests include deep learning and computer vision). Once you have characterized your interests, deep dive into a few specific projects and proposed directions. Towards the end, it’s a good idea to present a big-picture overview of your vision and the applications and/or insights your research might unlock.

Think about the full application package

Contextualize your SOP with respect to your application materials; don’t treat it as if it were an independent document. You can use your SOP to talk about things that aren’t necessarily on your CV (poor academic performance, career disruptions, etc.). Some universities require personal statements in which case that’s the best place for these discussions. Alternatively, if you have too many projects to talk about and too little space in your SOP, consider asking one of your letter writers to discuss a project in greater detail. A common mistake people tend to make is to treat the SOP as a verbose form of their CV. Remember that admissions committees always look at your CV.

Do your homework

Please, please, do your homework! A number of statements end up listing faculty that have either retired or aren’t taking on students in the application cycle. Ensure that you list faculty who are actively hiring. Another related trend is most SOPs end up listing the top (most sought-after) 3-4 faculty at a school, without actually sensing a research fit. It is important to ensure that you pick your prospective advisors wisely. Merely going after the top-k advisors implies stiffer competition and hence lower likelihood in receiving an offer. On an unrelated but serious note, please get all the names of your prospective advisors, labs, and schools right. I still receive so many applications to MILA, even after it has been renamed to Mila :)

Personalize

Personalize your statement to the school and program you are applying for. Many statements merely list 2-3 influential papers per faculty by peeking at their Google Scholar profile. It is important to asses whether your prospective advisors are still actively working in the area. What’s more important is to get a sense of your prospective advisor’s viewpoints about areas of your interest. One way to do this is to watch recent talks by your potential advisors. Look out for anything that might convey their thoughts on a particular problem. The Q&A during and after the talk might present you with additional insights about your potential advisor’s perspective.

With nearly all SOPs read on digital screens and readers, there are lots of little things you can do to make your SOP more readable. Hyperlink anything that can be hyperlinked – people/schools/labs/companies you’ve worked with, project pages, manuscripts, videos showcasing your work, media coverage; these are all great ways to supplement your statement. If you use colored text, images, or plots, ensure that your palettes are colorblind-friendly.

Explicit structure

When people often read two-page-long statements it is hard to remember each point conveyed along the way. It helps if your statement is organized into explicit sections and subsections – this way the reader can skim and get a sense of all the themes being discussed. (To appreciate the value of structure, just skim the subheadings of all the above paragraphs. Didn’t you think it helped you instantly recreate a gist of the blog post? Readers of your SOP will truly appreciate this gestalt.)

Edit, re-edit, re-(re-)edit

That’s it! The single most important takeaway of this post. As with any piece of scientific writing, take as many passes of the document as you can before you send it out for review or submit it. Get as many of your friends/mentees to review it as possible. It is important that your SOP is readable by people outside of your primary research area. This improves the chances of the SOP appealing to a larger fraction of prospective advisors.

Drafting and continuously editing out statements can be daunting and/or stressful. However, remember that the end result is worth the effort. I personally draw immense satisfaction from watching SOPs evolve over time; reminiscing Dorothy Parker’s quote “I hate writing, I love having written”.

Finally, remember that each school and review committee has different expectations from a statement of purpose. It is best to get each of your statements reviewed by students at the school it is geared towards.

If you have any thoughts to share about this, drop me a line. Twitter DMs welcome too!

Here are a few other articles in this series.

  1. How to pick your grad school
  2. Demystifying letters of recommendation
Krishna Murthy Jatavallabhula
Krishna Murthy Jatavallabhula
PhD Candidate

My research blends robotics, computer vision, graphics, and physics with deep learning.